{
 "name": "IB Maths IA ideas",
 "author": "Pietro Meloni",
 "url": "https://www.pietromeloni.com/ib-ia-ideas/maths",
 "updated": "2026-11-11",
 "count": 146,
 "ideas": [
  {
   "id": "building-and-breaking-a-small-rsa-system",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Building and breaking a small RSA system",
   "researchQuestion": "How does the difficulty of factorising the modulus grow with its size, and at what modulus size does a brute-force factoring attack stop being feasible on a school laptop?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Prime factorisation, modular arithmetic and congruences, Euler's totient φ(pq)=(p−1)(q−1), extended Euclidean algorithm for the private exponent, fast modular exponentiation by repeated squaring. Beyond syllabus: Fermat's little theorem as the proof that decryption returns the message.",
   "howToStart": "Choose p=61, q=53, e=17, compute n, φ(n) and d, then encrypt and decrypt a short number by hand. Next write a trial-division factoriser and time it on moduli of 10, 20, 30 digits.",
   "slVsHl": "An SL student can carry out the worked example and the timing study. HL depth comes from proving why decryption works with Fermat or Euler, and from analysing how trial division cost scales with sqrt(n).",
   "whereMarksAreLost": "E: a worked example alone is a demonstration, not exploration. C: RSA is very common, so a generic account feels copied. D: little evaluation of the model's limits, such as key size and weak prime choices.",
   "dataNote": "Easy to run with a short script, but the timing data need repeated trials to be credible.",
   "verdict": "Very common topic, so it only works with your own experiment on factoring times and a discussion of a bad choice such as p and q close together. Otherwise choose something else."
  },
  {
   "id": "hanoi-with-four-pegs-how-much-time-does-a-spare-peg-save",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Hanoi with four pegs: how much time does a spare peg save?",
   "researchQuestion": "How many moves does an optimal transfer of a tower of n discs need with three pegs, and by how much does a fourth peg reduce that count as n grows?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Recurrence relations T(n) = 2T(n-1) + 1 with T(1) = 1; solving by iteration and by adding 1 to both sides to get a geometric sequence; closed form 2^n - 1 and geometric series sums. Four pegs: splitting the tower into a top group of k discs and the rest, giving F(n) = min over k of [2F(k) + 2^(n-k) - 1]; triangular numbers to explain why the best k follows a pattern. HL extension: proof by induction of the three-peg formula, and comparing growth rates using limits or logarithms. Beyond syllabus: the Frame-Stewart strategy, proved optimal for four pegs only in 2014 and still unproved for five or more.",
   "howToStart": "Build a physical or online puzzle with 1 to 5 discs and record the shortest solution by hand. Then write a short program or spreadsheet that computes the shortest solution by breadth-first search for three pegs up to about 8 discs and four pegs up to about 10 discs. Compare the counts with the recurrence predictions before doing any algebra.",
   "slVsHl": "SL: derive and solve the three-peg recurrence, verify against small cases, and tabulate the four-peg strategy with a clear explanation of the split. HL: add a full induction proof, derive how fast the four-peg count grows (the exponent grows roughly like √(2n)) with justification, and check the strategy against your own exhaustive search for small n.",
   "whereMarksAreLost": "A and B: a page of move lists with no structure, and unexplained notation for the recurrence. C: retelling the legend without any personal angle. D: stating the four-peg values without saying how you know they are minimal. E: stopping at 2^n - 1, which is a textbook exercise.",
   "dataNote": "Data comes from a computed search or hand play and is easy to generate. The optimality proof for four pegs exists but is far beyond the course, so quote it and check it computationally for small n instead of attempting it.",
   "verdict": "The three-peg version alone is overdone and thin, but the four-peg comparison gives it a real question and room for computation. Worth choosing if you write the search program yourself; otherwise it reads as a textbook exercise."
  },
  {
   "id": "routing-a-postal-round-through-awkward-streets",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AI",
   "level": "HL",
   "title": "Routing a postal round through awkward streets",
   "researchQuestion": "How much extra distance must a postman walk to cover every street of a real neighbourhood, and how does the number of odd-degree junctions determine that extra cost?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Graphs, degrees of vertices, Eulerian circuits and trails, weighted graphs, the Chinese postman algorithm: pairing odd vertices by minimum total weight, with shortest paths between them found by inspection. Beyond syllabus: Dijkstra's algorithm for the shortest paths, a proof of the handshake lemma and why the pairing method is optimal.",
   "howToStart": "Take a map of the streets in your own area, mark junctions as vertices with lengths as weights, count odd-degree vertices and list all pairings of them with their path lengths.",
   "slVsHl": "This is AI HL content, since graph theory is not in the SL course. HL depth: compare your optimal route with a greedy hand-planned one, count how the number of pairings grows with the number of odd vertices, and test what happens when one street is closed.",
   "whereMarksAreLost": "E: shortest paths between odd vertices guessed rather than justified. B: graph diagrams cluttered and vertices unlabelled. A: too many pairing tables in the main text.",
   "dataNote": "Street lengths come from an online map and take time to collect, and the network must be kept small enough to check by hand.",
   "verdict": "A good personal choice because you can use your own streets. Keep the network to about 8 vertices, or the by-hand pairing becomes a slog."
  },
  {
   "id": "the-order-of-a-perfect-shuffle-for-decks-of-any-size",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "The order of a perfect shuffle for decks of any size",
   "researchQuestion": "For a deck of 2m cards, after how many perfect out-shuffles does the deck first return to its starting order, and can this number be predicted from m alone?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Permutations written in cycle notation; the out-shuffle as the position map x → 2x (mod 2m−1) on positions 0 to 2m−1; modular arithmetic and powers of 2 modulo n; the order of a permutation as the lowest common multiple of its cycle lengths; the multiplicative order of 2 modulo 2m−1. Beyond the syllabus, and reachable: Euler's theorem to bound the answer by φ(2m−1), and a comparison with the in-shuffle, which behaves like 2x modulo 2m+1.",
   "howToStart": "Write a spreadsheet or short Python loop that applies one out-shuffle to a list of 52 numbered cards and repeats until the list matches the start. Record the count for every even deck size from 2 to 60, then write out the cycles for 8 and 10 cards by hand to see why the counts differ.",
   "slVsHl": "An SL student can tabulate counts for small decks, describe the position rule and explain the pattern using modular arithmetic and lowest common multiples. HL depth or top-band sophistication comes from proving that the return count equals the multiplicative order of 2 modulo 2m−1, explaining why 52 cards need 8 shuffles, and characterising the decks where the count is as large as possible or unusually small. Using the in-shuffle as a contrast adds a second model to compare.",
   "whereMarksAreLost": "E: stopping at a table of counts with no proof of why the modular rule works. B: cycle notation and the position indexing (starting at 0 or 1) not defined clearly, so formulas look inconsistent. C: no personal angle, for example a card trick or magic performance to motivate it. D: little reflection on the limits, since real shuffles are imperfect. A: long tables for many deck sizes left unedited.",
   "dataNote": "Pure theory plus easy computation, since a short loop generates every result and the proofs stay at elementary number theory level.",
   "verdict": "A sound, well-defined choice where the mathematics genuinely does the work, though the topic appears on a couple of other sites so it needs your own twist. Make it personal by linking it to a card trick you can actually perform, and merge the second, randomness-focused angle only if you can keep the maths controlled."
  },
  {
   "id": "area-of-the-polygon-formed-by-the-nth-roots-of-unity",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AA",
   "level": "HL",
   "title": "Area of the polygon formed by the nth roots of unity",
   "researchQuestion": "What is the exact area of the polygon with vertices at the nth roots of unity, how does it approach π as n grows, and how does the product of the diagonals from one vertex behave?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Roots of unity, polar form, area of a triangle as ½ab sin C, sum of n triangles giving (n/2) sin(2π/n), limit using Maclaurin series, and the product of chord lengths from one vertex equal to n via factorising z^n − 1. Sum of the roots equal to zero.",
   "howToStart": "Compute the area for n = 3 to 12 by hand and with a spreadsheet, then derive the general formula and compare with π.",
   "slVsHl": "Complex numbers in this form are HL only in AA. Depth comes from the chord product result and the error bound as n increases.",
   "whereMarksAreLost": "E: only the trigonometry area formula is used and the complex link is missing. B: gaps in the polynomial factorisation. C: the result stays impersonal.",
   "dataNote": "Pure theory with tidy closed forms that can be checked numerically.",
   "verdict": "Neat and safe, but the area part alone is a textbook exercise. Adding the chord product proof and a comparison with Archimedes' polygon method for π gives it substance."
  },
  {
   "id": "bounding-the-best-delivery-round-for-local-addresses",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AI",
   "level": "HL",
   "title": "Bounding the best delivery round for local addresses",
   "researchQuestion": "How close to the true shortest round trip are the bounds given by nearest-neighbour and deleted-vertex methods for eight to ten real addresses?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Complete weighted graphs, Hamiltonian cycles, nearest-neighbour upper bound, minimum spanning tree by Kruskal or Prim, deleted-vertex lower bound, exhaustive check of all tours for small cases. Beyond syllabus: the triangle inequality and the factor of two guarantee from a spanning tree.",
   "howToStart": "Choose eight addresses, for example a paper round, and build the distance table from a map. Find a nearest-neighbour tour, then a lower bound by deleting one vertex and adding two edges to the spanning tree.",
   "slVsHl": "Graph theory and the travelling salesman bounds are AI HL content. HL: bounds for one set of addresses, several deleted vertices to tighten the lower bound, the triangle inequality argument, and how the gap between the bounds changes with the number of addresses. An SL student would be working outside the course and should pick another idea.",
   "whereMarksAreLost": "E: bounds found but never justified. B: terms such as Hamiltonian and spanning used without definition. C: addresses picked at random. D: no discussion of what real drivers care about besides distance.",
   "dataNote": "Only a distance table is needed, but a brute-force check beyond about nine points needs a short program.",
   "verdict": "A solid discrete maths choice with clear answers, but a tour of universities or local sights is named in the subject reports as a template seen too often. It is stronger when the gap between the bounds is tested on several address sets, and when the addresses are ones you actually visit."
  },
  {
   "id": "breaking-a-hill-cipher-with-matrices-and-modular-arithmetic",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AI",
   "level": "HL",
   "title": "Breaking a Hill cipher with matrices and modular arithmetic",
   "researchQuestion": "How many plaintext and ciphertext pairs are needed to recover the key of a Hill cipher, and how does the choice of modulus affect which keys are usable?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Modular arithmetic, matrix multiplication, determinants and inverse matrices (AI HL), inverses modulo 26, greatest common divisor, counting invertible keys. Beyond syllabus: modular inverses of matrices and Bezout's identity.",
   "howToStart": "Encode a short message with a 2 by 2 key modulo 26 by hand. Then solve for the key from two known plaintext pairs and check when the determinant has no inverse.",
   "slVsHl": "Matrices are AI HL, so this is an AI HL idea. Depth comes from the known-plaintext attack, a proof of when it recovers the key, and a count of the usable 2 by 2 keys. An SL student could do the affine cipher idea instead.",
   "whereMarksAreLost": "E: encrypting and decrypting only, with no analysis of attack. B: number theory stated but not justified. A: long tables. D: no comparison of cipher strength.",
   "dataNote": "Simulation in a spreadsheet or code is easy, but proving key counts needs care.",
   "verdict": "A good choice if the attack and key count are the centre. Common cryptography descriptions score poorly."
  },
  {
   "id": "check-digits-how-well-do-isbn-and-card-numbers-catch-typing-errors",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Check digits: how well do ISBN and card numbers catch typing errors?",
   "researchQuestion": "What proportion of single-digit and adjacent-swap errors is detected by the ISBN-13 and Luhn check-digit schemes, and why do some errors escape?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Modular arithmetic, divisibility, weighted sums, counting arguments and probability of detection, systematic listing of all error types. HL can add proof using congruences and the role of primes as moduli. Beyond syllabus: the Verhoeff scheme, which uses group theory.",
   "howToStart": "Write a spreadsheet or short program that takes 100 real ISBN numbers, applies every possible single-digit change and every adjacent swap, and counts how many still pass the check.",
   "slVsHl": "SL can count detections by simulation and explain with mod 10. HL can prove exactly which swaps go undetected and why modulus 11 in ISBN-10 behaves differently.",
   "whereMarksAreLost": "E: stopping at describing the scheme with no proof. B: unexplained modular notation. D: no comparison between schemes.",
   "dataNote": "Numbers can be generated or taken from books at home, and the proofs are short.",
   "verdict": "A good, safe topic with real mathematics if the detection rates are proved and not just quoted. Twist: test the numbers on your own bookshelf and on a card scheme you invent."
  },
  {
   "id": "comparing-savings-plans-and-loan-repayments-with-geometric-series",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Comparing savings plans and loan repayments with geometric series",
   "researchQuestion": "How long does it take a regular monthly saving plan to reach a target sum, and how does the answer change with the compounding frequency and a realistically varying interest rate?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Sum of a finite geometric series; compound interest with n compounding periods per year; solving for the number of periods with logarithms; comparison of nominal and effective rates and the limit as n tends to infinity giving continuous compounding with e. Sum to infinity for perpetuities.",
   "howToStart": "Take the published rates of two or three real savings accounts or a student loan. Derive the closed formula for a monthly deposit annuity, then check it against a spreadsheet that rolls the balance forward month by month.",
   "slVsHl": "SL students derive and use the annuity formula and logarithms for time to target. HL extension: continuous compounding as a limit, and modelling a rate that changes each year with a product of factors compared with a constant average rate.",
   "whereMarksAreLost": "E: using the formula from a textbook without deriving it or testing its assumptions. C: generic figures with no connection to a real decision. D: no reflection on inflation, tax and fixed rates.",
   "dataNote": "Bank rates and inflation figures are public and easy to cite.",
   "verdict": "Safe and accessible but risks being a textbook exercise. Make it personal with a real goal, such as a deposit for university, and test the model against varying rates."
  },
  {
   "id": "continued-fractions-and-the-best-rational-approximations",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AA",
   "level": "both",
   "title": "Continued fractions and the best rational approximations",
   "researchQuestion": "Why do the convergents of a continued fraction give the best rational approximations to √2, e and π, and how does the error compare with the size of the denominator?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Recurrence relations for convergents, sequences and limits, solving quadratics for periodic expansions such as √n, proof by induction of p_k q_{k−1} − p_{k−1} q_k = ±1, comparison of errors with 1/q² and a log scale plot. Beyond syllabus: Pell's equation from convergents.",
   "howToStart": "Expand √2 and π by the Euclidean-style algorithm, compute the first eight convergents and tabulate |x − p/q| against 1/q².",
   "slVsHl": "SL treats quadratic irrationals and the recurrence. HL adds the proof of the determinant identity and the bound |x − p/q| < 1/q².",
   "whereMarksAreLost": "E: listing Ramanujan's expansions with no explanation. A: unfocused wandering across patterns. D: no assessment of what is proved versus observed.",
   "dataNote": "Everything is computable by hand or spreadsheet and the proofs are short.",
   "verdict": "Quite a good idea once tied to approximation. Personal link: gear ratios, calendars or musical tuning make the application concrete."
  },
  {
   "id": "counting-steps-in-the-euclidean-algorithm-and-lam-s-bound",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Counting steps in the Euclidean algorithm and Lamé's bound",
   "researchQuestion": "What is the worst-case number of division steps in the Euclidean algorithm for numbers below N, and why do consecutive Fibonacci numbers produce it?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Euclidean algorithm and division with remainder, the Fibonacci sequence, the closed form with the golden ratio, inequalities and induction, logarithms to bound step count, average step count found by computation, Bézout's identity by back substitution. Beyond syllabus: Lamé's theorem.",
   "howToStart": "Code the algorithm, record the step count for all pairs (a,b) up to 1000, find which pairs give the maximum, and check that they are Fibonacci neighbours.",
   "slVsHl": "SL can do the experiment, the Fibonacci pattern and a log bound. HL adds induction for the exact bound and a comparison of average versus worst case with a fitted model.",
   "whereMarksAreLost": "B: the algorithm described but not analysed. E: nothing beyond computing a GCD, which is only textbook. A: code dumps.",
   "dataNote": "The data is simulated easily, though the average-case model is a bit harder to justify.",
   "verdict": "On its own, computing a GCD is a textbook exercise. The worst-case step count and the Fibonacci link turn it into real mathematics."
  },
  {
   "id": "do-different-voting-rules-crown-different-winners",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Do different voting rules crown different winners?",
   "researchQuestion": "Using ranked ballot data from a real preferential election, how often do plurality, Borda count, runoff and Condorcet rules disagree on the winner, and which fairness criteria fail?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Counting and permutations of preference orders; Borda scoring as arithmetic sequences; pairwise majority comparison in a table; probability of a Condorcet cycle for three candidates under random ballots, exactly 1/18 for three voters (countable by hand) and about 8.8% as the number of voters grows; simulation of random electorates. Beyond syllabus: Arrow's impossibility theorem stated, not proved.",
   "howToStart": "Find a published ranked ballot dataset, for example from Scottish council elections or a club or society election, and tabulate first preferences. Then compute Borda scores and the pairwise matrix for the top three candidates.",
   "slVsHl": "SL students can compare methods on one real dataset and construct a small paradox by hand. HL depth comes from computing the cycle probability analytically and comparing it to simulation across electorate sizes.",
   "whereMarksAreLost": "E: descriptive comparison with little calculation; A: too many methods introduced without focus; C: no personal link such as a school council vote; D: not questioning whether the data represent true preferences.",
   "dataNote": "Ranked ballot data is available but needs cleaning; a simulation fallback exists if real data is thin.",
   "verdict": "Rarely listed, so fresher than most, but it turns weak if it stays a list of methods. Keep it to a sharp question on disagreement rates and use real ballots, ideally from your own school."
  },
  {
   "id": "fibonacci-ratios-and-the-speed-of-convergence",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Fibonacci ratios and the speed of convergence",
   "researchQuestion": "Why do ratios of consecutive Fibonacci numbers settle at the golden ratio, and how many terms are needed before the ratio is correct to a chosen number of decimal places?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Second-order recurrence; fixed-point equation x = 1 + 1/x solved with the quadratic formula; limits of sequences; error analysis using geometric decay. HL extension: Binet's formula from the characteristic equation, giving the error term exactly as a multiple of (−1/φ²)ⁿ. Beyond syllabus: generalise to Fₙ = aFₙ₋₁ + bFₙ₋₂.",
   "howToStart": "Use a spreadsheet to list the first 30 Fibonacci numbers and their ratios, then plot the absolute error against n on a log scale and look at the slope before any algebra.",
   "slVsHl": "SL students find the fixed point, show the errors alternate in sign and fit a geometric model to the data. HL students derive Binet's formula, prove the error rate is φ⁻² per step, and extend to other starting values or other recurrences.",
   "whereMarksAreLost": "E: stopping at the fixed-point equation, which assumes the limit exists without justifying it. D: reflection limited to 'it works'. C: weak if the golden-ratio-in-nature story is just repeated.",
   "dataNote": "Pure theory with spreadsheet checks; the only risk is assuming convergence rather than showing it.",
   "verdict": "Very common and safe, but easy to make routine. It becomes worthwhile only if you generalise to other recurrences and quantify the error rate yourself."
  },
  {
   "id": "greedy-algorithm-for-egyptian-fractions-does-it-always-stop",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Greedy algorithm for Egyptian fractions: does it always stop?",
   "researchQuestion": "Does the greedy method of always subtracting the largest possible unit fraction terminate for every proper fraction, and how long can the resulting expansions become?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Fractions and inequalities; proof of termination by showing the numerator strictly decreases; ceiling function; comparison of greedy expansions with shortest ones for fractions such as 4/n; sequences with rapid growth of denominators. Beyond syllabus: the Erdős-Straus conjecture 4/n = 1/x+1/y+1/z tested computationally.",
   "howToStart": "Perform the greedy method by hand on 5/6, 4/13 and 5/121, then code it and record the number of terms and the largest denominator for every fraction with denominator below 100.",
   "slVsHl": "SL students prove termination and study the data on length and denominator size. HL adds a proof that the denominators can grow extremely fast (each one roughly the square of the previous at worst), or a search for the shortest expansions of 4/n.",
   "whereMarksAreLost": "E if the termination proof is missing and the work only shows examples. B if the notation for the algorithm is unclear. C if it is only a history of Egyptian mathematics.",
   "dataNote": "The proof is short and the computations are small; big denominators need exact integer arithmetic in code.",
   "verdict": "A neat idea that is only worth choosing with the termination proof and a real comparison of algorithms. Testing 4/n against the Erdős-Straus conjecture gives it a personal, open-ended feel."
  },
  {
   "id": "last-digits-and-cycles-powers-under-a-modulus",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AA",
   "level": "both",
   "title": "Last digits and cycles: powers under a modulus",
   "researchQuestion": "For which moduli n and bases a does the sequence a^k mod n repeat, how long is its period, and how can this predict the last two digits of large powers?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Congruence notation, proof by induction for a^k ≡ a^(k+p) statements, periods of powers, and the link between the period and n. Beyond syllabus: Euler's totient function and Euler's theorem, the Chinese remainder theorem to split a modulus into coprime parts, and the multiplicative order of an element.",
   "howToStart": "Tabulate a^k mod n for n from 2 to 30 and a coprime to n in a spreadsheet, record the period each time, then compare with φ(n) and conjecture a rule.",
   "slVsHl": "SL students can tabulate, conjecture and prove cycle lengths for prime moduli using simple algebra. HL depth comes from proving Euler's theorem and using the Chinese remainder theorem to predict periods for composite moduli.",
   "whereMarksAreLost": "E: stopping at tables with no proof. B: heavy notation left undefined. C: a generic 'modular arithmetic is everywhere' framing with no personal angle.",
   "dataNote": "Pure theory with easy computation, and the proofs are elementary if the scope stays on powers.",
   "verdict": "A good and safe choice if the scope stays on powers. A twist such as designing a check-digit scheme or predicting the last digits of a number that means something to you makes it yours."
  },
  {
   "id": "leslie-matrices-for-a-real-species-population",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AI",
   "level": "HL",
   "title": "Leslie matrices for a real species population",
   "researchQuestion": "Can a Leslie matrix built from real survival and fertility data predict the long-term growth and age structure of a chosen population?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Matrix multiplication and powers, eigenvalues and eigenvectors, dominant eigenvalue as growth rate, diagonalisation, sensitivity of the eigenvalue to survival rates, comparison with a Markov chain for a closed system. Beyond syllabus: Perron-Frobenius reasoning.",
   "howToStart": "Find an age-structured life table for one species, such as a fish or a bird, build a 3-class Leslie matrix and compute the first ten generations.",
   "slVsHl": "Matrices and eigenvalues are AI HL, so this is an AI HL idea. Depth comes from the exact eigenvalue analysis, the stable age distribution as an eigenvector, and a sensitivity test: which survival rate changes the growth rate most.",
   "whereMarksAreLost": "E: eigenvalue work done by software with no explanation. C: a generic species with no personal link. D: no comment on data reliability.",
   "dataNote": "Reliable fertility and survival rates take searching in published life tables.",
   "verdict": "A good choice with a real data set. Choosing a species you care about and testing how a conservation change alters the growth rate makes it personal."
  },
  {
   "id": "modular-arithmetic-behind-caesar-and-affine-ciphers",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Modular arithmetic behind Caesar and affine ciphers",
   "researchQuestion": "How many distinct affine ciphers exist over a 26-letter alphabet, and how does frequency analysis break them?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Modular arithmetic, congruences, greatest common divisor and Euclidean algorithm, modular inverses, counting arguments. Beyond syllabus: Euler's totient function.",
   "howToStart": "Encode a short sentence with a shift of 3, then with a map x to 5x+8 mod 26. Find which multipliers a allow decryption by testing gcd(a,26).",
   "slVsHl": "SL treats Caesar and affine ciphers, counts the 312 affine keys and proves which multipliers are invertible. AI HL students can move to a Hill cipher with 2 by 2 matrices mod 26.",
   "whereMarksAreLost": "E: a Caesar cipher alone is a textbook exercise. B: modular notation used loosely. C: no real message or attack of the student's own.",
   "dataNote": "Little data needed. Use a real English passage for frequency analysis of a cipher you break yourself.",
   "verdict": "Caesar alone is too thin. It becomes a good exploration once you prove invertibility conditions and break your own ciphertext with letter frequencies."
  },
  {
   "id": "parametrising-every-pythagorean-triple-and-counting-them",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Parametrising every Pythagorean triple and counting them",
   "researchQuestion": "Why does every primitive Pythagorean triple take the form (m²−n², 2mn, m²+n²), and how many triples exist with hypotenuse below a given bound N?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Divisibility and parity arguments; proof by contradiction; rational parametrisation of the unit circle using the gradient of a line through (−1,0); coprimality conditions on m and n; counting with sequences and inequalities. Beyond syllabus: the tree of primitive triples generated by three matrices.",
   "howToStart": "List all triples up to hypotenuse 100 by hand or spreadsheet, mark the primitive ones and tabulate the (m,n) that produce each. Then prove that in a primitive triple exactly one leg is even.",
   "slVsHl": "SL students derive the formula through parity and factorisation, then count triples up to N and compare with a fitted trend. HL students can add the unit circle parametrisation with a full proof, or the matrix tree generation.",
   "whereMarksAreLost": "E if the formula is quoted rather than derived. A and B if a table of triples replaces argument. D if there is no comment on the limits of the counting estimate.",
   "dataNote": "Pure theory with a small computed check; the proofs are elementary but need careful writing about coprimality.",
   "verdict": "A classic that moderators know, so it earns marks only through a genuine proof and the counting question. Adding the growth rate of the number of triples below N gives it a personal, quantitative twist."
