feat: infinite practice generator + Verify Solution (roadmap #5, #8)#80
Merged
Conversation
…ded answers
- New string-seeded RNG (xmur3 + mulberry32); seeds minted via
crypto.getRandomValues — no Math.random anywhere in generation
- templateEngine.generate(id, seed) is now pure and seed-required:
the RNG is a local stream threaded through parameter generation
(the mutable Math.random field and old LCG are gone)
- ProblemInstance carries its seed (URL round-tripping) and an optional
machine-gradable answer computed at generate() time from a new
per-template canonical() spec: exact roots via the closed-form
solver, expressions for CAS-equivalence grading, numeric values with
per-answer tolerance, and labeled assignments for systems
- canonical() added to 22 of 24 templates (prime-factorization stays
textual; linear-inequality answers carry an operator the equation
solver cannot support, so it deliberately has no graded answer)
- Fix latent system-linear-2x2 bug: the det != 0 constraint sat on c2,
which the determinant does not involve — retries could never repair a
singular draw; moved to b2 where at most one value in range is bad
- generateBatch(id, count, baseSeed?) derives reproducible per-item
seeds as `${baseSeed}:${i}`
- './problems/templates' and './equivalence' subpath exports
- Tests: seeded determinism across all 24 templates, 100-seed
constraint sweeps, and back-substitution of every graded root into
the ORIGINAL equation (|LHS-RHS| < 1e-6)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…bing
checkEquivalence(student, canonical): parse both, prove
simplify(student - canonical) == 0 via symbolic/simplify + CAS; fall
back to deterministic seeded numeric probing (24 points in [-5,5] plus
positive-domain points for sqrt/log; singularities skipped; >=8 valid
samples required; mixed abs/rel tolerance 1e-9). Conservative by
construction: any mismatch rejects, insufficient samples reports
'inconclusive' — it never claims equivalence it cannot support.
checkGradedAnswer grades free-form input against GradedAnswer kinds:
multi-solution lists ('x = 2 or x = 3', '2; 3'), numeric values
(expression input like 1/2 or 16*pi accepted), labeled assignments
('x = 2, y = 3' any order, swapped values rejected), and canonical
expressions with '+ C' and assignment-prefix stripping.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- DrillMode component: seed-deterministic problems from the template
engine, graded via checkGradedAnswer (equivalent forms accepted),
commonMistakes explanations on wrong answers, progressive hints,
per-problem timer (useEffectEvent — no legacy ref hacks), running
tally chips, Copy-link sharing, template picker grouped by category
- Practice page: ?mode=drill&template=…&seed=… deep links via
useSearchParams (Suspense-wrapped) with native
window.history.replaceState/pushState shallow URL updates; a missing
seed is minted (crypto.getRandomValues) straight into the URL so
every drill link reproduces its exact problem; Infinite Drill entry
card in the config UI
- Session persistence: startPracticeSession on first checked answer,
completePracticeSession on End drill with new optional correctCount +
topicSlug fields feeding TopicProgress (generated problems write NO
per-attempt rows — Attempt.problemId is a required FK to Problem);
graceful sign-in messaging when unauthenticated
- CheckWorkPanel ('Check my work'): verifies any equivalent form against
a canonical expression, reporting verified-symbolically/numerically,
parse-error and inconclusive states; embedded in InteractiveSolver
(gated on isValidExpression) and SolverPanel (new finalRaw on both
compute paths; equations use the zero-form LHS-RHS so factorings
verify)
- submitAnswer grading upgraded in place (contract unchanged): exact
match first, then conservative CAS equivalence in try/catch
- i18n: 21 practice.drill.* + 11 checkWork.* keys fully translated in
en/ru/es/uk/de/fr/ja/zh (identical structure, ICU plurals per locale)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
DrillMode: deterministic statement for fixed template+seed, equivalent expression form '(c-b)/a' graded correct, sign-mistake answer graded incorrect with the commonMistakes explanation surfaced, Generate-another mints a fresh 8-char crypto seed, progressive hint reveal with count. CheckWorkPanel: equivalent (symbolic and numeric captions), wrong form, parse-error, and verdict-reset-on-edit states. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Check-my-work equation canonical is now the solved root, not the LHS-RHS zero form (correct answers were graded "not equivalent"); panel offered only for single-real-root equations - "+ C" strip is opt-in (indefinite-integral problems only) instead of running on every problem type in grading and check-work paths - numeric probing refuses to claim equivalence when floor/ceil/round is present — sample agreement on step functions proves nothing - duplicate-padded root lists no longer pass distinct-root problems - drill "End & save" reads sessionId from state, not a ref (button was stuck disabled); saved status tracked per segment - switching drill templates flushes the tally to the topic actually practiced instead of crediting everything to the final selection - probability-basic tolerance 1e-3 so its own displayed 3-decimal answer grades correct (matches geometry templates) - ungradeable templates (no canonical) excluded from the drill picker Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
…e-verify Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
checkEquivalence(symbolicsimplify(student − canonical) == 0with seeded numeric-probing fallback)?mode=drill&template=&seed=), Check my work panel in the solver; 32 i18n keys × 8 localesAdversarial-review fixes included
Verification
math-engine 2738 tests, web 284 tests — green; typecheck + biome clean. Wave-1 roadmap items #5 and #8.
🤖 Generated with Claude Code