| name | mathematical-problem-execution |
| description | Drive a concrete pure or applied mathematics problem from exact statement through method selection, derivation, proof or computation, executable checks, error control, edge cases, and machine-readable evidence without promoting bounded tests to proofs. |
| license | MIT |
Mathematical Problem Execution
Decision contract
Restate the target, domain, quantifiers, givens, unknowns, regularity assumptions, conventions, requested output, and acceptable evidence level. Separate supplied facts from introduced assumptions. Classify the task as proof/refutation, exact computation, approximation, optimization, differential equation, inverse problem, or mathematical model.
Give every material claim a stable ID and decide which independent check would falsify it before doing the main derivation.
Reference usage
Read the problem execution and routing contract before selecting an executable check, assigning a proof status, or packaging a result. It maps problem classes to quantitative commands and defines evidence levels, required cross-checks, and stop conditions.
Record material reference hashes. Preserve computations with $science-provenance and use $science-review when the result supports a scientific, engineering, or published mathematical claim.
Workflow
- Specify the exact statement and identify missing conditions that change existence, uniqueness, or equivalence.
- Route to
$cx-proof-and-counterexample, $cx-linear-algebra-problem-solving, $cx-ode-pde-solving, $cx-numerical-analysis-error-control, $cx-dimensional-analysis-units, $cx-experimental-uncertainty-propagation, or $cx-formal-theorem-proving as needed.
- Work from definitions and state every theorem with the hypotheses actually used.
- Prefer exact arithmetic and symbolic structure until approximation is necessary. Track branches, domains, one-way implications, and lost solutions.
- Use bounded CLIs only for their documented purpose: counterexample, statistics, numerical verification, dimensions, uncertainty, or formal checking.
- Require at least two applicable checks: substitution, independent derivation, limiting cases, exact versus high-precision evaluation, invariant, residual, condition estimate, or formal kernel check.
- Package claims and receipts with the narrowest supported statuses and explicit limitations.
Outputs
- exact problem statement, assumptions, and claim IDs;
- derivation or proof with theorem dependencies;
- applicable
mathematical-claim, counterexample-search, proof-obligation-graph, proof-receipt, numerical-verification, dimension-check, or uncertainty-propagation artifacts;
- independent checks and failed cases;
- final result, validity region, error or uncertainty, and unresolved obligations;
- manifest and review receipt for material work.
Boundaries
- Computer algebra output is not proof until domains, branches, assumptions, and transformations are checked.
- Random examples can falsify a universal claim but cannot prove one.
- Numerical agreement does not establish exact equality or general convergence.
- A model consequence is conditional on the model assumptions and is not empirical evidence by itself.
- Do not hide non-uniqueness, singular cases, lost roots, or failed proof obligations.
- Stop and return
conditional, tested, unavailable, or a corrected theorem rather than inventing missing assumptions or unsupported digits.
Source basis
This workflow is an original synthesis of the openly licensed mathematical texts recorded in ../../docs/TEXTBOOK_SOURCES.md; it does not reproduce their prose or exercises.