  },
  {
   "id": "pyramid-schemes-how-fast-recruitment-runs-out-of-people",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Pyramid schemes: how fast recruitment runs out of people",
   "researchQuestion": "After how many levels must a pyramid scheme collapse, and what share of participants lose money?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Geometric sequences and series, sum to n terms, exponential growth against a fixed population, logarithms to solve for the number of levels, comparison with a logistic model. HL: sum to infinity for a scheme with declining recruitment, and differential equation for logistic growth.",
   "howToStart": "Set a scheme where each member pays 100 EUR and recruits 4 people. Build a spreadsheet of members and money by level, then find the level where the total exceeds the population of Italy.",
   "slVsHl": "SL does the geometric model, the collapse level with logarithms and the share of members in the last level, who never recruit anyone: with each member recruiting r people it is close to (r - 1)/r of everyone, so 3/4 when r = 4. HL adds a recruitment rate that falls by level and a logistic comparison.",
   "whereMarksAreLost": "E: only summing a geometric series, which is textbook level. D: no discussion of how unrealistic constant recruitment is. B: the definition of a loser left unclear.",
   "dataNote": "All data is constructed by the student, which is easy but needs sensible parameter choices justified.",
   "verdict": "Accessible and very clear, though it risks being too easy. It earns marks only if the recruitment ratio is varied and a realistic decay model is fitted."
  },
  {
   "id": "ranking-pages-with-pagerank-and-a-damping-factor",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AI",
   "level": "HL",
   "title": "Ranking pages with PageRank and a damping factor",
   "researchQuestion": "How does the damping factor change the ranking of pages in a small network, and why does the iteration always settle?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Matrices, powers of matrices, eigenvalues and eigenvectors, the stationary vector of a stochastic matrix, geometric series for convergence. Beyond syllabus: the Perron-Frobenius result, quoted but not proved.",
   "howToStart": "Draw a 6-page network from a small real site or a set of accounts, build the link matrix column by column, and compute the dominant eigenvector by repeated multiplication from an even start.",
   "slVsHl": "Matrices and eigenvectors are AI HL, so this is an AI HL idea. Depth: show that the damped matrix has a steady state, find it as an eigenvector, and show how the second eigenvalue sets the speed of convergence.",
   "whereMarksAreLost": "E: leaning on a software eigenvector command without showing meaning. B: unclear notation for the matrix. C: reads as a summary of Google. D: no discussion of manipulation.",
   "dataNote": "Small hand-built networks are enough, and the theory can be handled with matrices.",
   "verdict": "Very good HL choice as it is maths from the first page. It stays weak if the student ends up describing social media, so keep it to the matrices and test damping values."
  },
  {
   "id": "roots-of-unity-and-complex-roots-on-the-argand-diagram",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AA",
   "level": "HL",
   "title": "Roots of unity and complex roots on the Argand diagram",
   "researchQuestion": "How do the roots of z^n = w and of polynomials with real coefficients arrange themselves in the complex plane, and what happens to them when a coefficient changes?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Polar and exponential form, de Moivre's theorem, nth roots, conjugate root theorem, factorisation of real polynomials, sums of roots of unity, transformations as rotation and scaling. Beyond syllabus: how roots move as a parameter varies.",
   "howToStart": "Solve z^n = 1 for n up to 8, plot the roots, then pick a family such as z^3 + az + 1 = 0 and plot how the roots move as a changes using software.",
   "slVsHl": "This is HL only because complex numbers are HL content. The extra depth comes from proving results such as the roots of unity summing to zero, and from tracking roots with a parameter and finding when they collide.",
   "whereMarksAreLost": "E: only plotting known roots, which is the textbook level; B: skipped steps in de Moivre derivations; C: no reason for choosing this; D: no reflection on accuracy of numerical roots.",
   "dataNote": "Needs only a graphing tool or a few lines of code to solve polynomials numerically.",
   "verdict": "Sound but risks being a routine exercise. Choose a parameter family or a link to geometry, for example regular polygons from roots, so the exploration asks a genuine question."
  },
  {
   "id": "steady-levels-under-repeated-medication-doses",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Steady levels under repeated medication doses",
   "researchQuestion": "For a drug with a known half-life, how does the time between doses control the peak and trough concentrations that build up over many doses?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Exponential decay C = C₀e^(−kt) with k from the half-life; geometric series for the concentration after n doses; the limit as n→∞ giving a steady peak C₀/(1−e^(−kτ)) for dose interval τ; inequalities to find when a therapeutic window is reached. HL: the model dC/dt = −kC between doses solved by separation of variables, compared with continuous infusion.",
   "howToStart": "Choose a real drug, such as paracetamol or ibuprofen, look up its half-life and dosing advice from an official leaflet, and calculate concentrations after doses 1 to 5 by hand.",
   "slVsHl": "SL students build the series model, derive the steady-state peak and trough, and test different intervals. HL students derive the same result from a differential equation and compare with a continuous infusion model.",
   "whereMarksAreLost": "E: assuming instant absorption with no comment. A: lots of tables and no clear structure. D: not questioning the one-compartment assumption or the reliability of half-life values.",
   "dataNote": "Half-life values are easy to find, but check that they come from reliable sources and state clearly that this is not medical advice.",
   "verdict": "A good applied idea with real mathematics. Strong if you use a drug you know and compare your model with the printed dosing limits."
  },
  {
   "id": "true-cost-of-a-loan-simple-compound-and-hidden-fees",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "True cost of a loan: simple, compound and hidden fees",
   "researchQuestion": "For a realistic consumer loan, how much of the total repayment comes from the interest method and how much from fees, and at what term does the ranking of offers change?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Geometric sequences and series, compound interest with different compounding periods, the loan repayment formula (amortisation) derived from a geometric series, solving for the interest rate numerically, effective annual rate, comparison of offers via graphs and intersections. Beyond syllabus: the APR as the root of an equation for the cash flows.",
   "howToStart": "Take two or three published loan offers (bank or car finance sites). Model each in a spreadsheet, then derive the monthly payment formula by hand and check it against the spreadsheet.",
   "slVsHl": "SL derives the repayment formula and finds the effective rate of each offer. HL can extend to continuous compounding, or to solving for the true APR with an arrangement fee by numerical root finding. Top band comes from finding a crossover term where the cheapest offer switches.",
   "whereMarksAreLost": "E: using only the calculator's finance solver with no derivation. A: long tables of numbers with no structure. C: generic loan with no personal link.",
   "dataNote": "Offers are public, and all the maths can be verified independently.",
   "verdict": "A solid and practical choice, but a common one. It becomes strong when you derive the amortisation formula and find the APR yourself rather than quoting it."
  },
  {
   "id": "where-the-sum-1-2-3-1-12-breaks-down",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AA",
   "level": "HL",
   "title": "Where the sum 1+2+3+... = -1/12 breaks down",
   "researchQuestion": "Which steps in the common derivations of 1+2+3+... = -1/12 are legitimate, and what does the value actually mean for a smoothed or regularised sum?",
   "dataDifficulty": 3,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Geometric series and convergence, Maclaurin series, sigma notation, partial sums, Abel summation of 1-1+1-1+..., a smoothed sum using an exponential cutoff with limits (HL), then a simple form of the Riemann zeta function continuation as the beyond-syllabus idea.",
   "howToStart": "Take Grandi's series 1-1+1-1+... and show that Cesàro and Abel summation both give 1/2. Then apply the exponential cutoff e^(-nε) to 1+2+3+..., expand with Maclaurin series and find the constant term.",
   "slVsHl": "SL students can study partial sums, why the usual shifting trick fails and the Abel sum of Grandi's series. HL students can do the cutoff calculation with series expansions and limits, and compare it with the zeta value.",
   "whereMarksAreLost": "E: proofs that copy the zeta continuation without understanding it. B: unclear distinction between 'equals' and 'is assigned'. D: no critical view of why the value is useful in physics.",
   "dataNote": "The full zeta continuation needs complex analysis, so the scope must be held to the cutoff method.",
   "verdict": "Popular and often done badly. It is only worth choosing if you stay with the cutoff calculation and are honest about what -1/12 means. The twist is testing the three summation methods on the same series."
  },
  {
   "id": "where-the-2-irrationality-proof-breaks",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AA",
   "level": "both",
   "title": "Where the √2 irrationality proof breaks",
   "researchQuestion": "For which positive integers n is √n irrational, and which step of the classical √2 proof stops working when n is a perfect square?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Proof by contradiction; parity and divisibility arguments; prime factorisation and the fundamental theorem of arithmetic; generalising to √p, √(pq), and ³√2 and other roots. Beyond syllabus: the rational root theorem.",
   "howToStart": "Write out the √2 proof line by line, then try it word for word on √3, √4 and √6 and mark each line where it fails or needs a new argument.",
   "slVsHl": "Proof by contradiction is AA HL, so HL students handle √p for primes, then any non-square n using prime factorisation, and extend to kth roots and to sums such as √2 + √3. An SL student can study the step where the argument fails for perfect squares, but the proofs themselves sit in the HL course.",
   "whereMarksAreLost": "E: repeating the standard proof without extension. B: unclear wording in the contradiction steps. C: seen extremely often, so little personal input.",
   "dataNote": "Pure theory; the main danger is being a textbook exercise unless you find your own family of cases.",
   "verdict": "Well-worn and short. It only merits choosing if you push to general n and kth roots and explain why the parity argument needs replacing."
  },
  {
   "id": "which-exponents-give-mersenne-primes-testing-and-proving",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Which exponents give Mersenne primes? Testing and proving",
   "researchQuestion": "Why must n be prime for 2ⁿ−1 to be prime, and how does the Lucas-Lehmer test decide primality far faster than trial division?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Factorisation of 2^(ab)−1 using geometric series; modular arithmetic; proof by contradiction; recurrence sequence s(k+1) = s(k)²−2 mod (2^p−1); comparing running times with growth of exponential functions. Beyond syllabus: a sketch of why Lucas-Lehmer works via quadratic extensions.",
   "howToStart": "Factorise 2ⁿ−1 for n up to 30 by computer, and note which are prime and which prime n still fail (11, 23, 29). Then prove the necessity of n prime using a geometric series.",
   "slVsHl": "SL students prove the prime exponent condition and run the Lucas-Lehmer recurrence with modular arithmetic. HL adds a proof for the test itself or the link with perfect numbers.",
   "whereMarksAreLost": "C if the exploration only mentions distributed prime searches and does no mathematics of its own. E if the test is used without justification. A if the computational tables crowd out the argument.",
   "dataNote": "Small cases run on any laptop; the proof of the Lucas-Lehmer test is hard, so an honest scope must be chosen.",
   "verdict": "Good if it centres on the proof of the necessary condition and an efficiency comparison. Joining a distributed search is not mathematics and earns nothing in criterion E."
  },
  {
   "id": "which-numbers-are-sums-of-two-squares",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "AA",
   "level": "both",
   "title": "Which numbers are sums of two squares?",
   "researchQuestion": "Which positive integers n can be written as x² + y² with integers x and y, and how does the prime factorisation of n decide this?",
   "dataDifficulty": 2,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Prime factorisation, modular arithmetic mod 4, proof by contradiction, the identity (a²+b²)(c²+d²) = (ac−bd)² + (ad+bc)², counting solutions systematically. HL extension: complex numbers and Gaussian integers, where n = |z|² and the product identity is just |zw| = |z||w|.",
   "howToStart": "List all n up to 100 that are sums of two squares, factorise each, and look for what the failures share.",
   "slVsHl": "SL proves with arithmetic mod 4 that no number of the form 4k + 3 is a sum of two squares, and uses the product identity to build new sums from old. The full rule, that every prime of the form 4k + 3 must appear to an even power, needs a further argument that suits HL, and complex numbers make the product identity transparent.",
   "whereMarksAreLost": "E: quoting Fermat's two-square theorem with no proof of either direction. B: gaps in a proof by cases. C: no reason given for the interest.",
   "dataNote": "The full theorem needs care, so a student should plan which half to prove and check it with a program.",
   "verdict": "A proper number theory problem that you can genuinely solve in part. Plan which half of the theorem you will prove and check the rest with a program."
  },
  {
   "id": "winning-positions-in-nim-and-a-variant-of-my-own",
   "topic": "1",
   "topicName": "Number and algebra",
   "course": "both",
   "level": "both",
   "title": "Winning positions in Nim and a variant of my own",
   "researchQuestion": "How can the winning strategy in Nim be characterised using binary representations, and does it survive when a new move rule is added?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Binary representation, the XOR (nim-sum) operation as addition without carry, proof by induction of the P-position characterisation, recurrence relations for counting positions, combinatorial counting. Beyond syllabus: Sprague-Grundy values.",
   "howToStart": "Play two-heap and three-heap Nim by hand, tabulate all positions with small heaps as winning or losing, then look for the pattern before proving it.",
   "slVsHl": "SL can find and justify patterns for two heaps and small cases. HL or top-band work proves the nim-sum theorem by induction and analyses the modified game.",
   "whereMarksAreLost": "E if patterns are guessed and never proved; C if the game is copied without a personal variant; A if the tables are unreadable.",
   "dataNote": "Pure theory with small exhaustive tables.",
   "verdict": "Nim is well known, so the invented variant is what makes it yours. Solid if the proof is done and the variant is genuinely analysed."
  },
  {
   "id": "choosing-a-growth-model-for-a-real-population",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Choosing a growth model for a real population",
   "researchQuestion": "Which of a linear, quadratic, exponential or logistic model gives the most reliable description and forecast for the population of one chosen place from census records?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 4,
   "namedInReports": false,
   "mathsRequired": "Regression for linear, quadratic and exponential forms; residual plots; coefficient of determination; hold-out testing by fitting to early data and predicting later census values; logistic function with carrying capacity. HL: derive the logistic model from dP/dt = rP(1 − P/K) using separation of variables and partial fractions.",
   "howToStart": "Pick a city or country you know and download census counts for 100 years or more from a national statistics office. Fit each model to the first 70 percent of the data, then measure the prediction error on the rest.",
   "slVsHl": "SL compares the regression models and discusses the residuals. HL derives the logistic model and estimates its carrying capacity, using the hold-out test to expose overfitting.",
   "whereMarksAreLost": "C: a generic country with no reason for the choice. E: choosing the model with the highest R² without a hold-out test. D: no discussion of wars, migration or policy changes that break any smooth model. B: models used without defining the parameters.",
   "dataNote": "Census data is easy to find, but censuses are irregular and boundaries change, which needs careful handling.",
   "verdict": "Heavily used, so it only stands out with a place that matters to you and a hold-out test that shows models failing. A poorly understood logistic fit will cost marks."
  },
  {
   "id": "which-curve-describes-a-draining-phone-battery",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Which curve describes a draining phone battery?",
   "researchQuestion": "Which of a linear, exponential or piecewise model best describes battery percentage against time on my phone under one fixed activity, and how far can it be trusted for prediction?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 3,
   "namedInReports": false,
   "mathsRequired": "Linear and exponential regression; piecewise functions; residual analysis; root finding for the time to reach a given percentage; prediction intervals by hand from residual spread. HL: a differential equation dB/dt = −kB compared against a constant drain model.",
   "howToStart": "Fully charge the phone, run one repeated video loop at fixed brightness in flight mode, and note the percentage every 10 minutes until 5 percent. Plot the points and try the three models.",
   "slVsHl": "SL compares models by residuals and predicts the time to empty. HL derives the exponential model from a rate assumption and tests it against the data.",
   "whereMarksAreLost": "E: fitting a calculator regression with no reason for the chosen model. D: ignoring that the reading is rounded to whole percentages and that the drain is nonlinear near the end. A: several usage scenarios crammed in, so nothing is analysed deeply.",
   "dataNote": "Easy to collect, but controlling brightness, temperature and background apps is essential for a fair test.",
   "verdict": "Fine and personal if you keep to one controlled condition and take the rounding seriously. Trying to cover many usage types will make it thin."
  },
  {
   "id": "catenary-or-parabola-for-the-gateway-arch",
   "topic": "2",
   "topicName": "Functions",
   "course": "AA",
   "level": "both",
   "title": "Catenary or parabola for the Gateway Arch?",
   "researchQuestion": "Does a catenary or a parabola fit the outline of the Gateway Arch more closely, and how large is the difference over the height of the arch?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Hyperbolic cosine written with exponentials, y = a − b(e^(cx) + e^(−cx))/2; quadratic model; least-squares fitting by minimising the sum of squared errors; comparing residuals. Beyond syllabus: arc length by integration, and the tension argument for why a hanging chain is a catenary.",
   "howToStart": "Take the published height and width of the arch, and digitise the outline from a high-resolution photograph taken straight on, using a known dimension for scale. Fit both curves to the digitised points and plot the residuals.",
   "slVsHl": "Hyperbolic functions are in neither syllabus, so define cosh from exponentials, which is within reach at SL and HL. SL compares the two fits by residuals. HL can differentiate the catenary, show it satisfies the hanging-chain equation, and compute the arc length numerically.",
   "whereMarksAreLost": "E: fitting with software and no explanation of the curve. C: no link to the student's own interest. D: results reported without discussing the accuracy of the digitised image, perspective distortion or the fact that the actual arch is a weighted catenary. A: unclear coordinates.",
   "dataNote": "The digitising process introduces perspective error, so the scale and camera position must be stated carefully.",
   "verdict": "An attractive geometry problem, but the true arch is a flattened catenary, so the honest conclusion is subtle. Worth it if you handle the image error carefully."
  },
  {
   "id": "how-reliable-is-a-carbon-14-date-calibration-and-error",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "How reliable is a carbon-14 date? Calibration and error",
   "researchQuestion": "How much does the uncertainty in the measured C-14 activity and in the half-life change the calculated age of a sample, and how large is the correction needed for the changing atmospheric level?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Exponential decay, solving for t using natural logarithms, error propagation by differentiating t = (1/k) ln(A0/A), percentage uncertainty, interpolation of a calibration curve. Beyond syllabus but reachable: comparing the conventional age with the calibrated age using published IntCal data.",
   "howToStart": "Take a sample of known age, such as a tree-ring dated piece of wood, and compute its raw radiocarbon age from a stated activity ratio. Then differentiate the age formula to find how a 1 percent error in activity shifts the age.",
   "slVsHl": "SL students can use the age formula, compare the 5730 and 5568 year half-lives, and estimate the error from small changes in the ratio, using the derivative of a logarithm. HL students can fit a smoothed calibration curve to published data and study how the error behaves for very old samples.",
   "whereMarksAreLost": "E: only rearranging the decay formula for t. D: treating the answer as exact, with no discussion of error. B: skipping the derivation of the age equation. C: no personal angle.",
   "dataNote": "Calibration tables are public, but reading them and matching units takes effort.",
   "verdict": "The basic version is weak and overdone. Asking how wrong the date could be, and by how much calibration changes it, gives the mathematics real work to do."
  },
  {
   "id": "how-well-does-a-sinusoid-describe-a-real-ferris-wheel",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "How well does a sinusoid describe a real Ferris wheel?",
   "researchQuestion": "How closely does a function h = a sin(b(t − c)) + d fitted to timed video heights of a Ferris wheel cabin match the true motion, and where does the model systematically break down?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Sinusoidal functions, deriving amplitude, period, phase and midline from geometry, regression or least-squares fitting, residual analysis, and parametric circle equations. Beyond syllabus: differentiating the model to obtain vertical velocity and comparing it with video.",
   "howToStart": "Film a Ferris wheel or a large rotating object side-on with a known reference length. Use video analysis software such as Tracker to extract height every 0.5 s, then compute radius and period by hand before fitting.",
   "slVsHl": "SL fits the sinusoid, interprets parameters and analyses residuals with R squared and a plot. HL adds derivatives for velocity and acceleration, the projection of uniform circular motion, and a test of a non-uniform angular speed model such as a starting or stopping wheel.",
   "whereMarksAreLost": "E: fitting with software and reading off numbers without deriving the parameters. D: 'the model fits well' with no analysis of the pattern in residuals. A: heavy screenshot use without labelled axes.",
   "dataNote": "Data comes from one video, but calibrating scale and perspective needs care.",
   "verdict": "A classic that is safe but common. It becomes worthwhile if you derive the parameters from geometry independently and then use the residuals to find something real, such as speed variation."
  },
  {
   "id": "measuring-a-real-coastline-with-a-shrinking-ruler",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Measuring a real coastline with a shrinking ruler",
   "researchQuestion": "What is the box-counting or divider dimension of the coast of a chosen region, and how does the measured length scale with the ruler length?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Power law models L(ε) = kε^(1−D), logarithms to linearise, linear regression on log-log data, gradient and intercept interpretation, self-similarity with the Koch curve as an exact comparison (dimension log 4 / log 3). HL: geometric series showing that the Koch perimeter diverges.",
   "howToStart": "Print a high-resolution map outline of a coast such as Norway or Cornwall, walk dividers of decreasing length along it and record the number of steps. Then plot log L against log ε.",
   "slVsHl": "SL: log-log regression, estimate D, compare with the Koch curve. HL: prove with geometric sequences that the Koch length diverges, and discuss the reliability of the fitted gradient using residuals and the range of ruler lengths that behave like a power law.",
   "whereMarksAreLost": "E: only one measurement and no linearisation. D: no comment on map resolution limiting the smallest ruler. C: no personal choice of coastline.",
   "dataNote": "Manual divider walking is tedious and error prone, so consider using a scripted box count on a downloaded outline.",
   "verdict": "Common but sound if you collect your own measurements and compare several coastlines. Choose somewhere you know to make it personal."
  },
  {
   "id": "why-decibels-and-magnitudes-are-logarithmic-testing-scale-choices",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Why decibels and magnitudes are logarithmic: testing scale choices",
   "researchQuestion": "What do the decibel and moment-magnitude scales imply about ratios of real intensity and energy, and does a logarithmic scale match how we perceive or compare these quantities?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Laws of logarithms, solving exponential equations, base changes, inverse functions, composition, log-linear plots. Energy of an earthquake E proportional to 10^(1.5M), and decibels as 10 log10(I/I0). Beyond syllabus but reachable: the Weber-Fechner law from a differential equation dS = k dI / I.",
   "howToStart": "Collect intensity values for five familiar sounds and five earthquakes, convert to scale values, and calculate how many times more energy a magnitude 7 quake releases than a magnitude 5.",
   "slVsHl": "SL students can work with the laws of logs and compare scales. HL students can derive the logarithmic response from dS = k dI / I by integration and test it against a listening experiment.",
   "whereMarksAreLost": "E: only substituting into given formulae. D: no critique of where the scales fail, such as A-weighting. C: no personal experiment. B: unclear units.",
   "dataNote": "Sound and earthquake values are widely available.",
   "verdict": "Weak as a topic in its usual form, since it is largely descriptive. A listening test of your own, together with the derivation from the Weber-Fechner assumption, is what lifts it."
  },
  {
   "id": "equal-temperament-versus-just-intonation-as-a-ratio-problem",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Equal temperament versus just intonation as a ratio problem",
   "researchQuestion": "How closely can the twelve-note equal-tempered scale approximate the simple frequency ratios of just intonation, and why does twelve notes work better than other divisions of the octave?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Geometric sequences with common ratio 2^(1/12); exponential and logarithmic functions for cents, 1200 log2(f2/f1); comparison of ratios such as 3/2 and 5/4 with powers of 2^(1/12); continued fraction or rational approximation of log2(3/2) to justify divisions into 12, 19 or 31 notes. Beyond syllabus: the Pythagorean comma by comparing 3^12 with 2^19.",
   "howToStart": "Calculate the frequencies of one octave in equal temperament from A = 440 Hz, then compute the error in cents of each interval against its just ratio. Next, search for integers n and k with 2^(k/n) close to 3/2.",
   "slVsHl": "SL students can do the geometric sequence, the cents scale and the table of errors. HL depth comes from continued fractions of log2(3/2) or a proof that no power of 2 equals a power of 3, explaining why a perfect tuning is impossible.",
   "whereMarksAreLost": "E: an audio description with little calculation. A: Fourier analysis brought in at length, which is too heavy unless kept to a small, clearly explained part. C: no personal instrument or recording used.",
   "dataNote": "All calculation is from known ratios, and a recorded note from the student's own instrument can add a small measured check.",
   "verdict": "A good one if focused on the tuning question, which has a genuine mathematical answer. Recording your own instrument and measuring its frequencies gives the personal angle."
  },
  {
   "id": "fitting-piecewise-curves-to-the-batman-logo",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Fitting piecewise curves to the Batman logo",
   "researchQuestion": "How accurately can a small set of piecewise functions reproduce a chosen logo, and how does the area under them compare with the measured area of the printed logo?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Transformations of functions, piecewise definitions, regression or curve fitting for quadratics and ellipses, domain and range restrictions, integration for area between curves, percentage error. Beyond syllabus: least squares fitted by hand with matrices.",
   "howToStart": "Import a clear logo image into Desmos or GeoGebra, place it on a grid, choose about 15 points along the wings and fit each segment, then record the fitting error.",
   "slVsHl": "SL students can fit conics and quadratics with technology and integrate for area. HL students derive the least squares fit and compare polynomial with ellipse or trigonometric pieces using residuals.",
   "whereMarksAreLost": "E: plotting with no justification for the chosen functions. B: pages of equations with no explanation. D: no measure of how good the fit is.",
   "dataNote": "Easy to start, but a quantitative error measure must be added to make it more than drawing.",
   "verdict": "Drawing a picture with functions is a common choice and, done as plotting only, a weak one. It becomes acceptable with a fit error measure and an area comparison, and a logo you chose for a reason makes it personal."
