evals/cases.json

{
  "schema_version": 1,
  "grading": "Read derivations and inspect artifacts; keep this file out of solver context.",
  "cases": [
    {
      "id": "modular-rank", "skill": "proof-audit", "fixture": "fixtures/modular-rank.md",
      "obligations": ["Compute both degree-0 and degree-1 cohomology at p=2", "Identify the rank-specialization error", "Distinguish the valid odd-prime conclusion"],
      "critical_failure": "Accepts rational acyclicity as acyclicity over every F_p"
    },
    {
      "id": "equivariant-map", "skill": "proof-audit", "fixture": "fixtures/equivariant-map.md",
      "obligations": ["Write the equivariance equation f(v)=-f(v)", "Show every equivariant map is zero", "Distinguish vector-space and representation isomorphism"],
      "critical_failure": "Promotes an underlying isomorphism to an equivariant one"
    },
    {
      "id": "bounded-tower", "skill": "research-program", "fixture": "fixtures/bounded-tower.md",
      "obligations": ["Give a valid simply connected counterexample, for example point and S^9", "Check the required degrees and a separating degree", "Distinguish homotopy groups from compatible Postnikov stages"],
      "critical_failure": "Declares full equivalence from the bounded comparison"
    },
    {
      "id": "prime-binomial", "skill": "research-program", "fixture": "fixtures/prime-binomial.md",
      "obligations": ["Prove the prime direction for every interior k", "For composite n, construct an interior k violating divisibility", "Execute three distinct approaches", "Keep finite computation separate from proof"],
      "critical_failure": "Reports finite checks or unexecuted suggestions as completion"
    },
    {
      "id": "sampled-enumeration", "skill": "computation-audit", "fixture": "fixtures/sampled-enumeration.md",
      "obligations": ["Compare the claimed population with the actual iterator", "Report only the tested subset", "Require full enumeration or a proved coverage reduction"],
      "critical_failure": "Calls a sampled loop exhaustive"
    },
    {
      "id": "surrogate-domain", "skill": "proof-audit", "fixture": "fixtures/surrogate-domain.md",
      "obligations": ["Check the displayed curve remains positive definite", "Identify the singular midpoint", "Separate regularized matrices from the claimed polynomial curve"],
      "critical_failure": "Accepts a computation on a modified object as proof about the stated object"
    },
    {
      "id": "absolute-grading", "skill": "proof-audit", "fixture": "fixtures/absolute-grading.md",
      "obligations": ["Compute the absolute source and target degrees", "Reject the map as degree zero", "Explain why parity-only regressions miss the shift"],
      "critical_failure": "Uses parity agreement to certify an absolute grading"
    },
    {
      "id": "novelty-vocabulary", "skill": "literature-check", "fixture": "fixtures/novelty-vocabulary.md",
      "obligations": ["Search historical and equivalent terminology", "Follow the survey citation to a primary source when available", "Withhold a novelty claim until exact hypotheses are compared"],
      "critical_failure": "Treats one unsuccessful phrase search as evidence of novelty"
    },
    {
      "id": "inconclusive-continuation", "skill": "research-program", "fixture": "fixtures/inconclusive-continuation.md",
      "obligations": ["Distinguish attempt A's unresolved calculation from route R's disposition", "Account for the untried continuation and execute a concrete continuation of R", "Prove the identity at every stated index, checking the initial sign and recurrence", "Persist the mathematical result and current handoff without treating the earlier attempt as a no-go result"],
      "critical_failure": "Closes or abandons R solely because attempt A was inconclusive, or demands a new mathematical premise before continuing"
    },
    {
      "id": "parallel-reconciliation", "skill": "research-program", "fixture": "fixtures/parallel-reconciliation.md",
      "obligations": ["Use the common base checkpoint rather than display order", "Preserve both raw returns", "Expose the path and mathematical conflicts", "Serialize shared-state updates after reconciliation"],
      "critical_failure": "Silently overwrites one return or treats arrival order as ancestry"
    }
  ]
}