  },
  {
   "id": "fitting-the-keeling-curve-trend-plus-seasonal-cycle",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Fitting the Keeling curve: trend plus seasonal cycle",
   "researchQuestion": "Which function, a quadratic, an exponential or a linear plus sinusoidal model, best describes monthly atmospheric CO₂ at Mauna Loa, and what does each predict for 2050?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Regression, sinusoidal functions with amplitude and period, exponential and quadratic models, residual analysis, coefficient of determination, extrapolation. Beyond syllabus: differencing the data to compare growth rates.",
   "howToStart": "Download the monthly Mauna Loa series from NOAA and plot it. Fit a trend to the annual means first, then add a sine term with a 12-month period fitted to the residuals.",
   "slVsHl": "SL compares fitted trends and a sine term by regression and residuals. HL extends with a fitted growth-rate function and a discussion of what the exponential term claims about the rate of change of CO₂ growth.",
   "whereMarksAreLost": "E: choosing the model with the best R² and nothing else. D: no criticism of extrapolating 25 years. B: unexplained parameters from a Desmos fit.",
   "dataNote": "The data are clean and public, but 60+ years of monthly values need to be handled in a spreadsheet.",
   "verdict": "Sound, though common. The strength lies in comparing models on residuals and forecast, not in drawing one curve. Add a check against a later year held back from the fit."
  },
  {
   "id": "have-olympic-100-m-and-1500-m-times-improved-in-step",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Have Olympic 100 m and 1500 m times improved in step?",
   "researchQuestion": "Which function family best models Olympic winning times for the men's 100 m and 1500 m since 1900, and does the rate of improvement in the two events move together?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Regression with linear, exponential decay and logistic models, comparison by residuals and r², correlation between the two time series, derivative of the fitted model to give rate of improvement, limit or asymptote as a predicted physical bound.",
   "howToStart": "Copy winning times by year from the official Olympic results tables into a spreadsheet, then plot both series and fit a linear model to each.",
   "slVsHl": "SL fits and compares linear and exponential models with technology. HL differentiates the models, finds the asymptote, and questions extrapolation to future Games.",
   "whereMarksAreLost": "E: linear fit only. D: missing Games in 1916, 1940 and 1944 and different timing technology not discussed. B: treating time as independent variable in correlation without noticing shared trend.",
   "dataNote": "Data is easy to obtain. The 1500 m is a middle-distance race, so the two events test whether sprint and endurance times improve at the same rate.",
   "verdict": "Decent, provided the correlation of two trending series is treated carefully, since shared trends inflate r. The asymptote question gives it a real result."
  },
  {
   "id": "how-small-must-an-angle-be-for-sin-x-x",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "How small must an angle be for sin x ≈ x?",
   "researchQuestion": "For which angles does the small-angle approximation stay within a chosen relative error, and how does the period of a real pendulum depart from the simple model beyond that range?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Relative error function (x − sin x)/sin x; solving error equations numerically; graphs of the error against x in radians; pendulum period T = 2π√(L/g) versus measured periods at larger amplitude. For HL, Maclaurin series of sin x showing the error is about x²/6. Beyond syllabus: the series correction to the period involving θ₀²/16.",
   "howToStart": "Tabulate the relative error for angles from 1 to 40 degrees, find the angle where it passes 1%, 5% and 10%. Then time 20 swings of a string pendulum at several release angles to see when the period starts to grow.",
   "slVsHl": "SL: numerical solution of the error inequality and comparison with pendulum data. HL: derive the error using Maclaurin series, predict the thresholds without technology and compare with the period correction.",
   "whereMarksAreLost": "E: maths limited to a table on a calculator. D: no reasoning on whether the tolerance chosen is sensible. A: mixing degrees and radians. C: little sign of personal insight.",
   "dataNote": "The error analysis is pure calculation, and the pendulum data are easy to collect.",
   "verdict": "Accessible and neat, but light at SL unless the pendulum comparison is done carefully. Choose a tolerance that has a physical reason, such as timing precision of a stopwatch."
  },
  {
   "id": "modelling-how-fares-climb-towards-departure",
   "topic": "2",
   "topicName": "Functions",
   "course": "AI",
   "level": "both",
   "title": "Modelling how fares climb towards departure",
   "researchQuestion": "Which function best describes the price of one fixed flight as the departure date approaches, and how well does it predict the next week's price?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Exponential and power regression, logarithmic linearisation, residual plots, coefficient of determination, percentage change, prediction error on held-out data. Beyond syllabus: a piecewise model with a change of behaviour at a fixed number of days before departure.",
   "howToStart": "Choose one route and date about eight weeks ahead, record the cheapest listed fare at the same time every day, and plot price against days to departure.",
   "slVsHl": "SL: fit two or three models and compare them with residuals and R squared. HL: use logarithms to linearise, hold out the last week to test predictions, and consider a piecewise fit.",
   "whereMarksAreLost": "E: choosing a model by R squared alone with no residual check. C: no reason for the route. D: one route and one date used to generalise about all airlines.",
   "dataNote": "Needs daily collection for weeks and prices can jump for reasons that are not modelled, such as sales.",
   "verdict": "Personal and testable, and the held-out prediction makes it stronger than most fitting projects. Start early because the data takes weeks."
  },
  {
   "id": "sinusoidal-daylight-curves-and-what-latitude-does-to-them",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Sinusoidal daylight curves and what latitude does to them",
   "researchQuestion": "How do the amplitude and phase of a sinusoidal model of daylight length change with latitude, and how well does the model fit real sunrise and sunset data?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Sinusoidal functions y = a sin(b(x - c)) + d, parameters from data, regression with technology, residual analysis. For HL depth, derive the daylight length from the solar declination with the formula cos(H) = -tan(latitude) tan(declination), using inverse trigonometric functions. Beyond syllabus but reachable: spherical trigonometry.",
   "howToStart": "Download daily sunrise and sunset times for five cities from 0 to 65 degrees north, convert to daylight hours, and fit a sine to each. Tabulate amplitude against latitude.",
   "slVsHl": "SL students can fit sinusoidal models, compare parameters across cities and discuss the polar limits. HL students can derive the exact formula from geometry, compare it with the sine fit and show where the sine approximation fails near the Arctic Circle.",
   "whereMarksAreLost": "E: only fitting sine curves on Desmos with no reason for the parameter values. D: no comment on the fit failing near the poles. B: no explanation of how the data was converted.",
   "dataNote": "Sunrise tables are publicly available for any city.",
   "verdict": "Accessible and reasonable, but a simple sine fit is thin. Deriving the amplitude from the tilt of the Earth and comparing it with the data makes it worth doing."
  },
  {
   "id": "smooth-joins-for-a-roller-coaster-track",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Smooth joins for a roller coaster track",
   "researchQuestion": "What is the lowest degree of polynomial that joins two track sections so that height, gradient and curvature all match at the join?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Piecewise polynomial functions; derivatives f′ and f″; solving simultaneous linear equations or matrices for coefficients; curvature κ = |f″|/(1 + f′²)^(3/2). Beyond syllabus: clothoid transition curves used in real track design.",
   "howToStart": "Pick a straight incline meeting a flat top. Try a quadratic join and check the gradient, then a cubic and quintic, solving for coefficients at each stage.",
   "slVsHl": "SL can match f and f′ with quadratics or cubics and compare graphs. HL or top-band work uses the curvature formula and shows why matching f″ is felt as a jolt through acceleration.",
   "whereMarksAreLost": "E: only solving a linear system with no discussion of why the conditions matter. C: no reference to a real ride. B: unclear notation for pieces.",
   "dataNote": "Pure algebra with fully controllable results, though real track data is optional.",
   "verdict": "A pleasant and visual choice with real engineering meaning. Base it on the profile of a real coaster you have ridden and compare your design with the measured shape."
  },
  {
   "id": "testing-the-titius-bode-law-with-logarithmic-scales",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Testing the Titius-Bode law with logarithmic scales",
   "researchQuestion": "How well does an exponential model fit the semi-major axes of the planets, and does the Titius-Bode style rule survive when tested with log-linear regression and residuals?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Exponential and logarithmic functions, linearisation by taking logs, least-squares regression and the correlation coefficient, residual analysis, comparing model fits with and without the asteroid belt. HL extension: fit a parametric model a = a0 + b·c^n by minimising the sum of squares.",
   "howToStart": "Download semi-major axes in AU for the eight planets plus Ceres from NASA's planetary fact sheet. Plot a against planet index, then ln(a) against index, and find the regression line for each.",
   "slVsHl": "SL students can linearise, fit and interpret the regression, and compare with the classical rule. HL students can fit a nonlinear three-parameter model, discuss sensitivity to the index assigned to the asteroid belt, and test predictions with Uranus and Neptune held back.",
   "whereMarksAreLost": "E: stopping at a calculator regression with no analysis of residuals. D: not asking whether the pattern is numerology. B: unclear labelling of what the index means.",
   "dataNote": "The data is tiny and public; the challenge is judging the fit honestly, not obtaining numbers.",
   "verdict": "A good, safe choice if you commit to testing the law critically instead of just admiring a straight line. Twist: hold back the outer planets and see if the model predicts them."
  },
  {
   "id": "two-wave-model-of-tide-heights-at-my-local-harbour",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Two-wave model of tide heights at my local harbour",
   "researchQuestion": "How much does the fit of a tide-height model at one port improve when a second twice-daily wave with a slightly different period is added to a single wave, judged by residuals over one month of published predictions?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Sinusoidal functions f(t) = a sin(b(t − c)) + d: amplitude, period, phase shift and midline read from data, then refined by sinusoidal regression on a GDC or software. Sum of two sinusoids with periods of about 12.42 hours (lunar) and 12.00 hours (solar), residual analysis, root mean square error and percentage error. The sum-to-product identity for sin A + sin B explains the beat pattern, which is why spring and neap tides alternate about every fortnight.",
   "howToStart": "Download one month of hourly tide predictions for a nearby port from a national hydrographic or tide-table site. Plot height against hours, estimate the period (about 12.4 hours) and the midline by hand, fit a single sine, then plot the residuals and look for a slow pattern.",
   "slVsHl": "An SL student fits one sine wave carefully, justifies each parameter from the graph and tests the fit with residuals and percentage error. Top-band depth comes from adding a second wave, comparing the two models with an error measure, and explaining the beat pattern with the sum-to-product identity. HL students can go further with a small Fourier-style fit of several tidal components, or a least-squares argument for the parameters.",
   "whereMarksAreLost": "E: stopping at one regression from the calculator with no reasoning about why the model works or where it fails. B: parameters quoted without showing how they were found. C: a generic port with no personal reason for the choice. D: no critique of the fit, such as ignoring that the tide period is not exactly 12 or 24 hours. A: too many similar graphs and tables.",
   "dataNote": "Tide predictions are publicly available in tabulated form, though the file may need reformatting into hours and heights before plotting.",
   "verdict": "Common and easy to do at a basic level, so only worth choosing with the twist of a second wave and a proper error comparison at a place you know, such as where you sail or swim. Avoid daylight hours as the topic, as that version is heavily overused."
  },
  {
   "id": "where-bacterial-growth-stops-being-exponential",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Where bacterial growth stops being exponential",
   "researchQuestion": "Over which time window does an exponential model fit measured bacterial growth, and does a logistic model fit the full curve significantly better?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Exponential functions and their logarithmic linearisation, regression on ln(N) against t, residual analysis, doubling time. Logistic function and its parameters. HL extension: solve the logistic differential equation by separation of variables and partial fractions, and use the slope field. Beyond syllabus: the Gompertz model.",
   "howToStart": "Download an optical density or colony count time series from a public growth curve dataset, such as published yeast or E. coli plate reader data. Plot ln(N) against t first and see where the line stops being straight.",
   "slVsHl": "SL students can fit exponential and logistic curves with technology and compare residuals. HL students solve the logistic equation analytically and relate the fitted constants to the carrying capacity and the maximum growth rate.",
   "whereMarksAreLost": "E: only fitting a curve with software and no analysis of the fit. A: too much raw data pasted in. D: not questioning what optical density really measures. C: generic, borrowed dataset with no comment.",
   "dataNote": "Public plate reader data exists but needs cleaning and a stated conversion from optical density to population.",
   "verdict": "A solid and common choice. It works well if the focus is where the exponential model breaks, not just fitting it. Own kitchen experiment with yeast would lift personal engagement."
  },
  {
   "id": "will-the-3x3-world-record-ever-hit-a-floor",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Will the 3x3 world record ever hit a floor?",
   "researchQuestion": "Which function type fits the progression of the 3x3 single-solve world record best, and does it approach a positive limit or keep falling?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Fit exponential decay with a horizontal asymptote, reciprocal and logarithmic models to record times. Compare using residual plots and R squared, and the sum of squared residuals. Solve for the year the record passes a given time. HL: least squares derived by calculus, and a model of records as the minimum of many random attempts using extreme value ideas. Beyond syllabus: Gumbel distribution.",
   "howToStart": "Copy the list of official world record singles with dates from the public record history page into a spreadsheet. Plot time against years since 2003 and try an exponential with an asymptote.",
   "slVsHl": "SL fits and compares three models and extrapolates with caveats. HL derives the fit, explains why records behave like minima and tests predictions using earlier data only.",
   "whereMarksAreLost": "E: one regression on Desmos with no justification. D: extrapolation stated as fact, and the naming of a future record holder is meaningless. B: R squared quoted without meaning.",
   "dataNote": "Only around 20 to 30 data points exist, so fitting is fragile and this must be discussed.",
   "verdict": "Keep to the floor question; who will break the record next is not a mathematical question. Test the model by fitting to old records and predicting recent ones."
  },
  {
   "id": "writing-polygons-as-single-equations-using-absolute-value",
   "topic": "2",
   "topicName": "Functions",
   "course": "both",
   "level": "both",
   "title": "Writing polygons as single equations using absolute value",
   "researchQuestion": "Can a regular polygon be described by one equation, and how closely does the equation of a smooth curve approach a polygon as a parameter grows?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Modulus functions and their transformations, piecewise linear functions, linear inequalities, equation of a line, intersections of lines, exponents in |x|^p + |y|^p = 1 as p grows, area by integration. Beyond syllabus: polar description of a polygon using cos and floor.",
   "howToStart": "Start with the square |x| + |y| = 1, then find an equation for a hexagon by combining absolute values of linear expressions, and plot in graphing software.",
   "slVsHl": "SL students build equations for a square, rhombus and hexagon with modulus and transformations. HL adds an area comparison for |x|^p + |y|^p = 1 by numerical integration and a limit argument as p tends to infinity, or the polar formula for the n-gon.",
   "whereMarksAreLost": "E: trial and error without derivation of why the equation works; B: unexplained software output; D: no test of generality beyond one polygon; A: too many screenshots.",
   "dataNote": "Graphing software makes checking easy, and the algebra stays within the syllabus.",
   "verdict": "A good base with a real modelling question. The p to infinity limit gives it more depth than just fitting polygons."
  },
  {
   "id": "do-sunflower-seed-spirals-really-follow-fibonacci-counts",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Do sunflower seed spirals really follow Fibonacci counts?",
   "researchQuestion": "How closely do the clockwise and anticlockwise spiral counts on real sunflower heads, pine cones and pineapples match consecutive Fibonacci numbers, and does a divergence angle near 137.5 degrees reproduce them?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 3,
   "namedInReports": true,
   "mathsRequired": "Recurrence relations and the limit of the ratio of consecutive terms, solved for the golden ratio via a quadratic. Polar coordinates for Vogel's model (r = c√n, θ = n×137.5°) plotted in software. Comparison of observed counts with predictions using percentage error. HL: Binet's formula, and a proof that rational approximations of the divergence angle change the spiral pattern (continued fractions as an extension).",
   "howToStart": "Collect 20 to 30 specimens (pine cones, pineapples, sunflower photographs), count both spiral families, and tabulate the pairs. Then code Vogel's model in a spreadsheet or Desmos and vary the angle from 137.0° to 138.0° to see when the spirals appear or vanish.",
   "slVsHl": "SL: count, tabulate, compare with Fibonacci pairs, and simulate the model with a few angles. HL: derive Binet's formula, prove the ratio limit, and explain why the golden angle is the most irrational angle.",
   "whereMarksAreLost": "E: stops at 'the numbers match' with no real mathematics. C: generic web facts instead of self-collected specimens. D: never reports the cases that fail to match. B: unexplained golden ratio results.",
   "dataNote": "Specimens are cheap but photographs need careful counting, and the sample should include some non-matching cases to be honest.",
   "verdict": "Very overdone, so it only works with your own specimens and the angle-sensitivity model. Skip the vague 'Fibonacci in nature' angle."
  },
  {
   "id": "how-much-space-do-equal-circles-waste-in-a-box",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "How much space do equal circles waste in a box?",
   "researchQuestion": "How does the density of square-grid and hexagonal packings of identical circles in a rectangle change with the number of circles, and how quickly does each approach its limiting density?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 3,
   "namedInReports": false,
   "mathsRequired": "Area of circles and polygons, trigonometry in equilateral triangles, density as a ratio, limits of a sequence as the number of circles grows, modelling boundary loss with a formula. Beyond syllabus: extension to spheres, FCC and HCP density (pi/sqrt 18).",
   "howToStart": "Derive the densities pi/4 and pi/(2 sqrt 3) for the infinite lattices. Then count how many circles of radius r fit in a 10 by 10 box for each arrangement and compute the real density.",
   "slVsHl": "SL compares densities with area formulae and a counting table. HL derives an edge-loss model, finds the limit as the box grows and can extend to 3D lattices with vectors.",
   "whereMarksAreLost": "E: only the two textbook densities with no analysis. B: unclear definition of the boundary. C: generic 'optimal packing' with no personal angle.",
   "dataNote": "Counting is easy to compute; the edge model needs some careful algebra.",
   "verdict": "Overdone in its general form. It becomes worthwhile if you model the finite-box effect and test it on something real, such as coins or oranges in a tray."
  },
  {
   "id": "fixed-tilt-of-a-solar-panel-over-a-year",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Fixed tilt of a solar panel over a year",
   "researchQuestion": "For a fixed south-facing panel at my latitude, which tilt angle maximises the energy received over one year, and how close is a geometric model to measured irradiance?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Declination modelled as a sinusoid δ = 23.45° sin(360(284+d)/365), noon solar altitude, cosine of the angle of incidence, sum or integral of cos(incidence) over the year, differentiation or a graphing search for the optimum tilt, sinusoidal regression, comparison with data. Beyond syllabus: including hour angle to model the whole day, not only noon.",
   "howToStart": "Take the latitude of your town, build the noon-only model in a spreadsheet for tilts from 0 to 90 degrees, and plot the yearly sum; then download hourly irradiance data for the same place from a public European or US solar database.",
   "slVsHl": "SL uses noon incidence only and a graphical or derivative-based optimum. HL adds the full daily integral with the hour angle, a comparison of seasonal against yearly optima and error analysis against measured data.",
   "whereMarksAreLost": "E: noon-only models can look like a single cosine exercise, and ignoring cloud and diffuse light limits depth. A: many formulas without a clear diagram. D: not questioning the clear-sky assumption.",
   "dataNote": "Irradiance data is public but needs cleaning and matching to the model's assumptions.",
   "verdict": "Solid and practical, with a definite answer that can be checked. Worth choosing if you do the daily integral; a friend's or family's real roof panel gives the personal hook."
  },
  {
   "id": "angle-excess-on-a-sphere-and-the-area-of-a-triangle",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Angle excess on a sphere and the area of a triangle",
   "researchQuestion": "How is the amount by which the angles of a spherical triangle exceed 180° related to its area, and can this be used to find the area of a triangle joining three real cities?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Great circles, area of a lune as a fraction of 4πr², inclusion and exclusion of overlapping lunes to prove Girard's theorem A = r²(α + β + γ - π), radian measure, the spherical cosine rule to find the angles from coordinates. Beyond the syllabus: Gauss-Bonnet, mentioned as the general idea.",
   "howToStart": "Prove the lune area 2r²θ first, then draw three overlapping lunes on a globe to make the excess appear. Choose three cities and convert their latitudes and longitudes to vectors.",
   "slVsHl": "SL can prove Girard's theorem geometrically, apply it to small and large triangles and compare with flat-map estimates. HL can add vector scalar products and the spherical cosine rule to compute the angles from coordinates and check the result.",
   "whereMarksAreLost": "B: spherical diagrams are hard to draw clearly. E: using a formula without proving it. C: the cities are chosen carelessly.",
   "dataNote": "Pure theory, with coordinates for cities easy to find.",
   "verdict": "Strong for a geometry-minded student. Choosing a triangle with personal meaning, such as home, family and a holiday place, gives it ownership."
  },
  {
   "id": "bouncing-off-a-wall-by-the-shortest-route",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Bouncing off a wall by the shortest route",
   "researchQuestion": "What is the shortest path from A to B that touches a given wall, and how does the method extend to a path that must touch two walls?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Reflection of a point in a line using coordinate geometry, the triangle inequality, gradients and equations of lines, minimising a sum of two distances by differentiation to confirm the reflection result, repeated reflections for two walls. Beyond syllabus: Fermat's principle and the link to Snell's law.",
   "howToStart": "Choose A and B on the same side of the x-axis, write the total length as a function of the contact point x, minimise it with calculus, then check that the answer matches the reflected-point construction.",
   "slVsHl": "SL can compare the calculus answer with the reflection method for a single wall and try two perpendicular walls, as in a snooker table. HL can generalise to walls at arbitrary angles, prove the path length is invariant under the reflection, and connect to Fermat's principle for light.",
   "whereMarksAreLost": "E: too little beyond a textbook optimisation problem if only one wall is treated. A: many similar diagrams with weak labelling. C: generic unless tied to a game or sport.",
   "dataNote": "Pure theory, easy to control, but the depth comes only from the two-wall and general-angle cases.",
   "verdict": "Easy to start but on the light side. It becomes a credible exploration if it reaches walls at any angle, or snooker or squash shots that the student tests on a real table."
  },
  {
   "id": "closest-approach-of-two-aircraft-on-straight-courses",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Closest approach of two aircraft on straight courses",
   "researchQuestion": "For two aircraft flying straight at constant velocities, at what time is the distance between them smallest, and what is that minimum distance?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Vector equations of lines r = a + tb; separation vector d(t); minimising |d(t)|² by differentiation; scalar product condition d·v_rel = 0; comparison with the shortest distance between skew lines. Beyond syllabus: relative-velocity frame, which reduces the problem to a point and a line.",
   "howToStart": "Invent two flights with positions in km and velocities in km/min, write r₁(t) and r₂(t), form d(t), and find the minimum of |d|² by calculus. Then confirm the answer with the perpendicular condition d·(v₁−v₂)=0.",
   "slVsHl": "SL students work in the plane with calculus and a graph of separation against time. HL students use 3D vectors with different altitudes, and compare the calculus result with the skew-line formula. Top-band work adds a safety threshold (say 5 nautical miles) and finds the time window in which it is breached.",
   "whereMarksAreLost": "E: stopping at one worked example with no generalisation. C: no personal link, such as real flight data. B: sloppy vector notation. D: no comment on the assumption of constant velocity.",
   "dataNote": "Pure theory with invented or real flight-tracker values; the algebra is easily controlled.",
   "verdict": "A sound, safe choice, but it drifts towards a textbook exercise unless generalised. Twist: use two real flight paths from a tracker and test them against a separation rule."
  },
  {
   "id": "counting-lattice-points-to-find-polygon-areas",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Counting lattice points to find polygon areas",
   "researchQuestion": "For which lattice polygons does area equal half the boundary count plus interior count minus one, and can I prove this from triangles and rectangles?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Area of triangles by coordinate methods (shoelace formula), counting lattice points on a segment using gcd, induction on triangulation, linear regression on collected data to conjecture the formula. Beyond syllabus: Euler's formula for planar graphs as an alternative proof.",
   "howToStart": "Draw 30 lattice polygons on squared paper, count interior and boundary points, compute area with the shoelace formula, then fit area against both counts to see the pattern before proving it.",
   "slVsHl": "SL students conjecture the formula from data, prove it for rectangles and right triangles, then extend to general triangles. HL depth comes from a full induction proof via triangulation, or a link to Euler's formula.",
   "whereMarksAreLost": "E: stopping at verification with examples instead of proving. B: skipped steps in the triangulation argument. C: a generic treatment with no personal angle.",
   "dataNote": "Pure theory with hand-made examples, and the proof is manageable if built up from triangles.",
   "verdict": "A sound, popular choice that only earns high marks if the proof is genuinely yours. A twist such as extending to 3D lattices or to polygons with holes makes it personal."
  },
  {
   "id": "do-famous-facades-really-follow-the-golden-ratio",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "AI",
   "level": "both",
   "title": "Do famous facades really follow the golden ratio?",
   "researchQuestion": "Measuring photographs of several buildings, do the ratios claimed to be golden differ from 1.618 by more than measurement error would allow?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Ratio and proportion, scaling from photographs, perspective correction with trigonometry, mean and standard deviation of measured ratios, confidence interval or t-test against 1.618. Beyond syllabus and reachable: the golden ratio from the recurrence and Fibonacci limit.",
   "howToStart": "Choose 6 to 8 buildings with straight-on photographs and known real dimensions. Measure width to height ratios in pixels, repeat each measurement five times, and compute the mean.",
   "slVsHl": "SL compares measured ratios with 1.618 using percentage error and spread. HL adds a hypothesis test, a correction for camera angle and a comparison against other candidate ratios such as the square root of two.",
   "whereMarksAreLost": "E: measuring and admitting near 1.6 with no test. C: repeating a well known claim. D: ignoring cherry picking of features. A: no clear method.",
   "dataNote": "Good undistorted images and dimensions take effort to find, and choosing which edges to measure introduces bias.",
   "verdict": "The golden ratio is named in the subject reports among topics seen too often, so a confirming version earns little. A sceptical statistical test, with the choice of edges fixed before you measure, turns it into real work."
  },
  {
   "id": "flying-straight-versus-drawing-a-straight-line-on-a-map",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "HL",
   "title": "Flying straight versus drawing a straight line on a map",
   "researchQuestion": "By how much does the great-circle distance between two cities differ from the straight-line distance measured on a Mercator map, and how does the gap depend on latitude?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Converting latitude and longitude to 3D position vectors, scalar product to find the central angle, arc length rθ, Mercator coordinates y = R ln(tan(π/4 + φ/2)), scale correction with sec φ, percentage difference across several city pairs. Beyond the syllabus: the rhumb line as a second route.",
   "howToStart": "Pick six city pairs at different latitudes and east-west separations, take their coordinates from a public gazetteer, and compute the vector-based distances. Compare with published flight distances as a check.",
   "slVsHl": "Vectors are HL content in both courses. An SL student can use the spherical cosine rule as a stated tool and put the effort into the projection comparison. HL students can derive the distance with the scalar product and find how the error depends on latitude and longitude difference.",
   "whereMarksAreLost": "E: the comparison with the map is often done by measuring with a ruler, which is not mathematics. B: notation for coordinates and angle units is muddled. D: no discussion of why airlines use a different route again.",
   "dataNote": "City coordinates are easy to collect, but the Mercator distance needs a clear definition of what 'straight on the map' means.",
   "verdict": "Worth choosing if you define the map distance precisely and analyse a pattern across routes. Pick routes you have actually flown."
  },
  {
   "id": "geometric-proofs-of-the-am-gm-inequality-and-their-limits",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "AA",
   "level": "both",
   "title": "Geometric proofs of the AM-GM inequality and their limits",
   "researchQuestion": "Which geometric constructions prove that the arithmetic mean of two positive numbers is at least the geometric mean, and which of them extend to three numbers?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Circle geometry with a semicircle and the altitude theorem, similar triangles, algebraic proof by (√a − √b)² ≥ 0, equality case, optimisation by differentiation as a comparison (minimum perimeter of a rectangle of fixed area). HL option: induction or Jensen-style argument for n numbers, and 3D box geometry for three variables.",
   "howToStart": "Draw a semicircle of diameter a + b, split at a point, and prove that the perpendicular height is √(ab). Then compare with the algebraic and calculus proofs.",
   "slVsHl": "An SL student can present two or three proofs of the two-variable case and apply them to a rectangle problem. HL depth comes from the three-variable case, induction for n terms, and a discussion of where geometry stops working.",
   "whereMarksAreLost": "E: risk of one textbook proof shown three times. B: figures unlabelled. C: little personal voice. D: no comparison of what each proof gives beyond the result.",
   "dataNote": "Pure theory, but the known proofs are short, so extension must come from the n-variable case or applications.",
   "verdict": "Neat but thin as it stands. Worth choosing only with an extension, for example a real optimisation problem such as the cheapest packaging box."
  },
  {
   "id": "how-much-of-the-earth-can-you-see-from-a-given-height",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "How Much of the Earth Can You See From a Given Height?",
   "researchQuestion": "How does the fraction of Earth's surface visible from an altitude h depend on h, and at what heights does it reach one percent and one quarter?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Right-angled trigonometry on a tangent to a sphere, cosine of the horizon angle, area of a spherical cap 2πR²(1 − cos θ), arc length, inverse trig functions. The visible fraction simplifies to h/(2(R + h)). Beyond syllabus: deriving the cap area with calculus, or with Archimedes' hat-box argument.",
   "howToStart": "Draw a cross section of the Earth with an observer at height h and a tangent line to the horizon. Show that cos(theta) = R/(R + h), then find the cap area 2 pi R squared (1 - cos theta) and evaluate for a mountain, an aeroplane and the ISS.",
   "slVsHl": "SL derives the horizon distance and the visible fraction from the cap area formula, and finds the heights for 1% (about 130 km) and 25% (one Earth radius). HL can derive the cap area with calculus (beyond the syllabus but within reach) and add atmospheric refraction or the flattened shape of the Earth.",
   "whereMarksAreLost": "E: sticking with a single formula for the cap area. D: no check of the result for a known case such as the Moon or a satellite photograph. B: unclear diagram of the tangent. C: pure theory can feel impersonal.",
   "dataNote": "It is pure geometry with constants from any reference, so proof is easy but there is no data to test against.",
   "verdict": "A neat, uncommon question with clean mathematics. Add a comparison with real observations, such as the horizon seen from a flight, and derive the cap area yourself to avoid being purely a formula exercise."
  },
  {
   "id": "inversion-in-a-circle-proving-what-it-does-to-lines-and-circles",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "AA",
   "level": "HL",
   "title": "Inversion in a circle: proving what it does to lines and circles",
   "researchQuestion": "How does inversion in a circle transform lines and circles, and can it be used to solve a tangent-circle construction problem?",
   "dataDifficulty": 2,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Definition OP·OP' = r², coordinates of the image point, equations of circles and lines, complex numbers as z → r²/z̄ for HL, proof that circles through the centre map to lines, Apollonius-type tangency problem. Beyond syllabus: link to the Poincaré disc model at a light touch only.",
   "howToStart": "Write the inverse of (x, y) in the unit circle as (x, y)/(x²+y²), substitute into the equation of a general circle and see what equation results, then check with GeoGebra's inversion tool.",
   "slVsHl": "Best at AA HL through complex numbers and algebraic proofs. An SL student can do the coordinate proofs for lines and circles, but should keep away from hyperbolic geometry.",
   "whereMarksAreLost": "E if the hyperbolic link is asserted without any maths; B if proofs are hard to follow; C if there is no problem being solved.",
   "dataNote": "Proofs are manageable, but the non-Euclidean connection is easy to overreach and should be kept small.",
   "verdict": "Good for a student who likes proof, provided it solves something concrete. Without the tangent-circle application it becomes a list of theorems."
  },
  {
   "id": "measuring-a-tall-building-without-touching-it",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Measuring a tall building without touching it",
   "researchQuestion": "Which of three trigonometric methods gives the most reliable height for a building I cannot climb, once angle and distance errors are taken into account?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Right-angled trigonometry, sine rule and cosine rule in non-right triangles, angles of elevation from two points, percentage error, error bounds found by taking extreme values of the measured angles and lengths. Beyond syllabus: differentiating the height formula to get the error sensitivity, reachable at HL.",
   "howToStart": "Choose a building with a published height. Measure angles of elevation with a clinometer from one point, from two points in line, and from two points not in line with the base, recording lengths with a tape.",
   "slVsHl": "SL: three methods, repeated measurements and error bounds by extreme values. HL depth: derivatives of each height formula with respect to angle and baseline, which show why the method with a longer baseline or moderate angle is best.",
   "whereMarksAreLost": "E: trigonometry stays at basic right triangles. B: errors quoted without explaining how they were found. D: no evaluation of the measuring instrument. A: photos and tables without structure.",
   "dataNote": "Fieldwork is easy but needs enough repeated readings, and access to open ground around the building.",
   "verdict": "Well liked and practical, which helps personal engagement, but it can be a textbook exercise. The error analysis is what makes it a proper exploration."
  },
  {
   "id": "polygons-with-fixed-perimeter-chasing-the-circle",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Polygons with fixed perimeter chasing the circle",
   "researchQuestion": "If the perimeter is held at 1 unit, how quickly does the area of a regular polygon close the gap to the area of the circle of the same perimeter, and how many sides are needed to come within 1% or 0.1%?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Dividing the polygon into n isosceles triangles and using area = ½ab sin C to get A(n) = P²/(4n tan(π/n)). Radian measure and the exact values for n = 3, 4, 6. Comparison with the circle's area P²/(4π). The gap G(n) = P²/(4π) − A(n) studied numerically and graphically. Limit as n tends to infinity via sin x ≈ x and cos x ≈ 1 − x²/2, or Maclaurin series for tan x (HL) to show G(n) is roughly proportional to 1/n². Beyond syllabus: the isoperimetric inequality, and why the circle is the maximiser.",
   "howToStart": "Derive A(n) for the triangle, square and hexagon by hand, then check the general formula against them. Next, tabulate A(n) and the ratio A(n)/(circle area) for n from 3 to 1000 in a spreadsheet or Desmos, and plot the gap against n on a log-log scale to see the slope.",
   "slVsHl": "SL: derive the formula, tabulate, graph, and estimate the limit numerically with a clear argument from small-angle behaviour. HL: expand tan x as a Maclaurin series to show that the gap is about π²/(3n²) of the circle's area, so about 19 sides are needed for 1% and 58 for 0.1%, then test this against the data and compare fixed perimeter with fixed radius.",
   "whereMarksAreLost": "E: stopping at a table of values with no proof of the limit, or quoting the formula without deriving it. C: no personal angle, so it reads like a textbook. B: unclear notation for n, P and the apothem. D: no comment on why the fixed perimeter constraint matters or on the limitations of the approximations.",
   "dataNote": "Pure theory with easy numerical checks, and the algebra stays controllable throughout.",
   "verdict": "Sound but familiar, and it can easily become a textbook exercise. It is worth choosing only with a twist: study the rate of convergence, then compare fixed perimeter with fixed radius, or link it to the isoperimetric problem so the conclusion is not just 'it tends to a circle'."
  },
  {
   "id": "rolling-on-a-reuleaux-triangle-constant-width-curves",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Rolling on a Reuleaux triangle: constant width curves",
   "researchQuestion": "Why does every curve of constant width w have perimeter πw, and how do the areas of different such curves compare?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Arc length s = rθ with radians; sector and triangle areas; decomposition of the Reuleaux triangle; generalisation to Reuleaux polygons with an odd number of sides; comparison of areas. Beyond syllabus: Barbier's theorem proved via support functions.",
   "howToStart": "Construct a Reuleaux triangle of width 1 from three arcs. Compute its perimeter and area by hand, then repeat with a Reuleaux pentagon and compare with the circle.",
   "slVsHl": "SL works out perimeter and area for the triangle and pentagon and spots the pattern. HL can give a general formula for a Reuleaux n-gon, and use integration to model the rolling motion of the shape between two parallel lines.",
   "whereMarksAreLost": "E: only two or three shapes computed, with no general proof. B: unexplained geometry in the constructions. C: reads as a summary of known results. D: no evaluation of which claims were proved and which only observed.",
   "dataNote": "Pure theory; the general n-gon formula is doable but needs care with angles.",
   "verdict": "Visual and accessible, with a good twist: build a physical model and measure its width. Push beyond the triangle to keep the mathematics substantial."
  },
  {
   "id": "shoelace-formula-proof-orientation-and-real-land-plots",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Shoelace formula: proof, orientation and real land plots",
   "researchQuestion": "Why does the shoelace formula give the area of a polygon, why does the sign depend on vertex order, and how accurate is it on real map boundaries?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Coordinate geometry, area of a trapezium under each edge, signed area as a sum of determinants, proof by triangulation using the cross product, and checking with the integral of x dy around a boundary. HL option: link to Green's theorem idea using integration, and vector cross product in 3D for polygons in space. Beyond syllabus: Green's theorem, if kept qualitative.",
   "howToStart": "Take a simple quadrilateral, compute the area by splitting it into triangles, then by the formula. Then extract vertex coordinates for a local park or field from OpenStreetMap or Google Earth.",
   "slVsHl": "An SL student can prove the formula for triangles and quadrilaterals and apply it to a real plot. HL depth comes from the general proof by induction on vertices, the treatment of non-convex and self-intersecting polygons, and the integral view.",
   "whereMarksAreLost": "E: applying a formula without proof is only a routine calculation. B: proof by induction poorly explained. C: real data absent. D: no discussion of map projection or coordinate error.",
   "dataNote": "Coordinates from maps need converting from latitude and longitude to metres, which must be explained.",
   "verdict": "Decent if the proof and a messy real polygon are the core. Twist: measure a local park and compare with the official area."
  },
  {
   "id": "squeezing-with-doubling-polygons-and-a-numerical-trap",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Squeezing π with doubling polygons and a numerical trap",
   "researchQuestion": "How quickly do the perimeters of inscribed and circumscribed polygons, doubled repeatedly from a hexagon, trap π, and why does the naive recurrence give wrong answers after a few steps on a computer?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Half-angle identity tan(θ/2) = sinθ/(1+cosθ) or its sine form, recurrence relations for side lengths, upper and lower bounds, error ratio per doubling, rounding error and cancellation in subtraction of nearly equal numbers, rationalising to rewrite a formula. Beyond the syllabus: Richardson-style averaging of the two bounds.",
   "howToStart": "Start with a hexagon in a unit circle, write the side length recurrence s_{2n} = √(2 - √(4 - s_n²)), code it in a spreadsheet at full precision, and watch where it drifts away from π.",
   "slVsHl": "SL can derive the recurrence, tabulate the bounds and observe the error shrinking by roughly a factor of four per doubling. HL can prove that factor from series expansions and explain the cancellation quantitatively, then derive the stable rewritten recurrence.",
   "whereMarksAreLost": "E if the numerical failure is only reported and not explained algebraically. A: long tables of digits. D: the comparison with modern methods for π is often skipped.",
   "dataNote": "Computation only, but the student must reach enough digits to see the breakdown, which needs a tool with adjustable precision.",
   "verdict": "Good choice because the breakdown is a genuine discovery and the algebraic fix is real maths. It is a known Archimedes topic, so the numerical-stability angle is what makes it distinctive."
  },
  {
   "id": "steiner-points-and-the-cheapest-road-network",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Steiner points and the cheapest road network",
   "researchQuestion": "For three or four towns, how much shorter is the best network with an extra junction than the shortest network joining the towns directly?",
   "dataDifficulty": 2,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Cosine rule, trigonometry in triangles, proof that the optimal junction has three 120 degree angles, optimisation by differentiation of total length, coordinates and vectors. HL: Fermat point via vector sums equal to zero.",
   "howToStart": "Take an equilateral triangle with side 1, find the junction that minimises the sum of distances, and compare with the spanning tree length of 2. Then try an isosceles triangle with a varying apex angle.",
   "slVsHl": "SL differentiates the total length for a symmetric configuration and compares it with the direct tree. HL proves the 120 degree condition with vectors, and handles the square with two junctions.",
   "whereMarksAreLost": "E: quoting the 120 degree result without a derivation. B: unclear diagrams. C: no real map or context. D: not saying that the general problem is very hard.",
   "dataNote": "Small cases are fully provable, but four or more points get messy and should be limited to the square or a rectangle.",
   "verdict": "Strong if kept to three or four points, with a real map of the student's region as the personal angle. Avoid promising a general algorithm."
  },
  {
   "id": "stellar-parallax-how-much-error-can-a-distance-survive",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Stellar parallax: how much error can a distance survive?",
   "researchQuestion": "How does the uncertainty in a measured parallax angle propagate into the calculated distance of nearby stars, and at what distance does the method stop being useful?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Right-angled trigonometry with very small angles, small angle approximation tan θ ≈ θ in radians, unit conversion between arcseconds, radians, parsecs and light years, percentage error and error propagation, reciprocal relationship d = 1/p. HL option: differentiate d(p) to bound the error with a linear approximation.",
   "howToStart": "Download parallax values with their uncertainties for 15 to 20 stars from the Gaia archive or the SIMBAD database. Compute distances with exact tan and with the approximation and compare.",
   "slVsHl": "An SL student can compare exact and approximate distances and compute percentage errors. HL depth comes from differentiating d = 1/p, deriving relative error formulas, and modelling how the error grows with distance.",
   "whereMarksAreLost": "E: mostly a single trig ratio unless error analysis is done. C: astronomy facts with no personal decision. B: unit conversions unexplained. D: no critique of the small angle assumption.",
   "dataNote": "Gaia data is public but needs filtering to remove entries with poor precision.",
   "verdict": "Good if the error analysis is the centre and not the basic parallax triangle. Twist: pick the stars from a constellation the student can see from home."
  },
  {
   "id": "the-envelope-traced-by-a-sliding-ladder",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "AA",
   "level": "both",
   "title": "The envelope traced by a sliding ladder",
   "researchQuestion": "What curves are traced by points at different fractions along a ladder that slides down a wall, and what boundary do all the ladder positions enclose?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Parametric equations using the angle to the ground, Pythagoras and trigonometric identities to eliminate the parameter, ellipse equations, the astroid x^(2/3) + y^(2/3) = L^(2/3) as the envelope of the ladder positions. Beyond syllabus: the envelope condition, found by differentiating the family of lines with respect to the parameter, and the energy argument for when a real ladder leaves the wall.",
   "howToStart": "Place the ladder with its foot at (L cos t, 0) and top at (0, L sin t). Write the position of a point at fraction k along it, eliminate t, and show it is an ellipse. Plot k values in Desmos.",
   "slVsHl": "SL students derive the circle traced by the midpoint and the ellipses traced by other points, and find the astroid from the family of lines. HL students prove the envelope with the derivative condition and model a real ladder that leaves the wall.",
   "whereMarksAreLost": "E: only stating that the midpoint traces a circle without any derivation for other points. B: diagrams missing or unlabelled. C: no personal angle or experiment.",
   "dataNote": "Pure theory that is easy to check with dynamic geometry software.",
   "verdict": "A neat problem with plenty of room for proof. The twist is the real ladder, which leaves the wall, so the ideal model can be tested against a physical one."
  },
  {
   "id": "voronoi-cells-and-where-the-next-clinic-goes",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "AI",
   "level": "both",
   "title": "Voronoi cells and where the next clinic goes",
   "researchQuestion": "Where should one extra facility be located in a chosen town so that the greatest distance from any resident to their nearest facility is as small as possible, and how does this change when real road distance replaces straight-line distance?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Coordinates, distance and midpoint formulae, gradients and equations of perpendicular bisectors, construction of Voronoi cells, circumcentres of triangles, comparison of Euclidean and Manhattan metrics. Beyond syllabus if reachable: Delaunay triangulation.",
   "howToStart": "Take a map of a real town from OpenStreetMap, mark 5 to 8 existing facilities, and convert positions to coordinates. Draw the perpendicular bisectors by hand for three of them and check the cell edges.",
   "slVsHl": "SL builds the diagram and finds the vertex farthest from all sites as the candidate. HL proves why the circumcentre is equidistant, treats the boundary of the area, and models the road grid with a different metric.",
   "whereMarksAreLost": "E: only drawing diagrams with software and no algebra; B: unclear justification of why a Voronoi vertex is optimal; D: never testing the straight-line assumption.",
   "dataNote": "Coordinates come from an open map such as OpenStreetMap and the geometry is fully controllable.",
   "verdict": "Voronoi diagrams of universities or shops are named in the subject reports as a school template seen too often. It is worth choosing only if you do the algebra of the bisectors yourself, use a place you know, and test the straight-line assumption against road distance."
  },
  {
   "id": "where-to-put-ambulances-minimax-against-mean-distance",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "AI",
   "level": "both",
   "title": "Where to put ambulances: minimax against mean distance",
   "researchQuestion": "For a region I define, how do the best positions of two or three response units change when the aim is the smallest worst-case distance rather than the smallest average distance?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Coordinate geometry, perpendicular bisectors, Voronoi diagram construction, distance formula, minimising a sum of distances and a maximum distance, weighting by population. Beyond syllabus: the geometric median found by iteration.",
   "howToStart": "Take a small area with about 15 population centres, record coordinates and populations, and place two units on a grid. Compute both objectives for a few placements, then draw the Voronoi regions for the best ones.",
   "slVsHl": "SL: two units, straight-line distance, tables and graphs comparing both objectives. HL depth: a systematic search or iteration for the mean-distance optimum, weighting by population, and road distances in place of straight lines.",
   "whereMarksAreLost": "E: search by trial only, with no reasoning about why the optimum lies where it does. B: undefined objectives. D: the ethical comparison stays a single sentence. A: cluttered diagrams.",
   "dataNote": "Population and location data need cleaning, and the optimum for the mean objective may need software.",
   "verdict": "Rich and personal if you use your own region, and the ethical question of fairness gives real reflection. Keep the number of units small or the maths gets out of hand."
  },
  {
   "id": "which-frequency-ratios-close-a-lissajous-curve-and-how-many-crossings",
   "topic": "3",
   "topicName": "Geometry and trigonometry",
   "course": "both",
   "level": "both",
   "title": "Which frequency ratios close a Lissajous curve and how many crossings result",
   "researchQuestion": "For x = sin(at + φ) and y = sin(bt), how do the ratio a:b and the phase φ determine the period, the number of self-crossings and the symmetry of the curve?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Parametric equations, periodicity of sine, double-angle and sum-to-product identities, symmetry of functions, lowest common multiples and rational ratios. Differentiation of parametric curves for tangents at crossings (HL). Beyond syllabus: for x = cos t and y = cos(bt), y is a Chebyshev polynomial of degree b in x.",
   "howToStart": "Plot curves in Desmos or Python for a:b = 1:2, 2:3, 3:4 and several φ. Tabulate crossings and lobes, then guess a formula in a and b.",
   "slVsHl": "SL can plot, prove the closing condition and count lobes. HL can find the crossing points exactly, use dy/dx as a function of t, and derive the Chebyshev link for the cosine case.",
   "whereMarksAreLost": "E: pattern spotting without proof. A: too many pictures with no commentary. D: reflection limited to 'it looks nice'.",
   "dataNote": "Data are generated by the student's own plots, so nothing to collect.",
   "verdict": "Worth choosing. It is common enough not to be original by default, so the twist should be a proved crossing formula, or a physical link such as an oscilloscope or a double pendulum recording."
  },
  {
   "id": "monty-hall-with-n-doors-and-a-forgetful-host",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Monty Hall with n doors and a forgetful host",
   "researchQuestion": "How does the probability of winning by switching change when there are n doors, the host opens k losing doors, and the host may or may not know where the prize is?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 4,
   "namedInReports": false,
   "mathsRequired": "Conditional probability; tree diagrams; law of total probability; Bayes' theorem for the case when the host opens a door at random; generalising to a formula in n and k; verification by a simulation with a large number of trials. Beyond syllabus: expected values of strategies over repeated games.",
   "howToStart": "State the rules for the classic game and compute the switching probability with a tree diagram. Then write a short simulation of 100,000 games for n = 3, then for n = 5 with k = 1, 2, 3, and compare it with your formula.",
   "slVsHl": "SL builds trees and derives the general formula for n and k. HL adds Bayes' theorem in the case of a random host and the discussion of how the information given by the host's action changes the result.",
   "whereMarksAreLost": "C: a very familiar problem with no personal angle. E: staying on the 3-door case that appears in every textbook. D: no comment on the sensitivity of the result to unstated assumptions about the host. B: ambiguous rules.",
   "dataNote": "Pure theory with a simulation for checking, so no data is needed, but the rules must be stated with precision.",
   "verdict": "Overdone in its basic form. It only works with the forgetful host case and the general n and k formula, which is where the real mathematics is."
  },
  {
   "id": "queueing-at-a-real-junction-arrivals-service-and-green-time",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "HL",
   "title": "Queueing at a real junction: arrivals, service and green time",
   "researchQuestion": "Can a Poisson arrival model and a fixed discharge rate predict queue lengths at a local traffic light, and what green-time split minimises the average wait?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 3,
   "namedInReports": false,
   "mathsRequired": "Poisson distribution for arrivals, chi-squared goodness of fit test, expected queue length from a deterministic queue model, optimisation by differentiation of total delay with respect to green time. Beyond syllabus: the M/M/1 queue formula L = ρ/(1 − ρ) and its breakdown as ρ tends to 1.",
   "howToStart": "Film a junction for 30 minutes and count cars arriving in every 10-second interval. Test the counts against a Poisson distribution, then measure how many cars leave per second of green.",
   "slVsHl": "The Poisson distribution is AI HL. HL students fit and test the Poisson model, build a delay function and optimise the green split by differentiation, then compare with the M/M/1 result. An SL student could model arrivals per interval with a binomial distribution instead.",
   "whereMarksAreLost": "E: listing formulae without using them on real data. B: undefined symbols such as rho. C: no observation made by the student. D: no discussion of rush-hour counts breaking the Poisson assumption.",
   "dataNote": "Counting cars takes a few hours of recording and tidying, but produces authentic data.",
   "verdict": "With a real junction and a testable model this is strong. The twist is a junction you use every day."
  },
  {
   "id": "bluffing-frequency-in-a-simplified-poker-game",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Bluffing frequency in a simplified poker game",
   "researchQuestion": "In a one-card betting game, what mix of bluffing and calling makes a player's expected winnings unexploitable?",
   "dataDifficulty": 2,
   "dataSource": "theory",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Combinatorics for hand counts, conditional probability, expected value, payoff matrices, solving linear equations for the indifference condition, mixed strategy. Beyond syllabus: Nash equilibrium in a simplified Kuhn poker.",
   "howToStart": "Set up a three-card game with one bet size. List every deal and every action, then find the expected payoff of each pure strategy pair in a table.",
   "slVsHl": "SL works with the payoff table and expected values. HL solves the equilibrium mixing probabilities with algebra and may generalise to a variable bet size.",
   "whereMarksAreLost": "E: real Texas Hold'em is too big, so the maths becomes descriptive. B: game rules poorly explained. D: failure to test the model against play.",
   "dataNote": "Mathematics is controllable only if the game is cut down; testing against real play needs self-collected hands.",
   "verdict": "The full poker version is too broad and heavily used. A tiny game solved exactly, then played by classmates against the prediction, turns it into a good exploration."
  },
  {
   "id": "buffon-s-needle-as-a-way-of-finding",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Buffon's needle as a way of finding π",
   "researchQuestion": "How many needle drops are needed to estimate π to a given accuracy, and does the choice of needle length relative to line spacing make the estimate more efficient?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Uniform distributions for the centre distance and angle, P = 2l/(πd) by integrating over the angle, expectation and variance of a binomial count, standard error of the π estimate, simulation, dependence of variance on l/d. Beyond syllabus: using the delta method to approximate the variance of the π estimate.",
   "howToStart": "Derive P for l ≤ d, then write a spreadsheet or short program that drops 1000 random needles and estimates π; repeat 100 times to see the spread.",
   "slVsHl": "SL does the geometric derivation with a simple integral and a simulation. HL or top work examines variance against l/d, shows the best ratio and quantifies the convergence, including a confidence interval.",
   "whereMarksAreLost": "E: a simulation without a derivation or error analysis is too thin. C: a very common topic, so no physical experiment or personal angle. B: unclear model of what a random drop means.",
   "dataNote": "Simulation is easy; a real throw of needles is optional but slow.",
   "verdict": "Popular, so it needs the efficiency question to stand out. Adding a real drop of 500 matchsticks makes it your own."
  },
  {
   "id": "can-standing-height-predict-a-volleyball-player-s-spike-reach",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Can standing height predict a volleyball player's spike reach?",
   "researchQuestion": "How well does a linear model based on standing height predict the spike reach of professional volleyball players, and does the fit differ between outside hitters and middle blockers?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": true,
   "mathsRequired": "Scatter diagram, Pearson's correlation coefficient, least-squares regression line of y on x, the coefficient of determination, residual analysis with a plot of residuals against fitted values, interpolation versus extrapolation, and a two-sample t-test (AI) on the residuals of two positions, with a check of the normality assumption. Beyond syllabus: a confidence interval for the slope, or separate lines for each position with a comparison of gradients.",
   "howToStart": "Pick one recent international tournament, such as a Nations League or Olympic squad list, and copy height, spike reach and position for every player into a spreadsheet. Record the source of each row. Draw the scatter plot first and compute r and the regression line with a graphing calculator or spreadsheet before doing any testing.",
   "slVsHl": "An SL student can complete the scatter plot, r, the regression line and a careful interpretation of gradient and intercept, with a short residual discussion. HL depth comes from arguing why the test conditions hold, comparing position-specific models, and showing by hand why the least-squares line minimises the sum of squared residuals for a small data set.",
   "whereMarksAreLost": "E: stopping at r and a line of best fit with no test of the model's assumptions; treating a high r as proof of cause. B: unexplained calculator output and no definitions of spike reach. C: generic data with no personal link, such as no connection to the student's own team. D: no comment on the bias of using elite players only, or on measurement differences between sources.",
   "dataNote": "Data are public but need cleaning and consistent definitions, and a sample under about 30 per group weakens the t-test.",
   "verdict": "Two variables and a correlation is named in the subject reports as a common and usually weak format, so go beyond r. The residual analysis and position comparison give it substance, and measuring your own club team as an extra check makes it personal."
  },
  {
   "id": "cooperation-in-the-repeated-prisoner-s-dilemma",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Cooperation in the repeated prisoner's dilemma",
   "researchQuestion": "How does the probability that a game continues affect which strategy earns most in a repeated prisoner's dilemma?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Payoff matrices, expected value, geometric series for an infinite repeated game with continuation probability p, comparing sums of series, inequalities to find the critical p. Beyond syllabus: Markov chains for strategies with memory.",
   "howToStart": "Choose payoffs such as 3, 5, 1, 0. Compute by hand the total for always defect against tit-for-tat over n rounds. Then find the p above which tit-for-tat beats defection.",
   "slVsHl": "SL sums geometric series for a few strategy pairs. HL builds a round robin tournament and analyses strategies with memory using transition matrices.",
   "whereMarksAreLost": "E: one-shot game only, which is just reading a table. B: strategy definitions ambiguous. C: retelling Axelrod's tournament without original work.",
   "dataNote": "Simulation is simple; the series analysis gives the real mathematics.",
   "verdict": "The one-shot version is weak and known to moderators. The repeated version with a derived critical probability is a solid exploration."
  },
  {
   "id": "do-leading-digits-expose-invented-numbers",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Do leading digits expose invented numbers?",
   "researchQuestion": "Do leading digits in several real datasets match the logarithmic distribution of Benford's Law, and can the same test separate them from numbers I invent or that classmates invent?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Probability of a leading digit as log10(1 + 1/d), properties of logarithms, chi-squared goodness of fit test with degrees of freedom and p value, comparison of expected and observed frequencies. Beyond syllabus: why scale invariance forces the logarithmic law.",
   "howToStart": "Take populations of countries and a set of river lengths or share prices, tally first digits in a spreadsheet, then compute the expected counts from the formula and a first chi-squared value.",
   "slVsHl": "SL derives the expected proportions, runs the chi-squared test on two or three datasets and compares with invented data. HL or top SL adds a scale-invariance or geometric-series argument for why the law arises, and examines why some datasets fail, such as heights.",
   "whereMarksAreLost": "E: applying chi-squared with expected counts below 5 or with a huge sample that rejects everything. B: not defining hypotheses. D: claiming the test proves fraud. C: a purely descriptive account of the law.",
   "dataNote": "Plenty of open data; the care needed is in choosing sets that span several orders of magnitude.",
   "verdict": "Very popular, so a routine version will look familiar. Make it distinctive with an experiment where classmates fabricate data and you measure how well the test catches them."
  },
  {
   "id": "do-rock-paper-scissors-players-follow-a-predictable-markov-chain",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Do rock-paper-scissors players follow a predictable Markov chain?",
   "researchQuestion": "How well does a first-order Markov chain, fitted to recorded rock-paper-scissors games, predict a player's next throw, and does a strategy built on that chain beat a player choosing at random?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Relative frequencies estimated into a 3 by 3 transition matrix; conditional probability P(next | previous); expected value of a strategy from its payoff table; a chi-squared test of independence between consecutive throws; matrix powers to find the long-run distribution; HL: the steady state as an eigenvector with eigenvalue 1, and the speed of convergence from the second eigenvalue. Beyond syllabus: comparing first-order with second-order chains by counting parameters.",
   "howToStart": "Recruit 15 to 20 classmates, ask each for 60 to 100 throws against a computer or a hidden partner, and log each sequence in a spreadsheet. Then count the nine previous-to-next pairs for one player and turn the counts into a transition matrix.",
   "slVsHl": "SL builds the matrix from counts, computes conditional probabilities, and tests whether the win rate of a follow-the-pattern strategy exceeds one third using simulation or a simple test. HL derives the steady state as an eigenvector, links it to the fairness of the game, and compares the fitted chain with the null model of independent uniform throws, with a discussion of overfitting when only 60 throws feed nine parameters.",
   "whereMarksAreLost": "E: matrix and steady state done mechanically with no test of whether the Markov assumption fits; C: generic survey with no personal angle; D: no discussion of small samples, players who adapt when they lose, or the difference between a fitted pattern and a real one; B: notation for state, row and column convention left unclear.",
   "dataNote": "Collecting enough throws per player is the bottleneck, since nine matrix entries need several hundred throws in total for stable estimates, though pooling players or using an online game log solves it.",
   "verdict": "A sound choice that is only moderately overused, provided the test of the Markov assumption is real and not decorative. The personal twist is to play against one specific opponent, such as a sibling or a friend, and see whether the chain exposes their habits, then whether the habit disappears once they know."
  },
  {
   "id": "how-wrong-can-a-poll-be-sampling-error-against-actual-results",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "How wrong can a poll be? Sampling error against actual results",
   "researchQuestion": "How often did the published margin of error of UK opinion polls contain the actual election vote share, and what does this suggest about the assumptions behind a normal model?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Sampling distribution of a proportion, normal approximation, 95% confidence interval p ± 1.96√(p(1 − p)/n), comparison with actual results, mean error and root mean squared error, simulation of repeated polls. HL: hypothesis test or chi-squared on the frequency of misses, and the effect of design effects or house bias.",
   "howToStart": "Download the final polls before three or four past UK general elections from a public archive, with sample sizes, and compute each margin of error. Then tabulate whether the true party share lay inside the interval.",
   "slVsHl": "SL: compute intervals, count coverage and simulate. HL: test whether the miss rate exceeds 5% and model systematic bias using regression of error against date or pollster.",
   "whereMarksAreLost": "E: simply drawing a regression line through polls without justification. C: reflecting no personal interest, and stale politics. D: not discussing non-sampling error such as late swing.",
   "dataNote": "Data is public but pollster methodology and sample sizes have to be cleaned and matched to results.",
   "verdict": "Election poll ideas are common, so focus on the coverage question instead of predicting the next winner, and pick elections you have a reason to care about."
  },
  {
   "id": "income-and-life-expectancy-across-countries",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Income and life expectancy across countries",
   "researchQuestion": "Is life expectancy better modelled as linear in GDP per capita or in its logarithm, and how has that relationship changed between 1990 and the latest year?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": true,
   "mathsRequired": "Scatter plots, Pearson and Spearman correlation, least-squares regression, a log transformation of GDP, residual plots, coefficient of determination, comparison of the fits. Beyond syllabus: explaining why diminishing returns suggest a logarithmic model.",
   "howToStart": "Download GDP per capita and life expectancy for all countries for two years from Gapminder or a World Bank table, remove missing entries, and draw the raw scatter plot.",
   "slVsHl": "SL fits linear and log models and compares r and residuals. HL adds Spearman's rank correlation (AI), residual analysis and a non-linear fit; the extra depth comes from comparing years.",
   "whereMarksAreLost": "E: only computing r on raw data is too shallow. C: a very common topic, so choosing a niche question is essential. D: treating correlation as causation, and ignoring outliers or confounders such as health spending.",
   "dataNote": "Clean public data exists, but cleaning missing countries needs a stated rule.",
   "verdict": "Two variables and a correlation is named in the subject reports as a common and usually weak format. Choose it only for the log transformation and the change over time, or link it to a country you know."
  },
  {
   "id": "is-the-spreadsheet-s-random-number-generator-really-uniform",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Is the spreadsheet's random number generator really uniform?",
   "researchQuestion": "To what extent do the values from a spreadsheet's random function behave as independent draws from a uniform distribution on (0,1)?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Uniform distribution properties, frequency tables with equal bins, expected frequencies, chi-squared goodness-of-fit, correlation between consecutive values, a runs test or a chi-squared test on pairs. Beyond syllabus: a Kolmogorov-Smirnov statistic.",
   "howToStart": "Generate 5000 values, bin them into ten classes and calculate the chi-squared statistic and p-value. Then plot each value against the next and test the pair counts.",
   "slVsHl": "SL tests uniformity and correlation between consecutive values. HL adds a test on pairs with a two-dimensional chi-squared, a comparison with a simple linear congruential generator built by hand using modular arithmetic, and a power analysis.",
   "whereMarksAreLost": "E: one test on one sample with no consideration of repeated runs. D: treating 'p above 0.05' as proof of randomness. B: unclear hypotheses.",
   "dataNote": "Data is instant, but repeated runs and the correct degrees of freedom need care.",
   "verdict": "Neat and safe. The twist that lifts it is building your own poor generator and showing your tests catch it, a good way to judge test power."
  },
  {
   "id": "minimum-variance-weights-for-a-three-asset-portfolio",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Minimum-variance weights for a three-asset portfolio",
   "researchQuestion": "For three stocks chosen from different sectors, which weights minimise portfolio variance for a required expected return, and how far does the fitted frontier predict the next year's actual results?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Expected value and variance of a weighted sum of returns, which brings in covariance (beyond the syllabus, but a short step from variance); for two assets the variance is a quadratic in the weight, minimised by differentiation or completing the square; for three assets, substitute w₃ = 1 − w₁ − w₂ and solve the linear system from setting both derivatives to zero; out-of-sample comparison. AI HL students can write the variance in matrix form. Beyond syllabus: Lagrange multipliers and the efficient frontier.",
   "howToStart": "Download five years of monthly closing prices for three stocks from Yahoo Finance, convert to monthly returns, compute means and the covariance matrix, then find the minimum-variance weights for two assets by hand before moving to three.",
   "slVsHl": "An SL student can do two assets with the variance as a quadratic in w, then compare a few chosen three-asset weightings. HL depth comes from the three-asset system, the matrix formulation and testing the weights on a later period.",
   "whereMarksAreLost": "E: using Excel Solver as a black box; B: matrix notation introduced without explanation; D: no comment on the assumption that past covariances persist; C: generic choice of stocks with no reason.",
   "dataNote": "Prices are publicly available; cleaning is minor, though returns must be computed consistently.",
   "verdict": "Common, but it works if the optimisation is done by hand and validated on later data. Choose assets tied to your own interest and be honest that the model often fails out of sample."
  },
  {
   "id": "mixed-strategies-in-penalty-kicks-theory-against-real-shots",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Mixed strategies in penalty kicks: theory against real shots",
   "researchQuestion": "Which left/right mixture leaves a penalty taker with no exploitable weakness, and how close do professional kickers and keepers come to it in real shootout data?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Payoff matrices of scoring probabilities, expected value, indifference conditions that give a mixed-strategy equilibrium by solving linear equations, then a chi-squared goodness-of-fit test (AI) comparing observed direction frequencies with the predicted mix. Beyond syllabus: the minimax idea behind Nash equilibrium, and adding a third option (centre).",
   "howToStart": "Take penalty records for one league or tournament from a public football dataset, tabulate kicker side (natural or weaker foot) against keeper dive direction, and compute the four scoring rates. Then solve for the mix that makes the keeper indifferent.",
   "slVsHl": "SL solves the 2 by 2 game with expected value and simultaneous linear equations and compares it with observed proportions. HL adds a sensitivity analysis of the equilibrium to sampling error, AI HL confidence intervals on the scoring rates, and a 3 by 3 extension including the centre.",
   "whereMarksAreLost": "E: stopping at a textbook 2x2 game with no testing of the data. C: generic football enthusiasm with no personal angle. D: not questioning the assumption that kickers choose simultaneously and rationally. B: undefined game theory vocabulary.",
   "dataNote": "Penalty data exists online but cell counts can be small, and direction has to be coded as natural side versus weaker side consistently.",
   "verdict": "A sound idea if the data is tested honestly. The twist is to use your own club's penalties or a single tournament, and to report where the equilibrium fails."
  },
  {
   "id": "shared-birthdays-uniform-theory-against-real-birth-data",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Shared birthdays: uniform theory against real birth data",
   "researchQuestion": "For a group of n people, how does the probability of a shared birthday change when real, uneven birth frequencies replace the uniform assumption, and what group size gives an even chance in each model?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": true,
   "mathsRequired": "Complementary events and the product of conditional probabilities, solving 1 − 365!/((365−n)!·365^n) = 0.5 by table or logarithms, Poisson-style approximation using exp(−n(n−1)/730), a weighted model with daily probabilities p_i and the sum of p_i squared as a collision measure, Monte Carlo check. Beyond syllabus: a short argument that uniform probabilities minimise the collision chance (Cauchy-Schwarz).",
   "howToStart": "Download a public table of births per calendar day (for instance national birth statistics from a statistical office over several years), convert to relative frequencies, and first compute the uniform answer for n = 23 on a spreadsheet before touching the real data.",
   "slVsHl": "SL derives the uniform result, finds the threshold by trial and compares it with a simulation. HL or top-band work adds the weighted model, the approximation by exponentials and a proof or numerical demonstration that any unevenness raises the match probability.",
   "whereMarksAreLost": "C: the topic is the standard one, so personal engagement depends on using real data and a twist of one's own. E: staying with the uniform calculation only is too thin. B: unexplained use of factorials on a calculator. D: no comment on ignored leap days and clustering of births.",
   "dataNote": "Births by day are publicly available, though leap day and rounding need a stated decision.",
   "verdict": "The birthday problem is named in the subject reports among topics seen too often, so only choose it if real birth data and the comparison of models are the centre. Testing your own year group or school against the prediction makes it personal."
  },
  {
   "id": "sleep-hours-and-school-results-what-does-a-survey-show",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Sleep hours and school results: what does a survey show?",
   "researchQuestion": "How strongly is self-reported sleep duration associated with a defined measure of school performance in my school, and do a numerical and a banded analysis reach the same conclusion?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 2,
   "namedInReports": true,
   "mathsRequired": "Scatter plot, Pearson and Spearman correlation, least-squares regression, grouping into bands and a chi-squared test for independence with expected counts and degrees of freedom, a check that expected counts exceed 5, a discussion of the effect of banding choices.",
   "howToStart": "Design a short anonymous questionnaire with one question on average sleep and one on a clearly defined result (for example last term's mean mark banded), get school permission and collect at least 60 responses.",
   "slVsHl": "SL does correlation, regression and one chi-squared test. HL or top work tries different bands, checks sensitivity, uses a hypothesis test on the correlation and considers confounders such as stress or age.",
   "whereMarksAreLost": "E: using only r without testing or modelling. D: causal claims, self-report bias and the sample not being random. A: vague definition of the performance measure. C: a common topic with generic data.",
   "dataNote": "You need to collect enough honest responses and handle privacy and small expected counts.",
   "verdict": "Two survey variables and a correlation is named in the subject reports as a common and usually weak format. It is workable only with your own data, a strict definition of the performance measure, and honesty about association and cause."
  },
  {
   "id": "the-secretary-problem-and-the-1-e-stopping-rule",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "The secretary problem and the 1/e stopping rule",
   "researchQuestion": "For a sequence of n ranked candidates seen one at a time, how does the chance of picking the best or one of the top three change with the number rejected first, and how quickly does the best cut-off approach n/e?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Conditional probability and the sum over positions of the best candidate; harmonic sums; comparing exact probabilities for n = 5 to 50 with integer optimisation; Riemann sum turning the sum into an integral of (1/x) to give x ln(1/x) and maximising by differentiation (HL); Monte Carlo simulation to check the exact results. Beyond syllabus: the dynamic programming view of the same rule.",
   "howToStart": "Write out the exact success probability for n = 4 by listing all 24 orderings, then build a spreadsheet or Python function P(n,k) and simulate 100000 random permutations to confirm it. Then tabulate the best k for n up to 50.",
   "slVsHl": "SL students can do the exact discrete case, tabulate the best cut-off and compare with simulation. HL adds the integral limit, the derivation of 1/e by differentiation, and a variant where any of the top three counts as success, which changes the best rule.",
   "whereMarksAreLost": "E: stating the 37% rule without deriving the probability sum; B: unclear notation for 'best so far'; C: no personal context such as choosing a flat or a partner; D: not testing the rule's assumptions, such as no recall of rejected candidates.",
   "dataNote": "Pure theory plus a simple simulation; the only care needed is a clear derivation of the sum.",
   "verdict": "Well known, so moderators will have seen it, but the derivation is solid. Make it yours with a real setting, such as choosing a flat from viewings, and with a variant of the success rule."
  },
  {
   "id": "are-30-percent-rain-forecasts-really-right-30-percent-of-the-time",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Are 30 percent rain forecasts really right 30 percent of the time?",
   "researchQuestion": "How well calibrated are the daily rain probabilities issued for my town over a season, judged by observed outcomes?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Relative frequency, binned calibration table, conditional probability, binomial distribution, chi squared goodness of fit, Brier score as a mean of squared differences. Beyond syllabus: reliability diagram.",
   "howToStart": "Record or download forecast rain probability and whether it rained for at least 100 days. Group forecasts into bins of 10 percent and compare each bin's stated probability with the observed frequency.",
   "slVsHl": "SL builds the calibration table, compares frequencies and computes the Brier score. HL adds a hypothesis test per bin with the binomial and a comparison of two forecasters with a test on the mean scores.",
   "whereMarksAreLost": "E: only counting days without any test. B: unclear definition of what counts as rain. C: data from nowhere in particular.",
   "dataNote": "Archived forecasts are harder to obtain than observations, so you may have to log them daily yourself.",
   "verdict": "A good question with real data and a clear answer. The twist is collecting the forecasts yourself, so start now."
  },
  {
   "id": "bayes-and-the-prosecutor-s-fallacy-in-court",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Bayes and the prosecutor's fallacy in court",
   "researchQuestion": "How far can a match in DNA evidence mislead a jury when the probability of a match given innocence is confused with the probability of innocence given a match?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Conditional probability; tree diagrams; Bayes' theorem; sensitivity to the prior probability using tables and graphs of the posterior. Beyond syllabus: odds form of Bayes with likelihood ratios.",
   "howToStart": "Choose one real case where a statistical argument was used in court, such as the Sally Clark case. Write the two conditional probabilities that were confused, then compute the posterior for a range of priors.",
   "slVsHl": "SL builds tree diagrams and computes posteriors for several priors. HL or top band work adds the odds form, combining several independent pieces of evidence, and how the posterior grows with each.",
   "whereMarksAreLost": "E: only one Bayes calculation, which is a textbook exercise. C: no reason why the case matters to the student. D: no discussion of how priors are chosen. A: a long case narrative that buries the maths.",
   "dataNote": "Pure theory with published figures; the danger is making up figures without justifying them.",
   "verdict": "A sound choice if you vary the prior and treat several pieces of evidence. Keep the legal story short and let the algebra lead."
  },
  {
   "id": "breaking-vigenere-with-chi-squared",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Breaking Vigenere with chi-squared",
   "researchQuestion": "How accurately can a chi-squared statistic against English letter frequencies recover the key length and key letters of a Vigenere cipher, and how does accuracy depend on the message length?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Modular arithmetic for shifts; relative frequencies; chi-squared statistic and its comparison with critical values; simulation of accuracy against text length. Beyond syllabus: index of coincidence as an alternative.",
   "howToStart": "Take a public domain English text and encrypt extracts with random keys. Split a ciphertext into columns for a guessed key length and compute the chi-squared of each shift against English frequencies.",
   "slVsHl": "SL cracks a single cipher by hand and computes chi-squared for each shift. HL or top band work simulates thousands of ciphers, plots success rate against length and compares with the index of coincidence.",
   "whereMarksAreLost": "E: cracking one message shows only a procedure. B: code listings without explanation of the statistic. D: no discussion of short messages, where the test fails.",
   "dataNote": "All data can be generated by simulation; coding skill is the only requirement.",
   "verdict": "Strong, because the statistics does real work and you can measure your own method's success. The simulation of accuracy against length gives it personal depth."
  },
  {
   "id": "building-datasets-with-the-same-summary-statistics-but-different-shape",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Building datasets with the same summary statistics but different shapes",
   "researchQuestion": "Can I construct my own data sets that share mean, variance and correlation with a given set yet look completely different, and what does that say about summary statistics?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Mean, variance, Pearson correlation, least squares regression line, residuals, coefficient of determination. Beyond syllabus and reachable: solving simultaneous equations to force chosen statistics on a data set.",
   "howToStart": "Compute the statistics for the four classic Anscombe sets. Then try to build a new set of 8 points with a curved pattern and adjust two points algebraically until the mean, variance and correlation match a target.",
   "slVsHl": "SL computes and compares the statistics and plots residuals. HL treats the construction as a system of equations, and can add rank correlation or a transformation that restores a linear fit.",
   "whereMarksAreLost": "C: only reproducing the famous example. E: stopping at calculating values already published. B: no explanation of why the correlation coefficient misses curvature.",
   "dataNote": "Small data sets by hand, so the only difficulty is the construction step, which needs careful algebra.",
   "verdict": "Very familiar to moderators, so simply re-computing the quartet earns little. Constructing your own matching sets makes it personal and mathematical."
  },
  {
   "id": "do-pea-crosses-follow-mendel-s-ratios-testing-with-real-counts",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Do pea crosses follow Mendel's ratios? Testing with real counts",
   "researchQuestion": "How well do published offspring counts from two-trait pea or fruit fly crosses agree with the 9:3:3:1 ratio predicted by independent assortment, and what does a linked-gene model change?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Binomial and multinomial probability, conditional probability, tree diagrams and Punnett grids for dihybrid crosses, expected counts from a model, chi-squared test for goodness of fit with degrees of freedom and p-value (calculator), hypothesis testing. Beyond syllabus but reachable: a recombination fraction estimated by maximum likelihood, compared with the 9:3:3:1 model.",
   "howToStart": "Download the count tables from Mendel's 1866 dihybrid peas (round/wrinkled, yellow/green: 315, 101, 108, 32) and a fruit fly linkage dataset such as Morgan's white eye and miniature wing crosses. First compute the probability of each phenotype from a Punnett grid, then scale to the sample size and compare with the observed numbers.",
   "slVsHl": "SL: build the probability model for one and two traits, then run a chi-squared test on the peas and interpret the p-value. HL depth: use the fly data where the model fails, fit a linkage parameter by maximising the likelihood, and discuss why Mendel's fit seems suspiciously good, with a clear look at the assumptions behind each test.",
   "whereMarksAreLost": "E: stopping at Punnett squares, which is basic probability, and never testing a model against data. B: not stating hypotheses, degrees of freedom or the significance level. C: a generic 'genetics is interesting' voice with no personal choice of organism or dataset. D: failing to question the independence assumption or the small sample effects.",
   "dataNote": "Classic count tables are publicly available and small, so there is no cleaning, but the student must check the expected counts are large enough for the test.",
   "verdict": "It works when it tests a model against real counts, not as a probability exercise. The personal twist is picking a trait you can observe yourself, or comparing a case where the ratio holds with one where linkage breaks it."
  },
  {
   "id": "do-supermarket-sugar-bags-really-contain-1-kg",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Do supermarket sugar bags really contain 1 kg?",
   "researchQuestion": "Is the mean mass of 1 kg bags of sugar from different brands at least the labelled amount, and are the masses normally distributed?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Mean and standard deviation, normal model, probability of a bag below 1000 g, two-sample t-test to compare brands, confidence interval for the mean and one-sample t-test against 1000 g (AI HL), a normal quantile plot or chi-squared goodness of fit. Beyond syllabus: the 'average quantity' packaging rule (the mean must not be below the label and few bags may fall below a tolerance).",
   "howToStart": "Buy 30 or more bags of each of two brands and weigh each on a scale accurate to 1 g, checking its calibration with a known mass. Record the packaging mass separately if you weigh a bag opened.",
   "slVsHl": "SL models the masses as normal, estimates the proportion under 1000 g and compares two brands with a two-sample t-test. HL adds a confidence interval for each mean, a one-sample t-test against the label and a check of normality.",
   "whereMarksAreLost": "E: t-test used without checking the assumptions. A: unclear measuring method or calibration. D: no comment on scale error or whether the brand meets legal rules.",
   "dataNote": "Requires buying dozens of bags and a decent scale, but the data are quick to collect once done.",
   "verdict": "Good and practical if you compare brands and link the result to the legal rule; the personal touch could be a brand you buy regularly."
  },
  {
   "id": "do-sweet-colour-proportions-match-the-packet-claims",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Do sweet colour proportions match the packet claims?",
   "researchQuestion": "Do colour counts in several bags of sweets agree with the proportions stated by the manufacturer, and do people's favourite colours differ from what is in the bag?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Chi-squared goodness-of-fit test with degrees of freedom and expected-count check, chi-squared test for independence on a survey table, hypotheses and p-value, combining bags into one sample. Beyond syllabus: multinomial confidence intervals or a simulation of random bags.",
   "howToStart": "Buy at least 15 bags, count colours in each, and pool the totals. Separately, survey about 100 classmates for favourite colour. State the claimed proportions first.",
   "slVsHl": "SL applies the goodness-of-fit test to pooled counts and interprets the result. HL can add a test per bag, a simulation of the null model and a discussion of how bag-to-bag variation affects the test.",
   "whereMarksAreLost": "E: one test and nothing else. D: not noticing that comparing claims with preferences are two different questions. C: shallow motivation. A: unlabelled tables.",
   "dataNote": "Data is cheap to collect but the sample size must be large enough for expected counts of at least 5.",
   "verdict": "Accessible but quite common and easily too shallow. The twist is the simulation of random bags to check whether the chi-squared results agree with what chance alone would give."
  },
  {
   "id": "do-volcanic-eruptions-arrive-at-random-a-poisson-check",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "HL",
   "title": "Do volcanic eruptions arrive at random? A Poisson check",
   "researchQuestion": "Do eruptions of a chosen volcano, such as Etna or Kīlauea, occur as a Poisson process, judged by the counts per decade and the gaps between them?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Poisson distribution and its mean and variance, estimating λ from data, chi-squared goodness of fit with merged classes, comparison of variance with mean. Beyond syllabus: exponential waiting times between events.",
   "howToStart": "Download the eruption record of one volcano from the Smithsonian Global Volcanism Program. Count eruptions starting in each decade, estimate λ, and compare the observed counts with Poisson predictions.",
   "slVsHl": "The Poisson distribution is AI HL. An SL student could use the binomial distribution on years with or without an eruption. HL depth comes from testing the fit and asking whether the variance exceeds the mean.",
   "whereMarksAreLost": "E: fitting λ and stopping there without a goodness of fit test. B: unclear definition of an eruption event. D: ignoring that older records are incomplete, which biases counts.",
   "dataNote": "Records are publicly available but need cleaning, and historical under-reporting must be handled explicitly.",
   "verdict": "Good, provided you test the fit instead of assuming it. The twist is to compare a volcano that clusters with one that looks random."
  },
  {
   "id": "do-wet-days-come-in-streaks-testing-rainfall-models",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Do wet days come in streaks? Testing rainfall models",
   "researchQuestion": "Does a binomial or a Poisson model describe the number of wet days per month in one chosen city, and does a two-state Markov chain of dry and wet days explain the misfit better?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Relative frequency estimates of p, binomial distribution B(n, p) with mean np and variance np(1-p), Poisson distribution with mean equal to variance, comparing observed counts with model counts using a chi-squared test for goodness of fit (degrees of freedom, merging small cells, p-value, significance level), variance-to-mean ratio as a quick diagnostic. Beyond syllabus if reachable: a two-state transition matrix for dry and wet days, with its long-run proportion found as an eigenvector (HL matrices, or simple simultaneous equations for SL), and the resulting run-length distribution being geometric.",
   "howToStart": "Download daily precipitation records for one station over 25 to 30 years from a national meteorological archive, for example the Met Office historical station data, NOAA Climate Data Online or the ECA&D dataset. Choose a single month, such as October, to avoid seasonal mixing. Define a wet day with a stated threshold, say at least 1 mm. Count wet days per October for each year, compute the sample mean and variance, and estimate p as the total wet days divided by total days. Then tabulate the observed frequency of each count against the binomial prediction.",
   "slVsHl": "An SL student can do the binomial and Poisson fits, calculate the test statistic with the GDC or a spreadsheet, and interpret the result with a clear discussion of why independence fails. HL depth comes from building the Markov transition matrix from consecutive days, finding its steady state, simulating months from it, and showing the simulated counts fit the observed spread of counts far better than the binomial. Estimating the parameters from the same data and adjusting degrees of freedom shows top-band care.",
   "whereMarksAreLost": "A Presentation: lots of raw tables pasted in with no selection. B Mathematical communication: not defining a wet day, not stating hypotheses and degrees of freedom, unexplained GDC output. C Personal engagement: a generic city with no reason for the choice. D Reflection: stating the model fails without explaining why or testing the independence assumption. E Use of mathematics: a single test with expected counts below 5 left unmerged, or only a binomial fit with no second model.",
   "dataNote": "Daily rainfall archives are public and clean, though 25 or more years of data need a little tidying for missing days and a decision on the wet-day threshold.",
   "verdict": "Worth choosing because a model that fails for a physical reason makes a much better exploration than one that fits neatly. The personal twist is picking a place with real meaning for you, and comparing a wet-spell-prone climate with a steadier one to show the clustering is measurable.\n"
  },
  {
   "id": "does-caffeine-or-distraction-change-reaction-time-a-two-sample-test",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Does caffeine or distraction change reaction time? A two sample test",
   "researchQuestion": "Is the mean reaction time on an online ruler or screen test different between two conditions, once I allow for the spread in each group?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Sample mean and standard deviation, two-sample t-test, hypotheses and p-value, checking normality with a histogram. Beyond syllabus for SL: the paired t-test (AI HL) and a permutation test.",
   "howToStart": "Choose one condition, such as with and without music. Split about 30 volunteers at random between the two conditions, record the median of 10 trials for each person, plot the data, then run the test and check it against a GDC.",
   "slVsHl": "SL applies the two-sample t-test with clear hypotheses and interpretation. HL compares this with a paired design where everyone does both conditions, and with a permutation test, and discusses which design answers the question better.",
   "whereMarksAreLost": "E: running the test without checking assumptions or plotting. D: ignoring practice effects and small samples. A: cluttered raw data tables. C: no reason given for the chosen condition.",
   "dataNote": "Data must be collected with volunteers and consent, and practice effects need controlling.",
   "verdict": "A very common design, though it is workable if the condition is something you care about. Careful control of confounders is what separates a good one from a weak one."
  },
  {
   "id": "does-retrieval-practice-beat-rereading-a-paired-class-experiment",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "HL",
   "title": "Does retrieval practice beat rereading? A paired class experiment",
   "researchQuestion": "Do students score higher on a short quiz after studying by self-testing than after rereading, and is the difference larger than chance variation?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Paired design, differences of scores, paired t-test with hypothesis and p-value, confidence interval for mean difference, checking normality with a histogram or normal plot. Beyond syllabus and reachable: effect size using Cohen's d.",
   "howToStart": "Ask about 20 classmates for consent, give each two comparable word lists, one studied each way, and record scores. Get school approval first and anonymise names.",
   "slVsHl": "The paired t-test and confidence intervals are AI HL. HL work adds a confidence interval for the mean difference, an effect size and a discussion of order effects, with a check of the assumptions. An SL student could run a two-sample t-test, but that ignores the pairing, which should then be discussed honestly.",
   "whereMarksAreLost": "E: test run with no check of assumptions. A: messy tables. D: ignoring ethical and design limits. C: no personal link to studying.",
   "dataNote": "Collecting enough consented pairs takes effort and the sample stays small.",
   "verdict": "Worth choosing when the design is paired and controlled. Comparing two different classes taught differently is a weak design."
  },
  {
   "id": "how-many-reviews-before-an-average-rating-can-be-trusted",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "How many reviews before an average rating can be trusted",
   "researchQuestion": "How many reviews does a product need before its mean rating settles within ±0.2 of the long-run value, and does this differ between two review sites?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Sample mean and standard deviation, standard error, confidence intervals for a mean (t-interval), repeated random subsampling to show spread shrinking as 1/√n. HL adds a comparison of variances or a bootstrap-style argument and discussion of skewed, J-shaped rating distributions that break normality assumptions.",
   "howToStart": "Pick 5 to 10 items with several hundred public reviews each, record the star ratings in a spreadsheet, then draw random subsamples of sizes 5, 10, 20, 50 and compute the spread of the means.",
   "slVsHl": "SL measures how the spread of subsample means shrinks with n and compares it with σ/√n. HL adds t-intervals for the mean (AI HL), tests the normal approximation on skewed data and compares two platforms.",
   "whereMarksAreLost": "E: stopping at descriptive means and one interval. B: not explaining why ratings are not normal. C: safe topic with little voice. D: ignoring selection bias in who writes reviews.",
   "dataNote": "Collecting a few hundred ratings by hand is slow, and platform terms may limit scraping, so use a dataset or manual sampling.",
   "verdict": "Sound and doable, with a clear answer. The twist that makes it personal is choosing products or games you actually buy and checking whether the reviews would have changed your choice."
  },
  {
   "id": "how-many-seats-to-oversell-on-a-180-seat-flight",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "How many seats to oversell on a 180-seat flight",
   "researchQuestion": "What number of tickets maximises expected profit for a 180-seat flight when passengers fail to show up independently with a stated probability, and how sensitive is this number to the costs?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Binomial distribution and its cumulative probabilities, expected value of a piecewise profit function, comparing consecutive expected values to find an integer optimum, sensitivity analysis. Beyond syllabus: a normal approximation to the binomial with a continuity correction, checked against the exact values.",
   "howToStart": "Find a published no-show rate for airlines and set plausible ticket price and compensation cost. Compute expected profit for every ticket count from 180 to 200 in a spreadsheet using the binomial distribution.",
   "slVsHl": "SL: one no-show rate, one cost pair, exact binomial and a table. HL depth: normal approximation compared with exact values, sensitivity of the optimum to the no-show probability and cost ratio, and a test of the independence assumption using groups travelling together.",
   "whereMarksAreLost": "E: costs chosen with no source. B: profit function not defined precisely. D: independence assumption never questioned. C: no link to the student's own interests.",
   "dataNote": "Real no-show rates and compensation figures are patchy, so state clearly which values are assumed.",
   "verdict": "A clean optimisation with a real answer, and the sensitivity study is where the marks are. Weak if the costs are just invented."
  },
  {
   "id": "is-a-basketball-player-s-hot-streak-real",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Is a basketball player's hot streak real?",
   "researchQuestion": "Do a player's shooting results over a season show more streaks than a fixed shooting probability would produce by chance?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Binomial distribution, probability of runs, simulation of the model, chi-squared test for independence between consecutive shots, expected value and variance. HL extension: Markov chain with make and miss states and its transition matrix. Beyond syllabus: the runs test.",
   "howToStart": "Download shot-by-shot logs for one player from a public basketball statistics site for a full season. Compute overall shooting percentage, then the proportion of shots made after a make and after a miss.",
   "slVsHl": "SL compares the two conditional proportions with a chi-squared test and a simulation. HL builds a two-state Markov chain and compares its predictions with the data.",
   "whereMarksAreLost": "E: only means and percentages reported. B: hypothesis test steps not explained. D: no discussion of shot difficulty as a hidden variable. C: player chosen with no reason given.",
   "dataNote": "Shot logs are public but need cleaning into ordered sequences.",
   "verdict": "Turned into a streak test it has a clear answer and real statistics. Choose a team or player you follow."
  },
  {
   "id": "is-any-lottery-ever-worth-playing",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Is any lottery ever worth playing?",
   "researchQuestion": "For which jackpot sizes, if any, does the expected return on a ticket in a national lottery exceed its price once rollovers and shared prizes are allowed for?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Combinations to count outcomes for each prize tier; discrete random variable and expected value; binomial distribution for the number of jackpot winners; conditional expectation given a shared jackpot. Beyond syllabus for HL: Poisson approximation to the number of winners.",
   "howToStart": "Take the official prize structure and ticket price of one lottery (for example EuroMillions or the UK National Lottery). Compute the probability of each tier with nCr, then the expected return for a fixed jackpot.",
   "slVsHl": "SL computes the expected value tier by tier and finds the break-even jackpot. HL adds the number of other winners as a binomial or Poisson variable, so the jackpot is split, and compares the corrected break-even figure with the naive one.",
   "whereMarksAreLost": "E: stopping at one expected value calculation with no modelling choice. B: unexplained notation for tiers. C: generic wording with no personal angle. D: no comment on what the model ignores, such as tax and utility.",
   "dataNote": "Prize tables and past jackpot sizes are published, so the data is easy; the sharing correction is the real work.",
   "verdict": "Common topic, so it only earns marks with the shared jackpot twist and a break-even analysis. Add historical draw data to check whether the number of winners fits your model."
  },
  {
   "id": "is-reaction-time-really-normal-testing-a-model-on-ruler-drop-data",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Is reaction time really normal? Testing a model on ruler-drop data",
   "researchQuestion": "Do reaction times measured by an online test or ruler-drop follow a normal distribution, or is a skewed model such as the log-normal a better fit?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Mean, standard deviation, histograms, normal probabilities and inverse normal, chi-squared goodness of fit test (AI), transformation of data by taking logarithms, comparison of quantiles. Beyond syllabus: the log-normal distribution.",
   "howToStart": "Collect at least 100 reaction times from yourself and classmates using a ruler-drop or a simple browser test. Plot a histogram and compute the mean and standard deviation, then compare observed frequencies with those predicted by a fitted normal curve.",
   "slVsHl": "SL fits a normal model, checks the proportion within one and two standard deviations, and comments on skew. HL adds a formal goodness of fit test and a comparison with the log-normal fit.",
   "whereMarksAreLost": "E: assuming normality rather than testing it. A: too many raw tables. D: no discussion of measurement error, learning effects or outliers from lapses in attention.",
   "dataNote": "Needs enough clean repeated measurements, and the first few trials per person should be discarded because of practice effects.",
   "verdict": "Fine, as long as the question is whether normality fails rather than an assumption that it holds. The skew is the interesting result."
  },
  {
   "id": "predicting-league-results-with-a-poisson-goals-model",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "HL",
   "title": "Predicting league results with a Poisson goals model",
   "researchQuestion": "How well does a Poisson model of goals, using team attack and defence strengths, predict the results of a full league season?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Poisson distribution, expected value, independent events, chi-squared goodness of fit test, probability of a win, draw or loss by summing products. Fitting means from historical averages. HL extension: maximum likelihood ideas or a Markov chain for a league table.",
   "howToStart": "Download one past Premier League season from football-data.co.uk, compute average goals for and against for each team, then compute expected goals for one match and the win, draw and loss probabilities.",
   "slVsHl": "The Poisson distribution is AI HL. HL students build the Poisson model, test it with chi-squared and compare predicted with real points; extra depth comes from the distribution of goal difference or from testing the model on a season it was not fitted to.",
   "whereMarksAreLost": "E: doing repeated arithmetic in a spreadsheet without justifying the Poisson assumption. D: failing to see that goals are not fully independent. C: generic, no personal team. A: pages of tables.",
   "dataNote": "The data is public, tidy and ready to use.",
   "verdict": "Popular, but a well-tested model with an honest goodness of fit check is respectable. Pick the student's own team and test the model on a season it was not fitted to."
  },
  {
   "id": "screening-tests-and-the-base-rate-problem",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Screening tests and the base rate problem",
   "researchQuestion": "How does the prevalence of a disease change the probability that a positive screening result is a true positive, and at what prevalence does a test become useful?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Conditional probability, Bayes' theorem, tree diagrams, sensitivity and specificity, expected values for a screened population. Beyond syllabus: a comparison with a two-test sequence using independence assumptions.",
   "howToStart": "Take published sensitivity, specificity and prevalence figures for a screening test from a public health agency such as the NHS or WHO, and compute the positive predictive value for a population of 10,000 people.",
   "slVsHl": "SL students can use tree diagrams and conditional probability with published figures. HL students can vary prevalence as a parameter, derive the PPV function, and analyse a two-stage screening design with its assumptions.",
   "whereMarksAreLost": "E: using only one calculation with no variation of parameters. D: no discussion of the independence assumption or data quality. A: long unstructured tables.",
   "dataNote": "Figures are public but need careful sourcing and a check that sensitivity and specificity come from the same study.",
   "verdict": "A strong choice when the parameter study is the centre. Twist: use a test from a health issue you know, with sources cited."
  },
  {
   "id": "size-against-sale-price-linear-power-or-log-model",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Size against sale price: linear, power or log model?",
   "researchQuestion": "Which of a linear, power and logarithmic model best predicts the sale price of flats in my town from floor area, and how reliable is it outside the sampled range?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Scatter plots, Pearson and Spearman correlation, least squares regression, log-log transformation for a power model and semi-log for a logarithmic one, residual plots, coefficient of determination, prediction error on held-out sales. Beyond syllabus: inference on the gradient and a second explanatory variable.",
   "howToStart": "Download about 60 recent sales of one property type from a public sold-prices site or a listings portal in one postcode area. Plot price against area and check the residuals before choosing a model.",
   "slVsHl": "SL compares linear and power fits with residuals and interprets the coefficients. HL adds the logarithmic model through linearisation, tests predictions on held-out sales and discusses which model generalises best.",
   "whereMarksAreLost": "E: only a linear regression from a spreadsheet. C: little personal engagement when the topic is generic. D: no critique of outliers or sample choice. B: unexplained software output.",
   "dataNote": "Data are available but need cleaning, such as removing duplicates and mismatched property types.",
   "verdict": "Common and easy to do badly. It becomes worthwhile through comparing several models, testing predictions on held-out sales and being honest about the many missing factors such as location."
  },
  {
   "id": "snakes-and-ladders-as-a-markov-chain-expected-game-length",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Snakes and Ladders as a Markov chain: expected game length",
   "researchQuestion": "For a real 100-square board, what is the expected number of turns to finish, and how does moving one ladder or snake change it?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Probability of dice outcomes, expected value, transition matrix, matrix powers and solving linear systems for the expected number of turns (AI HL matrices). Beyond syllabus: absorbing Markov chains with the fundamental matrix, and the geometric distribution as a simple model.",
   "howToStart": "Code a 10-square toy board first and solve by hand. Then build the 100-square transition matrix and check it with a Monte Carlo simulation of 10000 games.",
   "slVsHl": "SL can use the toy board, tree diagrams and simulation. HL can handle the full matrix, solve for expected turns exactly and compare with simulation.",
   "whereMarksAreLost": "E: only simulation with no analytic result. B: unexplained code. D: no discussion of the exact-roll rule at the end. The 'strategy' claim is empty because players make no choices.",
   "dataNote": "Everything is computed from the board rules, though the finishing rule must be stated clearly.",
   "verdict": "A good, achievable choice, but popular. Stand out by comparing several real commercial boards or by designing a board with a target game length."
  },
  {
   "id": "testing-zipf-s-law-on-english-italian-and-latin-texts",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "AI",
   "level": "both",
   "title": "Testing Zipf's law on English, Italian and Latin texts",
   "researchQuestion": "How well does frequency proportional to 1/rank^s fit word counts in three languages, and does the exponent s differ between them?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Log-log transformation, linear regression and least squares, correlation coefficient r, coefficient of determination, residual analysis, chi-squared goodness of fit. Beyond syllabus: comparison with a harmonic distribution.",
   "howToStart": "Download three full novels from Project Gutenberg, count words with a short Python script or spreadsheet, rank them, and plot log frequency against log rank.",
   "slVsHl": "SL fits a straight line to the log-log data and interprets the gradient. HL examines the deviation at low and high ranks, tests the fit, and compares texts of different lengths.",
   "whereMarksAreLost": "E: regression done by software with no justification of the transformation. A: too many unlabelled plots. D: no discussion of why the tails deviate.",
   "dataNote": "Public domain texts are easy to count, but tokenisation choices such as apostrophes and accents must be stated.",
   "verdict": "A good choice with real data and a clear question. The personal twist is using a language you speak or your own writing, and being honest about where the law fails."
  },
  {
   "id": "testing-a-claimed-psychic-with-a-binomial-model",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Testing a claimed psychic with a binomial model",
   "researchQuestion": "How many correct guesses out of n Zener-style card trials are needed before chance can be rejected at the 5% and 1% levels, and how does the power of that test change with the true success rate?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Binomial distribution with P(X ≥ k) tail probabilities, null and alternative hypotheses, critical region, significance level, Type I and Type II errors. Beyond syllabus: power curve as a function of the true p, or a Bayesian comparison using a prior.",
   "howToStart": "Design a card-guessing task with a fixed chance probability (for example five symbols, p = 0.2). Run 100 trials with 5 to 10 volunteers, then compute P(X ≥ observed) under the null hypothesis by hand and check with a calculator.",
   "slVsHl": "SL does the binomial tail probabilities and a one-tailed test with a clear critical region. HL or top-band work adds the power calculation for several alternative values of p and a discussion of multiple testing when many volunteers are tried.",
   "whereMarksAreLost": "E: stopping at 'p-value below 0.05 so psychic' without Type II error. D: no comment on the multiple comparison problem or on cheating and poor randomisation. C: generic if the student did not design their own trials.",
   "dataNote": "Data are easy to collect, though the number of trials must be planned in advance to avoid stopping when results look good.",
   "verdict": "Sound and accessible, but common as a hypothesis test, so the depth has to come from power and multiple testing. A twist: test yourself against a shuffled deck you control, then discuss why one lucky volunteer among many proves nothing."
  },
  {
   "id": "typical-distance-after-n-coin-flip-steps",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Typical distance after n coin-flip steps",
   "researchQuestion": "After n independent steps of ±1, how far from the start is a walker typically, and does the typical distance grow like √n?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Expectation and variance of sums of independent variables; binomial distribution for the position; E(X²)=n; probability of returning to the origin; simulation for verification. Beyond syllabus: Stirling's approximation for the central binomial coefficient (HL) and the normal approximation.",
   "howToStart": "Write a short spreadsheet or Python simulation of 10 000 walks of n = 10, 100, 1000 steps, and record the mean absolute distance. Then compute E(X²) exactly and compare.",
   "slVsHl": "SL derives E(X)=0 and Var(X)=n and checks √n growth by simulation. HL uses the binomial distribution to compute E|X| exactly and applies Stirling to show it behaves like √(2n/π), or extends to biased steps.",
   "whereMarksAreLost": "E: relying on simulation only. B: confusing E|X| with √E(X²). A: unstructured code output. D: no comment on simulation error.",
   "dataNote": "Simulated data are easy to generate; the exact E|X| calculation needs care.",
   "verdict": "Strong if the exact calculation is done and compared with simulation. Twist: model a drunk walk on a real corridor with a wall, or a stock price."
  },
  {
   "id": "when-to-stop-rolling-optimal-strategy-in-a-push-your-luck-dice-game",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "When to stop rolling: optimal strategy in a push-your-luck dice game",
   "researchQuestion": "In a game where you keep adding dice totals until you roll a 1 and lose the turn, what stopping rule maximises the expected score per turn?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Probability of independent events, expected value, geometric distribution, conditional probability, comparing the expected gain and loss of one more roll to find a threshold. Beyond syllabus: Markov chains for a whole game against an opponent.",
   "howToStart": "Compute by hand the expected change in score from one more roll at a running total T (gain 4 x 5/6 on average, loss T x 1/6). Then simulate 100000 turns for each threshold to check it.",
   "slVsHl": "SL finds the threshold with expected values and a simulation. HL uses a recursion or a Markov chain, and can vary the die or the target score to model the winning probability rather than the score.",
   "whereMarksAreLost": "E: a single threshold found by trial and error with no algebra. B: unclear notation for states. D: no criticism of maximising the mean when a player only needs to win.",
   "dataNote": "Simulation is easy; exact results need care but stay tractable.",
   "verdict": "A good accessible topic. Make it personal by using a game you actually play and by asking whether maximising expectation is the right aim when trailing."
  },
  {
   "id": "where-does-e-appear-when-random-numbers-are-added-up",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Where does e appear when random numbers are added up?",
   "researchQuestion": "How many random numbers from 0 to 1 must be added on average for the total to exceed 1, and why is the answer e?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Uniform continuous distribution, expected value, volume of the simplex 1/n! as a probability in n dimensions, E(N)=sum of P(N>n), Maclaurin series of e^x, Monte Carlo simulation. Beyond syllabus: connection to the Irwin-Hall distribution.",
   "howToStart": "Simulate 100000 trials in a spreadsheet or Python of adding uniform random numbers until the sum passes 1 and record the mean count. Then show P(sum of n numbers <1)=1/n! for n=2 and 3 by integration.",
   "slVsHl": "SL uses simulation plus n=2,3 areas and volumes and the known series for e. HL proves the general 1/n! by induction and integration and discusses convergence of the simulation.",
   "whereMarksAreLost": "E: presenting simulation as proof. B: unclear link between the hypercube and the sum. D: no error estimate for the simulated mean.",
   "dataNote": "Simulation is easy. The general proof by induction is the hard part but manageable.",
   "verdict": "Elegant and less overdone than most. Keep it anchored in the sum problem and make the proof yours."
  },
  {
   "id": "why-doubling-your-bet-never-beats-the-house",
   "topic": "4",
   "topicName": "Statistics and probability",
   "course": "both",
   "level": "both",
   "title": "Why doubling your bet never beats the house",
   "researchQuestion": "Why does the doubling strategy give a near certain small win yet a negative expected gain, and how does a finite bankroll or table limit change the answer?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Geometric series and their sums, expectation of a discrete random variable, geometric distribution for the length of a losing run, probability of ruin, simulation of many sessions. Beyond syllabus: the martingale property and optional stopping.",
   "howToStart": "Write down the profit after n consecutive losses, work out the probability of a run of length n on a fair and on a roulette style bet with p = 18/37, then simulate 10000 sessions with a bankroll cap.",
   "slVsHl": "SL computes the expected gain for a fixed number of rounds using geometric series and simulates. HL derives the expected gain analytically, explains why an unlimited bankroll seems to beat even a fair game by looking at the size of the stakes needed, and compares bankrolls with a table cap.",
   "whereMarksAreLost": "E: a simulation with no exact calculation to check it. B: unclear notation for a round and a run. D: not linking the finite result to the paradox. C: no distinct twist, such as applying it to a real trading style.",
   "dataNote": "Pure theory plus simple simulation; the maths is reachable and easy to control.",
   "verdict": "A sound and pleasing choice if you show the exact calculation and the simulation agreeing. Keep away from claims about real trading, where data is hard to justify."
  },
  {
   "id": "reproduction-number-and-peak-of-an-sir-epidemic",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "Reproduction number and peak of an SIR epidemic",
   "researchQuestion": "How does the peak proportion infected in an SIR model depend on the reproduction number, and how well does a fitted model reproduce a real outbreak?",
   "dataDifficulty": 2,
   "dataSource": "database",
   "sitesListingIt": 6,
   "namedInReports": true,
   "mathsRequired": "Coupled differential equations, separation of variables after dividing dI/dt by dS/dt, integration, logarithms, Euler's method with step-size refinement, finding the maximum where dI/dt = 0. Beyond syllabus: least-squares fitting of the two rates.",
   "howToStart": "Download daily case counts of a school flu outbreak or a national epidemic from a health agency. Write the SIR equations, find the exact peak formula analytically, and compare with an Euler solution.",
   "slVsHl": "Differential equations are HL in both courses. HL depth: eliminate time analytically, derive I_max = 1 − (1 + ln R₀)/R₀ for a fully susceptible start, and compare it with an Euler solution and with real data. An SL student would be outside the course.",
   "whereMarksAreLost": "E: only running a solver with no derivation. C: generic COVID topic with no personal angle. B: unexplained assumptions such as constant population. D: no discussion of the model's limits.",
   "dataNote": "Data are easy to find but noisy, and reported cases rarely match the infected compartment.",
   "verdict": "Extremely overdone: listed on six public lists and named in the subject reports among topics seen too often. Only choose this narrow version, with the analytic peak and a critical fitting exercise on an outbreak you have a reason to study."
  },
  {
   "id": "cooling-of-a-hot-drink-does-one-exponential-fit",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Cooling of a hot drink: does one exponential fit?",
   "researchQuestion": "How well does a decaying exponential with a fixed room-temperature asymptote describe a cooling mug of tea, and does the fitted rate constant stay the same as the drink cools?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 4,
   "namedInReports": true,
   "mathsRequired": "Exponential model T = Ta + (T0 − Ta)e^(−kt); linearising by taking ln(T − Ta) against t; least-squares regression; residual plots; percentage error. HL: separable differential equation dT/dt = −k(T − Ta) solved analytically. Beyond syllabus: compare with a two-term exponential to capture early rapid loss.",
   "howToStart": "Pour boiling water into one mug, log temperature every 30 seconds for 40 minutes with a digital thermometer, and record room temperature at the start and end. Compute ln(T − Ta) for each reading and check whether the points lie on a line.",
   "slVsHl": "SL fits the exponential and linearised forms and judges them by residuals and R². HL derives the model from the differential equation and tests whether k drifts with temperature, which is where real insight appears.",
   "whereMarksAreLost": "C: a copy of a common topic with no personal twist. E: only fitting a formula without testing its assumptions. D: reporting a good R² without asking why residuals show a pattern. A: too many raw data tables.",
   "dataNote": "Data is easy to collect, but the room temperature must be logged and the sensor error stated.",
   "verdict": "Newton's law of cooling is named in the subject reports among topics seen too often, so it only scores with a personal twist, such as comparing the mug you actually use with a flask, and an analysis of the pattern in the residuals rather than praise for the fit."
  },
  {
   "id": "why-drinks-cans-are-not-the-minimum-surface-area-cylinder",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Why drinks cans are not the minimum surface area cylinder",
   "researchQuestion": "For a can of fixed volume 330 ml, what dimensions minimise metal, and how must the cost model change to explain the shape of real cans?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 4,
   "namedInReports": true,
   "mathsRequired": "Volume and surface area of a cylinder, elimination of a variable by substitution, differentiation of S = 2πr² + 2V/r, second derivative test, then a weighted cost function with a thicker material in the lid and base and rim allowance. Beyond syllabus: Lagrange multipliers to confirm the result, or comparison with other prisms.",
   "howToStart": "Measure the height, diameter and wall thickness of five different real cans, then work out the ideal ratio h : r for pure surface area and compute how far each can deviates.",
   "slVsHl": "SL optimises the plain and weighted cost functions with the derivative and compares with measured cans. HL adds sensitivity analysis on the cost weights, a numerical or implicit approach, and a comparison with a spherical or hexagonal packing constraint.",
   "whereMarksAreLost": "E: the simple h = 2r result is a textbook exercise, so marks depend on the extension. C: no personal measurements. D: no sensible reasons why real cans deviate, such as stacking and grip.",
   "dataNote": "Real cans are easy to measure, and the extension is a modelling choice.",
   "verdict": "Minimising the surface of a can is named in the subject reports among topics seen too often. It is only worth choosing with measured cans and a clearly justified cost model, which lifts it above a textbook problem."
  },
  {
   "id": "football-flight-with-quadratic-drag-solved-numerically",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "Football flight with quadratic drag solved numerically",
   "researchQuestion": "How far does a kicked football fall short of the vacuum range when quadratic drag is included, and what launch angle maximises range?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 3,
   "namedInReports": true,
   "mathsRequired": "Set up the coupled equations for horizontal and vertical velocity with a drag term proportional to speed squared. Solve using Euler's method, then improve with a second-order scheme such as Heun's. Compare with the linear-drag case, which has an exact solution by separation of variables. Estimate the optimal angle numerically. Check against measured range from video or published data. Beyond syllabus: non-dimensionalise the system.",
   "howToStart": "Film a ball being kicked or thrown against a gridded background and extract positions with tracking software such as Tracker. Then code the drag model in a spreadsheet and fit the drag coefficient.",
   "slVsHl": "Differential equations and Euler's method are HL in both courses. HL depth comes from the coupled equations, the exact linear-drag solution as a check, step-size error analysis and the optimal-angle search.",
   "whereMarksAreLost": "E: using a spreadsheet without justifying the step size. C: choosing a textbook example with no personal data. B: not explaining why drag is modelled as proportional to v squared. D: no discussion of spin or wind.",
   "dataNote": "Video tracking is achievable, but fitting the drag coefficient reliably needs several throws and careful scaling.",
   "verdict": "Projectiles in sport are named in the subject reports among topics seen too often, and three public lists carry this one. It only works with your own tracked data and a clear comparison of two drag laws."
  },
  {
   "id": "lotka-volterra-cycles-fitted-to-lynx-and-hare-pelts",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "Lotka-Volterra cycles fitted to lynx and hare pelts",
   "researchQuestion": "How closely does a Lotka-Volterra system, with parameters fitted to the Hudson Bay lynx and hare records, reproduce the observed cycle length, and where does it fail?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 3,
   "namedInReports": false,
   "mathsRequired": "Coupled first-order differential equations, equilibrium points, phase portraits, numerical solution by Euler's method, the conserved quantity that makes the orbits closed. AI HL students can use eigenvalues for the linearised system. Beyond syllabus: linearisation with the Jacobian matrix, and adding a logistic limit on prey so that the closed orbits become a spiral.",
   "howToStart": "Download the Hudson Bay Company lynx and hare pelt counts (about 90 years, publicly available in R datasets and many university pages). Plot both series, measure the average period, then estimate the four parameters from the equilibrium values and the period before coding Euler in a spreadsheet.",
   "slVsHl": "An SL student can do a single-species version and a step-by-step simulation of the pair, comparing the cycle with the data graphically, though this sits outside the SL course. HL depth comes from the equilibria, the conserved quantity, the stability of the equilibrium, and a comparison of the pure model with the logistic-prey version.",
   "whereMarksAreLost": "E: stopping at 'the graph looks cyclic' with no analysis of equilibrium or stability. B: pasting solver output without explaining the steps. C: generic ecology with no personal link. D: no comment on the pelt counts being a proxy for true populations.",
   "dataNote": "The pelt data is public and clean, but choosing a fitting method for four parameters needs care.",
   "verdict": "Popular and easily done badly, so choose it only if you commit to fitting real data and criticising the model. The twist is testing why the real cycle is not the neat closed orbit the equations promise."
  },
  {
   "id": "catenoid-versus-two-discs-soap-film-area",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "Catenoid versus two discs: soap film area",
   "researchQuestion": "For two coaxial rings at a given distance apart, when does a catenoid film have less area than two flat discs, and at what separation does the catenoid cease to exist?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Surface area of revolution by integration (beyond the syllabus), the catenary y = a cosh(x/a) written with exponentials, solving a transcendental equation numerically, comparing areas. Beyond syllabus: the Euler-Lagrange equation to derive the catenary.",
   "howToStart": "Dip two wire rings of equal radius into soap solution, pull them apart and record the ring separation when the film snaps. Then compute the theoretical catenoid area for each separation.",
   "slVsHl": "Best at HL. Take the catenary as given, integrate to get the area and compare with the discs numerically; find the separation where the discs win (about 1.06 times the radius) and where the catenoid stops existing (about 1.33 times the radius), and compare with your measurements.",
   "whereMarksAreLost": "E: the variational derivation is only quoted. C: no real experiment. D: failing to explain why measured and predicted breaking distances differ. A: too many pages on the physics.",
   "dataNote": "Measurements are simple but noisy; the critical ratio needs numerical solving of cosh equations.",
   "verdict": "Strong if the experiment is real and the catenary is derived. The broad 'soap bubble minimal surface' framing is weak, so use this narrow question."
  },
  {
   "id": "fitting-logistic-growth-to-census-data-and-predicting-a-limit",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Fitting logistic growth to census data and predicting a limit",
   "researchQuestion": "How reliable is the carrying capacity predicted by a logistic model fitted to census data up to a chosen year, compared with later observed values?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Logistic differential equation dP/dt = rP(1 - P/K), solution by separation of variables and partial fractions, the point of inflexion at P = K/2, nonlinear regression or linearisation, comparison of predictions from different cut-off years. Beyond syllabus but reachable: adding a delay or an Allee effect.",
   "howToStart": "Download census populations for one country, such as the USA from 1790 or a smaller region, and fit a logistic curve using only data before 1900. Compare the predicted curve and K with what happened afterwards.",
   "slVsHl": "SL students can use the logistic function form, fit it with technology and compare to an exponential model. HL students can derive the solution, prove the inflexion at K/2, and study how the estimate of K moves as more years are included.",
   "whereMarksAreLost": "E: using regression software as a black box with no derivation. D: not admitting that the fit is poor for demographic data. B: unexplained parameter values. A: too many similar graphs.",
   "dataNote": "Census data is easy to obtain; the tricky part is choosing a fitting method and explaining it.",
   "verdict": "A sound choice if you test predictions using old data against what really happened. The twist is showing how unstable K is when estimated early on."
  },
  {
   "id": "half-life-from-a-differential-equation-and-real-decay-counts",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Half-life from a differential equation and real decay counts",
   "researchQuestion": "How well does the exponential decay law, derived from the assumption that decay rate is proportional to the number of nuclei, fit counts from a real decaying source, and how large is the error in the estimated half-life?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Separable differential equation dN/dt = -kN and its solution, half-life as ln2/k, natural logarithms to linearise data, least-squares regression on ln(count), residual analysis. Beyond syllabus but reachable: Poisson noise in counts and the resulting uncertainty in k.",
   "howToStart": "Use a public dataset of Geiger counter readings for a short-lived isotope such as barium-137m, or the well-known dice or coin decay simulation with 200 dice. Take ln of the counts, fit a line and get k and its uncertainty.",
   "slVsHl": "SL students can use the exponential model, logarithmic linearisation and regression, and compare with the dice simulation. HL students can solve the differential equation by separation, add background radiation as a constant term, and treat the random scatter with a Poisson argument.",
   "whereMarksAreLost": "E: quoting N = N0 e^(-kt) without deriving it or testing it. A: a very short exploration that repeats standard textbook working. C: no reason given for the choice. D: no evaluation of the fit or of background counts.",
   "dataNote": "Real counts need a source or a public dataset; a dice simulation is an easy backup but is less authentic.",
   "verdict": "Very common and close to a textbook exercise unless real data and uncertainty are the focus. The twist is comparing a physical decay with a dice model and asking which is closer to the law."
  },
  {
   "id": "measuring-inequality-gini-by-lorenz-curve-integration",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Measuring inequality: Gini by Lorenz curve integration",
   "researchQuestion": "How much do trapezium-rule and fitted-function integration of the Lorenz curve differ in the Gini coefficient for national income data, and which is more reliable?",
   "dataDifficulty": 1,
   "dataSource": "database",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Definition Gini = 1 − 2∫L(x)dx; trapezium rule on grouped quintile or decile data; fitting a function such as L(x) = x^p and integrating it exactly; error comparison between methods; comparing countries or years. HL: an integration of the fitted function and a discussion of the relationship to the mean absolute difference.",
   "howToStart": "Download income share by decile for two countries from the World Bank. Build the Lorenz points, compute the Gini with the trapezium rule, then fit L(x) = x^p and integrate it, and compare the two values with the published one.",
   "slVsHl": "SL uses the trapezium rule and exact integration of a fitted power function, for which the Gini is (p − 1)/(p + 1). HL adds a richer fitted family and compares the integral formula with the mean absolute difference definition.",
   "whereMarksAreLost": "E: plugging numbers into a formula with no comparison of methods. C: a generic choice of countries. D: no note that grouped data underestimates inequality inside each group. B: the Lorenz curve not defined properly.",
   "dataNote": "Decile data is publicly available and clean, so the challenge is in the analysis.",
   "verdict": "Good if you focus on comparing numerical methods and their error, not describing inequality. Choose countries or years you have a reason to care about."
  },
  {
   "id": "straight-ramp-or-cycloid-timing-the-descent",
   "topic": "5",
   "topicName": "Calculus",
   "course": "AA",
   "level": "HL",
   "title": "Straight ramp or cycloid: timing the descent",
   "researchQuestion": "By what percentage is the descent time along a cycloid shorter than along a straight line joining the same two points, and can a physical track confirm it?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Energy conservation v = √(2gy), time as ∫ ds/v with ds from parametric derivatives, cycloid x = a(t − sin t), y = a(1 − cos t), a substitution to evaluate exactly, numerical integration for the straight line and other paths (parabola, circle arc), percentage comparison. Beyond syllabus: an outline of why the Euler-Lagrange equation leads to the cycloid, quoted rather than proved.",
   "howToStart": "Choose two points, write the descent time integral for a straight line and evaluate it, then do the cycloid; afterwards build two tracks and time a ball with slow-motion video.",
   "slVsHl": "The time integral along a curve is beyond both syllabi, so this suits a confident HL student; a strong SL student could time straight and polygonal paths with numerical sums only. Top marks come from comparing several families of paths and reconciling rolling friction with the model.",
   "whereMarksAreLost": "E: claiming the calculus of variations without understanding it, so quote it honestly. B: parametric notation. D: ignoring that a rolling ball has rotational energy, which changes constants.",
   "dataNote": "Mathematics is clean; the physical test needs careful timing and a correction for rolling.",
   "verdict": "Attractive and rich, and a track experiment makes it personal. Do not attempt to prove optimality; keep to comparing paths."
  },
  {
   "id": "where-to-sit-in-a-cinema-for-the-widest-view",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Where to sit in a cinema for the widest view",
   "researchQuestion": "At what distance from the screen is the vertical angle subtended by the screen largest, and how does this compare with where people actually choose or are advised to sit?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 2,
   "namedInReports": false,
   "mathsRequired": "Angle as a difference of two arctangents, θ = arctan((h+b)/x) − arctan(b/x), differentiation of arctan, solving dθ/dx = 0 to give x = √(b(b+h)), second-derivative or sign test, alternative proof by the tangent subtraction formula, sensitivity to h and b. Beyond syllabus: a sloping floor with rake, so that the eye height changes with distance.",
   "howToStart": "Measure or find plans for a real auditorium: screen height, height of its lower edge above eye level. Derive the optimum symbolically first, then check it numerically with a graph.",
   "slVsHl": "SL can do the tangent-difference and a graphical or numerical maximum with a derivative. HL handles arctan derivatives directly, does the general result in b and h, and adds the sloping floor.",
   "whereMarksAreLost": "E: this is very close to a textbook optimisation, so it needs the extension (rake, horizontal angle, a survey). C: generic wording with no personal source. B: not defining eye height clearly.",
   "dataNote": "Only a few measurements are needed; a survey of seat preferences would add data but is optional.",
   "verdict": "Weak alone because it is a known textbook problem; make it worthwhile by modelling a real cinema with a raked floor and comparing with a small survey of favourite rows."
  },
  {
   "id": "best-cone-from-a-fixed-sheet-of-paper",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Best cone from a fixed sheet of paper",
   "researchQuestion": "Which cone has the greatest volume when it is made from a sector of a circular sheet of fixed radius, and does the answer change when the area of paper is fixed instead?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Curved surface area πrl, l = √(r² + h²), eliminating a variable, differentiation of a volume function involving a square root, second derivative test, sector angle and arc length. With a fixed slant height (a sheet of fixed radius) the best sector angle is 2π√(2/3), about 294°; with a fixed area of paper it is 2π/√3, about 208°.",
   "howToStart": "Cut sectors of several angles from a disc of card of radius 15 cm, form the cones, and measure their volumes with rice. Then write V as a function of r with l fixed, find the maximum, and compare.",
   "slVsHl": "SL derives both optima and checks them with paper cones. HL proves the optimum angles do not depend on the size of the sheet, and analyses the experimental error from overlap and seams.",
   "whereMarksAreLost": "E: stopping at one calculus problem. D: no explanation of why real cups differ. A: messy geometry diagrams.",
   "dataNote": "Easy to model and to test with simple materials.",
   "verdict": "A decent optimisation idea when the experiment is real. Fairly standard otherwise, so the physical comparison is what makes it yours."
  },
  {
   "id": "best-throwing-angle-from-a-raised-release-point",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Best throwing angle from a raised release point",
   "researchQuestion": "How does the launch angle that gives the greatest range change as the release height above the landing level increases, and does it tend to 45° only when that height is zero?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Parametric equations x = u cos θ · t, y = h + u sin θ · t − ½gt², solving a quadratic for flight time, range R(θ), implicit differentiation, trigonometric identities. HL: deriving tan θ = u / √(u² + 2gh), limiting behaviour as h → 0 and h → ∞.",
   "howToStart": "Derive R(θ) for h = 0 and confirm 45°. Then include h, use a graph to find the optimal angle numerically for several h, and compare with a video analysis of a ball thrown from a balcony or a shot put.",
   "slVsHl": "SL computes the optimum numerically for chosen heights and plots it. HL derives the closed form for the optimal angle and studies the limits. Top marks come from comparing with real shot put release data and discussing air resistance.",
   "whereMarksAreLost": "E: numerical results only, without the general formula. B: long algebra with no signposting. D: ignoring air resistance and human release limits.",
   "dataNote": "Mainly theory, with optional video tracking data that is easy to collect.",
   "verdict": "Projectiles in sport are named in the subject reports among topics seen too often, but the closed form for the best angle is demanding and tractable. Using release data from your own sport gives it a strong personal angle."
  },
  {
   "id": "fastest-route-across-sand-and-sea",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Fastest route across sand and sea",
   "researchQuestion": "How does the fastest crossing point on a straight beach change as the ratio of running speed to swimming speed varies, and how well does the calculus model match timed trials on a real beach or playing field?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Pythagoras to build the total time T(x) = √(a²+x²)/v₂ + √(b²+(d−x)²)/v₁; differentiation of composite functions with the chain rule; solving T′(x)=0 and justifying a minimum with the sign of T′ or the second derivative; sensitivity of the optimum to the speed ratio, using a graphing tool; comparison with the straight-line path and the path that runs as far as possible on the fast surface. HL extension: rewrite the condition as sin θ₁/sin θ₂ = v₁/v₂ and link it to Snell's law. Squaring T′(x) = 0 gives a quartic, which is why a numerical solution is the honest route.",
   "howToStart": "Pick a scenario with two surfaces: for example a grass pitch next to a sandy strip, or a pool edge and water. Time yourself over 20 m on each surface to get v₁ and v₂. Then write T(x) for one fixed start and finish, plot it in Desmos or GeoGebra, and find the minimum before touching any derivative.",
   "slVsHl": "SL students can complete a sound piece by setting up T(x), differentiating, solving numerically, and testing the model against measured times. HL depth comes from expressing the optimum as an angle condition, proving it is Snell's law, and exploring a family of speed ratios.",
   "whereMarksAreLost": "E: stopping at a single worked example, which is a textbook optimisation problem, without generalising over the speed ratio. C: using invented speeds instead of measured ones. D: no discussion of model limits such as acceleration, a swimmer drifting, or wading speed. B: leaving the algebra of the derivative unexplained.",
   "dataNote": "The model needs no data, and the optional timing experiment is simple to run with a stopwatch and a measuring tape.",
   "verdict": "A classic and safe choice, but well known, so it only scores well if you gather your own speeds and go beyond one solved case. Set it on a course you really use, such as your school field and a sandy area, and study how the optimum moves with the speed ratio."
  },
  {
   "id": "finding-the-basketball-shot-with-most-room-for-error",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Finding the basketball shot with most room for error",
   "researchQuestion": "For a free throw, which release angle allows the widest range of release speeds or angles to still score, and how does this differ from the angle that needs the least speed?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Projectile motion written as y = x tanθ − gx²/(2v²cos²θ); ball and rim geometry giving the entry angle condition; differentiation to optimise both minimum speed and tolerance window; sensitivity analysis using partial changes in v and θ. Beyond syllabus: a simple comparison with a no-drag versus linear-drag model, only if numerical work is done in a spreadsheet.",
   "howToStart": "Set the hoop at 3.05 m, release height at about 2.0 m, free-throw distance 4.57 m. Solve for the speed needed at each angle from 40 to 70 degrees, then for each angle find the range of speeds that still lets a ball of 24 cm diameter pass a 45 cm rim. Plot the width of that range against angle.",
   "slVsHl": "SL: work out the trajectory, find the minimum speed angle by differentiation and compare it with the widest tolerance window found numerically. HL: derive the tolerance width analytically using implicit differentiation, and add air resistance through a differential equation solved by Euler's method.",
   "whereMarksAreLost": "E: stopping at one optimum and not justifying why tolerance is a fair objective. B: unclear notation for the two different optimisation problems. D: no comment on ignoring spin and drag. C: generic sports link with no personal shots or video evidence.",
   "dataNote": "Everything can be simulated from known heights and distances, though filming a few real shots with a phone would add credibility.",
   "verdict": "Projectiles in sport are named in the subject reports among topics seen too often, but the idea that different objectives give different best angles is genuinely interesting. Make it personal by filming your own shots and comparing your angle with the predicted one."
  },
  {
   "id": "gabriel-s-horn-paint-paradox-and-improper-integrals",
   "topic": "5",
   "topicName": "Calculus",
   "course": "AA",
   "level": "HL",
   "title": "Gabriel's Horn: paint paradox and improper integrals",
   "researchQuestion": "Why does rotating y = 1/x for x at least 1 give a solid of finite volume but infinite surface area, and does the paradox survive when the curve is changed to y = 1/x^p?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Integrals with an infinite upper limit, treated as limits; volume π∫y² dx; surface area 2π∫y√(1 + y′²) dx (beyond the syllabus); comparison by bounding the integrand from below; convergence of ∫ x^(−p) dx. Numerical integration to watch the divergence.",
   "howToStart": "Compute the volume exactly with a limit at b, then bound the surface integrand below by 2 pi/x to prove divergence. Then repeat for general p and find the range where each integral converges.",
   "slVsHl": "Volumes of revolution are HL, and the surface area integral is beyond the syllabus but needs only a lower bound, so this suits HL. Top marks come from the general p analysis and an honest discussion of what the paradox means physically.",
   "whereMarksAreLost": "E: only reproducing the standard argument. B: skipped steps in the divergence bound. C: no personal angle beyond repeating a famous result.",
   "dataNote": "Pure theory with clean algebra, but the surface area formula needs careful justification.",
   "verdict": "Very well known, so the plain version is weak. Generalising to 1/x^p and finding where the paradox exists makes it your own."
  },
  {
   "id": "how-container-shape-decides-the-time-to-drain",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "How container shape decides the time to drain",
   "researchQuestion": "How does the profile of a container determine the time for water to drain through a small hole, and why do real draining times exceed the ideal prediction?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Torricelli's law v = √(2gh); conservation of volume giving A(h) dh/dt = −a√(2gh); separation of variables and integration for cylinder, cone and paraboloid; comparison of predictions with measured times; introduction of a discharge coefficient fitted from data. Beyond syllabus: shaping a container so the water level falls at a constant rate.",
   "howToStart": "Drill a 4 mm hole near the base of a cylindrical bottle, record water height against time on video, and derive the ideal dh/dt. Compare the predicted curve with the data and estimate the correction factor.",
   "slVsHl": "Separable differential equations are HL, so the main analysis is HL. An SL student could study only the cylinder using a function fit to the data and a simple integral, but with less depth.",
   "whereMarksAreLost": "E: solving the equation but not linking to the experiment. D: not explaining the discrepancy in terms of contraction of the jet or viscosity. A: video timing errors ignored. B: units and notation for area not defined.",
   "dataNote": "Needs a careful experiment and timing from video, but equipment is cheap.",
   "verdict": "Very good for an HL student who likes experiments. The discrepancy with the ideal model is the part that shows real reflection, so plan enough trials."
  },
  {
   "id": "how-many-maclaurin-terms-are-enough",
   "topic": "5",
   "topicName": "Calculus",
   "course": "AA",
   "level": "HL",
   "title": "How many Maclaurin terms are enough?",
   "researchQuestion": "How many terms of the Maclaurin series of eˣ, sin x and cos x are needed to stay within 10⁻⁶ of the true value across intervals of increasing width, and why do the functions behave differently?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Maclaurin series of eˣ, sin x, cos x; partial sums and error against the true function; Lagrange remainder or alternating series error bound; comparison of bound with actual error; growth of the required n with interval half-width. Beyond syllabus: range reduction as used in calculators.",
   "howToStart": "Write partial sums in a spreadsheet or Python for x in [−1,1], then [−3,3] and [−10,10]. For each, find the smallest n that keeps the maximum error below 10⁻⁶ and plot n against the interval width.",
   "slVsHl": "Maclaurin series are HL only, so this is an HL idea. It gains depth by proving the remainder bound and asking how a calculator can avoid huge n by using periodicity.",
   "whereMarksAreLost": "E: numerical results without proving the bound. A: repetitive tables for each function. D: not asking why the bound overestimates. C: no motivation beyond a textbook exercise.",
   "dataNote": "Pure theory with easy computation, but the remainder proof needs to be explained carefully.",
   "verdict": "A solid HL idea, though it can drift into an exercise. The range reduction question, or how a calculator actually computes sin, gives it a personal purpose."
  },
  {
   "id": "how-quickly-numerical-integration-closes-in-on",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "How quickly numerical integration closes in on π",
   "researchQuestion": "How does the error in estimating π from the area under 4/(1+x²) fall with the number of intervals for left, midpoint, trapezium and Simpson's rules?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Definite integral ∫₀¹ 4/(1+x²) dx = π; left Riemann sums, midpoint and trapezium rules; log-log plots of error against n to estimate the order of convergence. HL: the error of one trapezium strip from a Taylor expansion. Beyond syllabus: Simpson's rule and its error term.",
   "howToStart": "In a spreadsheet or Python, compute each rule for n = 2, 4, 8, ... 1024. Record |estimate − π| and plot log(error) against log(n) to read off the gradient for each method.",
   "slVsHl": "SL: implement the rules, find the gradients experimentally and explain them using the shape of the curve. HL: derive the h² error of the trapezium and midpoint rules from Taylor expansions over one strip, compare with the observed gradients, and explain why the midpoint error is about half the trapezium error with opposite sign.",
   "whereMarksAreLost": "E: pages of tables with no explanation of why the gradients appear. A: too many similar tables. D: no comparison between measured and predicted convergence rates. C: little personal angle unless the student links to how calculators find π.",
   "dataNote": "Pure computation, no data collection, though care is needed with floating point rounding for very large n.",
   "verdict": "Solid and safe, but common in spirit, so the log-log analysis and theoretical prediction must carry the work. A twist is to compare with a series method for π, such as Machin-type arctan formulas."
  },
  {
   "id": "how-the-step-size-governs-euler-s-method-error",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "How the step size governs Euler's method error",
   "researchQuestion": "How does the global error of Euler's method depend on the step size h, and can the proportionality be explained using Taylor's theorem?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Euler's method; comparison with exact solutions of dy/dx = y and dy/dx = x + y; Taylor expansion to identify local error of order h²; accumulation to global error of order h; log-log plot of error against h. Beyond syllabus: the improved Euler or midpoint method.",
   "howToStart": "Code Euler in a spreadsheet for dy/dx = y with y(0) = 1, using h = 0.5, 0.25, 0.125, and so on. Compute the error at x = 1 against e.",
   "slVsHl": "Euler's method is HL content. Depth comes from the Taylor argument and from comparing with a second order method to show the changed scaling.",
   "whereMarksAreLost": "E: observing that halving h halves error without explaining it. B: undefined local and global error. D: no discussion of rounding error at very small h.",
   "dataNote": "Exact solutions exist to test against, so the numerical data is clean.",
   "verdict": "A good, honest numerical exploration. The extra step of a second order method and a case where the exact solution is unknown makes it personal."
  },
  {
   "id": "minimising-cost-of-a-pipe-or-cable-route",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Minimising cost of a pipe or cable route",
   "researchQuestion": "What is the cheapest route for a cable or pipe across ground of two different costs per metre, and how sensitive is the optimum to changes in those costs?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Building a cost function with Pythagoras, differentiation, stationary points and second derivative test, constraint handling, sensitivity analysis by varying a parameter, graph checks. Beyond syllabus: Lagrange multipliers for a constrained version.",
   "howToStart": "Take a real situation such as a cable from a house to a point across a river with different costs on land and in water. Sketch it, define the variable, write cost as a function and graph it before differentiating.",
   "slVsHl": "SL can work with a single variable cost function, GDC graphing and a table for the minimum. HL treats it fully with derivatives, implicit relations or a family of cost ratios and a general formula for the optimum.",
   "whereMarksAreLost": "E: an optimisation problem that is exactly a textbook one. C: invented numbers with no source. D: no sensitivity or realism check, for example ignoring that real routes cannot be perfectly straight.",
   "dataNote": "Pure theory with real costs looked up, so data are simple and the maths is well controlled.",
   "verdict": "Workable but close to a stock exercise unless you make the sensitivity analysis the core. Use real quoted prices for a local project to make it yours."
  },
  {
   "id": "newton-raphson-against-bisection-convergence-speed-compared",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Newton-Raphson against bisection: convergence speed compared",
   "researchQuestion": "How many iterations do bisection and Newton-Raphson need to reach a given accuracy for different functions, and when does Newton-Raphson fail?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Differentiation, tangent lines, iteration and fixed points, error analysis, sign change and the intermediate value idea, order of convergence from log of error. HL extension: Taylor series argument for quadratic convergence and analysis of failures near a zero derivative.",
   "howToStart": "Code both methods in Python for f(x) = x³ - 2 and cos x - x. Record the error after each step and plot log(error) against iteration number.",
   "slVsHl": "SL students can compare the methods on several functions and explain results using tangent geometry. HL students can prove quadratic convergence with Taylor expansion and study cycles or divergence for awkward starting values.",
   "whereMarksAreLost": "E: presenting code as the maths. B: undefined error terms. D: no discussion of when each method is preferable.",
   "dataNote": "Simulation is quick and reproducible; rigour lies in the error analysis.",
   "verdict": "A solid, answerable exploration when it is a comparison of two algorithms with measured and predicted convergence. Make it yours with an equation you met elsewhere, such as in a physics problem."
  },
  {
   "id": "optimal-pricing-for-a-small-business-using-a-fitted-demand-curve",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Optimal pricing for a small business using a fitted demand curve",
   "researchQuestion": "What price maximises weekly profit for a school-run product such as cakes or badges, and how sensitive is that price to the demand model chosen?",
   "dataDifficulty": 2,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Linear and exponential regression for demand, revenue and profit functions, differentiation to find stationary points, second derivative test, price elasticity of demand, sensitivity analysis on the model's parameters. Beyond syllabus: compare with Lagrange-type constraints when stock is limited.",
   "howToStart": "Sell one item at four or five different prices over several days or run a survey of stated willingness to pay. Fit linear and exponential demand models and write the profit function for each.",
   "slVsHl": "SL fits a linear demand and optimises profit with the derivative. HL compares model families, uses price elasticity, or adds a capacity constraint.",
   "whereMarksAreLost": "E: a single quadratic profit function is a textbook exercise. C: generic business setting. D: model never questioned. A: too many tables.",
   "dataNote": "Real sales data is small and noisy, so honesty about uncertainty is needed.",
   "verdict": "With real prices and a comparison of demand models it is decent, and a genuine venture of your own, such as a school club stall, is what lifts it."
  },
  {
   "id": "period-doubling-and-chaos-in-the-logistic-map",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Period doubling and chaos in the logistic map",
   "researchQuestion": "At which parameter values does the logistic map x(n+1) = r x(n)(1 - x(n)) change its long-term behaviour, and how fast do these changes accumulate towards chaos?",
   "dataDifficulty": 1,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Iterated functions and recurrence relations; fixed points found by solving f(x) = x; stability from the sign and size of |f'(x)| at the fixed point; period-two points by solving f(f(x)) = x; bifurcation diagram from spreadsheet or Python iteration; estimating the Lyapunov exponent as the mean of ln|f'(x_n)|. Beyond syllabus but reachable: Feigenbaum's constant from successive bifurcation values.",
   "howToStart": "Iterate the map in a spreadsheet for r = 2.8, 3.2, 3.5 and 3.9 from two starting values that differ by 0.001. Then derive the fixed points and the condition |f'| < 1 by hand and compare with what the iteration shows.",
   "slVsHl": "SL students can do the fixed point and derivative stability analysis and a simulated diagram. HL adds the algebra of the period-two cycle, the Lyapunov exponent as an average of logarithms and the ratio of successive bifurcation gaps.",
   "whereMarksAreLost": "E: relying on plots with no analytic derivation of the first bifurcation at r = 3. A: too many similar diagrams. D: no comment on floating point error limiting the chaotic results.",
   "dataNote": "Data is generated by iteration and the analysis is algebraic, so the only care needed is with rounding in long runs.",
   "verdict": "Well known but still good if the algebra is done properly. Narrowing to one map, with a derived first bifurcation and a measured Lyapunov exponent, makes it a real investigation."
  },
  {
   "id": "sine-waves-that-add-up-to-a-square-wave",
   "topic": "5",
   "topicName": "Calculus",
   "course": "AA",
   "level": "HL",
   "title": "Sine waves that add up to a square wave",
   "researchQuestion": "How does the error in a partial Fourier sum of a square wave fall as more harmonics are added, and why does an overshoot remain near the jump?",
   "dataDifficulty": 2,
   "dataSource": "simulation",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Integration by parts; integrals of products of sines and cosines over a period (orthogonality); Fourier coefficients bₙ = 4/(nπ) for odd n; partial sums; mean-square error; maximum of the overshoot from the first turning point. Beyond syllabus: Gibbs constant ≈ 1.179 via the sine integral.",
   "howToStart": "Derive the coefficients by integration. Then plot partial sums with 1, 3, 5, 15 terms in Desmos or Python and record the mean-square error and the peak height near the jump.",
   "slVsHl": "This is AA HL, because of integration by parts and of trigonometric products over a period. Depth comes from showing numerically that the overshoot stays near 9% of the jump as n increases, and explaining why.",
   "whereMarksAreLost": "E: coefficient formulae quoted without derivation. B: unclear definitions of the periodic function. A: plots with no labelled axes. C: purely textbook.",
   "dataNote": "Calculations are manageable but the Gibbs proof is hard to control; numerical evidence can support it.",
   "verdict": "Impressive if the integrals are done properly and the overshoot is measured. Twist: link to audio synthesis by playing the partial sums."
  },
  {
   "id": "the-cheapest-open-box-for-a-fixed-volume",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "The cheapest open box for a fixed volume",
   "researchQuestion": "For an open-topped box with a square base and fixed volume, which dimensions minimise the material cost, and how do they change as the base becomes more expensive than the sides?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Constraint substitution to reduce to one variable, differentiation of x² + k/x type functions, first and second derivative tests, parameter k for the cost ratio, sketching the cost function. Beyond syllabus: showing the optimal height to base ratio depends on the cost ratio only.",
   "howToStart": "Set V = 1000 cm³, write the surface area as A(x) = x² + 4V/x, and find its minimum. Then multiply the base term by a cost ratio c and re-derive. Build two cardboard boxes to test the result.",
   "slVsHl": "SL solves the single case and one or two cost ratios with a graph and the derivative. HL treats the cost ratio as a general parameter, adds a rectangular base by fixing the ratio of its sides and optimising again, and studies how the solution scales with V.",
   "whereMarksAreLost": "E: a textbook exercise if kept at one case. D: no discussion of real packaging limits. C: nothing personal unless a real product is used.",
   "dataNote": "Pure algebra and calculus, with optional cardboard measurements.",
   "verdict": "Minimising the material of a box or can is named in the subject reports among topics seen too often, and the plain version is a standard exercise. It becomes acceptable only with the general cost parameter and a real package you compare against."
  },
  {
   "id": "the-goat-problem-grazing-area-with-a-rope-round-a-barn",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "The goat problem: grazing area with a rope round a barn",
   "researchQuestion": "What is the grazing area of an animal on a rope of length L tethered to the wall of a rectangular barn, and how does the tether point maximise it?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Area of a sector and circle segments, arc length, intersection of circles solved with trigonometry, optimisation by differentiation to choose the tether position, and a transcendental equation solved numerically for the circular field version. Beyond syllabus: the exact solution of the circular field version, published only in 2020.",
   "howToStart": "Draw a rope of length L tied at the corner of a square barn of side a, split the region into sectors and compute the area in the cases L < a and a < L < 2a.",
   "slVsHl": "SL does the rectangular barn with sectors and one optimisation. HL adds the circular field, where the intersection area gives an equation with an angle that must be solved numerically, with integration for the lens region.",
   "whereMarksAreLost": "E: just applying a sector formula. A: unclear diagrams. C: playful but impersonal. D: no check of the limit cases (L very small or very large).",
   "dataNote": "Pure geometry, but the circular case has no closed form and needs a numerical solution.",
   "verdict": "A classic that moderators know well, but the barn version with case splitting gives real work. Make it your own with a real field or shed dimensions."
  },
  {
   "id": "the-safety-parabola-boundary-of-all-projectile-paths",
   "topic": "5",
   "topicName": "Calculus",
   "course": "AA",
   "level": "both",
   "title": "The safety parabola: boundary of all projectile paths",
   "researchQuestion": "For a fixed launch speed, what curve bounds every reachable point, and how can it be derived by two different methods?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Write the trajectory y = x tan θ − gx²(1 + tan²θ)/(2u²). Treat it as a quadratic in tan θ and set the discriminant to zero to find the boundary. Compare the vertex and focus of the resulting parabola. Show by differentiation that each trajectory touches it. Extend to a launch from a height or up a slope. Beyond syllabus: the envelope condition by differentiating with respect to the parameter θ.",
   "howToStart": "Take the trajectory equation with tan θ as the parameter, find the discriminant condition, and plot a family of trajectories in Desmos with the envelope on top.",
   "slVsHl": "An SL student can do the discriminant method and the graphs with a clear geometric interpretation. HL depth comes from the second derivation by the envelope condition, the tangency proof and the sloped-ground extension.",
   "whereMarksAreLost": "E: presenting only the discriminant trick with no justification of why it gives the envelope. B: unclear notation for parameter versus variable. C: no personal angle; a real scenario such as a water fountain or a sports throw helps.",
   "dataNote": "Pure theory with a manageable proof; the challenge is explaining envelopes clearly to a reader.",
   "verdict": "Elegant and less overdone. Worth choosing if the two derivations are compared and a real context, such as a garden sprinkler, gives it a personal hook."
  },
  {
   "id": "volume-and-surface-area-of-a-torus-by-two-methods",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "Volume and surface area of a torus by two methods",
   "researchQuestion": "Do the disc/washer method and Pappus's theorem give the same volume and surface area for a torus, and how does a horn or spindle torus change the result?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Volume of revolution about the y-axis by washers, integration of √(r² − y²) using the area of a semicircle or a trigonometric substitution, Pappus's theorem as a second method, surface area of revolution (beyond the syllabus), parametric equations of a circle.",
   "howToStart": "Write the circle (x − R)² + y² = r², rotate it about the y-axis using washers, and integrate with respect to y to get 2π²Rr². Then check with a physical torus such as a ring or bagel measured by water displacement.",
   "slVsHl": "Volumes of revolution are HL in both courses. HL students get the volume by washers, confirm it with Pappus's theorem, add the surface area, and study the case R < r where the shape intersects itself.",
   "whereMarksAreLost": "E: the maths reduces to one integral with no comparison of methods. A: unclear diagrams of the rotation. D: no check of the result against a real object.",
   "dataNote": "Theory with an optional water displacement check.",
   "verdict": "Volumes of revolution are named in the subject reports among topics seen too often. Workable if kept to two methods that check each other, and a measurement of a real ring makes it personal."
  },
  {
   "id": "volume-of-a-decanter-solids-of-revolution-against-water",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "HL",
   "title": "Volume of a Decanter: Solids of Revolution Against Water",
   "researchQuestion": "How closely does the volume of a real decanter, found by rotating a fitted profile curve, agree with the volume measured by filling it with water?",
   "dataDifficulty": 1,
   "dataSource": "experiment",
   "sitesListingIt": 1,
   "namedInReports": true,
   "mathsRequired": "Volume of revolution about the x-axis, curve fitting to measured radii (cubic or trigonometric regression), definite integration, numerical methods such as the trapezium rule and a stack of cylinders or frustums for comparison, percentage error. HL: integration by substitution or parts for a sinusoidal profile, volume rotated about the y-axis.",
   "howToStart": "Photograph your decanter against a ruler and trace its outline, or measure the radius every 1 cm of height. Fit a function to the radii, integrate to get the volume, then measure the true volume with a measuring jug.",
   "slVsHl": "Volumes of revolution are HL in both courses. HL uses a fitted profile, the volume integral and a comparison with a stack of cylinders or frustums and with the measured volume, with error analysis across several fitted models. An SL student could still compare a stack of cylinders with the measured volume, but the integral itself is HL.",
   "whereMarksAreLost": "E: an easy shape fitted with a single function and no comparison of methods. C: shape chosen without personal reason. D: no reflection on the wall thickness that reduces the internal volume. A: photographs poorly scaled.",
   "dataNote": "All data come from a simple measurement at home.",
   "verdict": "Volumes of revolution are named in the subject reports among topics seen too often, but this is a solid version if you measure the wall thickness and compare at least three methods. Use an object with a story for you, such as a family bottle."
  },
  {
   "id": "why-a-parabolic-mirror-sends-light-to-one-point",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Why a parabolic mirror sends light to one point",
   "researchQuestion": "For the curve y = x²/(4p), how can the reflection law, applied with tangent gradients, be shown to send every beam travelling parallel to the symmetry line through the single point (0, p), and how does a spherical mirror of similar width compare in the spread of its reflected beams?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Differentiation to find the tangent and normal gradient at a general point; angle between two lines using tan of a difference or the scalar product; reflection law as equality of angles with the normal; focus and directrix definition of the parabola; equation of a line and substitution to test a fixed point. HL extension: vector reflection formula r = d − 2(d·n)n, and the spread of reflected rays from a circular mirror. Beyond syllabus: the envelope (caustic) of rays reflected in a circle, a nephroid.",
   "howToStart": "Take y = x²/(4p) with p = 1. Pick the point (2, 1), find the tangent gradient, the normal, and the reflected direction of a vertical ray using the equal angles law. Check by hand that the reflected line passes through (0, 1). Then repeat with a general point (2pt, pt²) and simplify. Plot in Desmos or GeoGebra with several rays.",
   "slVsHl": "SL: do the general-point algebra with tan of the angle between lines, prove the focus property, and test it numerically on a few rays, then compare with a circle of matching curvature at the vertex. HL: use vector reflection, derive the caustic of the circular mirror as an envelope of lines, and quantify the focusing error (distance of reflected rays from the ideal point) as the beam widens, for a strong evaluation.",
   "whereMarksAreLost": "E: stopping at the standard proof found in any textbook, so no real choice of mathematics. C: no personal link, so it reads as copied. D: reflection that only repeats the result instead of judging limits such as real mirrors, ray width and the idealisation of light as rays. A/B: unlabelled diagrams and undefined notation for the angles.",
   "dataNote": "Pure theory with a graphing tool for checking, so no data is needed, but the proof alone is too short for 12 to 20 pages unless the circle comparison is added.",
   "verdict": "The classic proof is a textbook exercise, so it is only worth choosing with a twist: a comparison of parabolic and spherical mirrors, with the focusing error as the beam widens, or a real solar cooker or satellite dish with measured dimensions."
  },
  {
   "id": "why-athletics-tracks-are-not-circles",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Why athletics tracks are not circles",
   "researchQuestion": "For a track of fixed 400 m perimeter with straights of at least a given length, how does the maximum enclosed area compare with a circle, and how does the optimal bend radius change with the constraint?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Perimeter constraint 2L + 2πr = 400, area A = 2rL + πr², substitution, differentiation, endpoint analysis on a constrained interval, comparison with circle area 400²/4π. Beyond syllabus: showing a stadium shape is the best under a straight length restriction.",
   "howToStart": "Write A(r) after eliminating L, sketch it on the allowed interval and find where it is largest. Then check the real IAAF standard of 36.5 m radius and compare its area.",
   "slVsHl": "SL optimises with one constraint and comments on the result. HL introduces a required minimum straight length, studies how the optimum moves along the boundary and explains when the constraint is active.",
   "whereMarksAreLost": "E: the unconstrained result is trivial (a circle, or a degenerate track). B: unclear about which constraint is binding. D: little link to real track design.",
   "dataNote": "Pure theory with real track dimensions available from athletics rules.",
   "verdict": "Neat and the maths is honest, but the answer without constraints is dull. Its strength is the endpoint reasoning and the comparison with the real standard."
  },
  {
   "id": "why-e-compound-interest-limit-and-the-derivative-of-a-x",
   "topic": "5",
   "topicName": "Calculus",
   "course": "both",
   "level": "both",
   "title": "Why e: compound interest limit and the derivative of a^x",
   "researchQuestion": "Why is e the unique base for which the gradient of a^x equals a^x at every point, and how does this link to the limit of (1 + 1/n)^n?",
   "dataDifficulty": 1,
   "dataSource": "theory",
   "sitesListingIt": 1,
   "namedInReports": false,
   "mathsRequired": "Limit definition of the derivative, numerical limits, binomial expansion for the limit of (1 + 1/n)^n, derivative of a^x from first principles, inverse function and ln x, Maclaurin series for e^x (HL).",
   "howToStart": "Compute the gradient of 2^x, 3^x and 2.7^x at x = 0 numerically by shrinking h, and find the base that gives exactly 1. Then relate this to compounding n times a year.",
   "slVsHl": "SL students can do the numerical and compound interest route and show the derivative result. HL students prove convergence with the binomial theorem, use Maclaurin series and prove d/dx ln x = 1/x from the inverse.",
   "whereMarksAreLost": "E: statements taken from the textbook without proof. C: generic history of e. B: limits written loosely.",
   "dataNote": "Pure theory; the challenge is proving the limit exists rather than assuming it.",
   "verdict": "A very common topic and risky as a plain description. It only earns marks if it proves the limit converges and explains why the derivative property forces this base."
  }
 ]
}