{
 "e111_skills": [
  {
   "acquisition": "synthesized from failing-test feedback; no model weights changed",
   "adversarial": 1.0,
   "baseline": 0.4231,
   "baseline_correct": 11,
   "baseline_only": 0,
   "budget": {
    "cpu_s": 30,
    "memory_mb": 128
   },
   "dev_n": 4,
   "domain": "software",
   "effects": "none outside the temp dir",
   "held_n": 26,
   "held_out": 0.8846,
   "inputs": "candidate source files + tests",
   "limits": "cannot repair defects outside the edit grammar; tests are the only oracle",
   "mcnemar_exact_p": 0.00048828,
   "name": "Test-guided repair inside a bounded edit grammar",
   "outputs": "a source that passes, or ABSTAIN",
   "per_case_base": [
    0,
    1,
    1,
    0,
    0,
    0,
    0,
    1,
    0,
    1,
    0,
    1,
    1,
    0,
    0,
    0,
    0,
    0,
    0,
    1,
    0,
    1,
    0,
    1,
    1,
    1
   ],
   "per_case_proc": [
    1,
    1,
    1,
    1,
    1,
    0,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    0,
    1,
    0,
    1,
    1,
    1
   ],
   "permissions": "read-only temporary directory; no network; subprocess timeout 10 s",
   "preconditions": [
    "pure function",
    "tests supplied",
    "bounded candidate set"
   ],
   "procedure_correct": 23,
   "procedure_only": 12,
   "promoted": true,
   "revocation": "registry record -> REVOKED; callers must consult the registry",
   "safety": "never returns a candidate that did not pass every supplied test",
   "skill": "S1_software_repair",
   "status": "PROMOTED",
   "version": "1.1.0",
   "wilson_95": [
    0.7102,
    0.96
   ]
  },
  {
   "acquisition": "synthesized from controlled-intervention design",
   "adversarial": 1.0,
   "baseline": 0.9,
   "baseline_correct": 36,
   "baseline_only": 0,
   "budget": {
    "cpu_s": 1,
    "memory_mb": 32
   },
   "dev_n": 4,
   "domain": "science",
   "effects": "none",
   "held_n": 40,
   "held_out": 1.0,
   "inputs": "rows (x, z, y, intervened)",
   "limits": "binary variables; no continuous SCMs; unmeasured confounding not modelled",
   "mcnemar_exact_p": 0.125,
   "name": "Intervention-based causal identification (do-test)",
   "outputs": "X_CAUSES_Y / CONFOUNDED / ABSTAIN",
   "per_case_base": [
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    0,
    0,
    0,
    0
   ],
   "per_case_proc": [
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1
   ],
   "permissions": "none",
   "preconditions": [
    "interventional rows present",
    "binary treatment/outcome"
   ],
   "procedure_correct": 40,
   "procedure_only": 4,
   "promoted": false,
   "revocation": "registry record -> REVOKED",
   "safety": "abstains when no intervention is available instead of guessing",
   "skill": "S2_science_causal",
   "status": "NOT_PROMOTED",
   "version": "1.1.0",
   "wilson_95": [
    0.9124,
    1.0
   ]
  },
  {
   "acquisition": "synthesized from data-quality invariants (unique key, robust centre)",
   "adversarial": 1.0,
   "baseline": 0.027,
   "baseline_correct": 1,
   "baseline_only": 0,
   "budget": {
    "cpu_s": 1,
    "memory_mb": 32
   },
   "dev_n": 4,
   "domain": "data",
   "effects": "none",
   "held_n": 37,
   "held_out": 0.973,
   "inputs": "rows",
   "limits": "assumes at most one row per id and a unimodal clean distribution",
   "mcnemar_exact_p": 0.0,
   "name": "Duplicate/outlier-aware summary statistic",
   "outputs": "clean mean or ABSTAIN",
   "per_case_base": [
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    1
   ],
   "per_case_proc": [
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    0,
    1,
    1,
    1,
    1,
    1
   ],
   "permissions": "none",
   "preconditions": [
    "tabular (id, value) rows"
   ],
   "procedure_correct": 36,
   "procedure_only": 35,
   "promoted": true,
   "revocation": "registry record -> REVOKED",
   "safety": "does not silently drop values without a stated rule",
   "skill": "S3_data_quality",
   "status": "PROMOTED",
   "version": "1.1.0",
   "wilson_95": [
    0.8618,
    0.9952
   ]
  },
  {
   "acquisition": "synthesized from formal semantics",
   "adversarial": 1.0,
   "baseline": 0.3333,
   "baseline_correct": 13,
   "baseline_only": 0,
   "budget": {
    "cpu_s": 1,
    "memory_mb": 16
   },
   "dev_n": 4,
   "domain": "mathematics",
   "effects": "none",
   "held_n": 39,
   "held_out": 1.0,
   "inputs": "formula string",
   "limits": "propositional only; no quantifiers or arithmetic theories",
   "mcnemar_exact_p": 3e-08,
   "name": "Propositional validity by exhaustive truth table",
   "outputs": "VALID / INVALID / ABSTAIN",
   "per_case_base": [
    0,
    1,
    0,
    0,
    0,
    1,
    1,
    0,
    0,
    1,
    0,
    0,
    1,
    0,
    0,
    0,
    1,
    0,
    0,
    0,
    1,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    1,
    0,
    1,
    1,
    1,
    0,
    1,
    0,
    0,
    1
   ],
   "per_case_proc": [
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1
   ],
   "permissions": "none",
   "preconditions": [
    "formula over at most 3 atoms",
    "fixed grammar"
   ],
   "procedure_correct": 39,
   "procedure_only": 26,
   "promoted": true,
   "revocation": "registry record -> REVOKED",
   "safety": "abstains on malformed input; never evaluates untrusted code",
   "skill": "S4_math_validity",
   "status": "PROMOTED",
   "version": "1.1.0",
   "wilson_95": [
    0.9103,
    1.0
   ]
  },
  {
   "acquisition": "acquired in software change control; applied to business/ops decisions",
   "adversarial": 1.0,
   "baseline": 0.7222,
   "baseline_correct": 26,
   "baseline_only": 0,
   "budget": {
    "cpu_s": 1,
    "memory_mb": 16
   },
   "dev_n": 4,
   "domain": "cross_domain",
   "effects": "none; a human applies changes",
   "held_n": 36,
   "held_out": 1.0,
   "inputs": "change with gain, invariants, applicability",
   "limits": "invariants are supplied; the skill does not discover them",
   "mcnemar_exact_p": 0.00195312,
   "name": "Invariant-gated operational change (software rollback practice -> ops decisions)",
   "outputs": "APPLY / REJECT",
   "per_case_base": [
    1,
    0,
    0,
    1,
    0,
    1,
    1,
    1,
    1,
    0,
    1,
    1,
    1,
    1,
    0,
    1,
    0,
    1,
    1,
    1,
    0,
    1,
    0,
    1,
    0,
    1,
    1,
    1,
    1,
    0,
    1,
    1,
    1,
    1,
    1,
    1
   ],
   "per_case_proc": [
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1
   ],
   "permissions": "none (recommendation only)",
   "preconditions": [
    "declared invariants",
    "explicit applicability flag"
   ],
   "procedure_correct": 36,
   "procedure_only": 10,
   "promoted": true,
   "revocation": "registry record -> REVOKED",
   "safety": "checks applicability first; refuses to apply a rule outside its valid domain",
   "skill": "S5_ops_invariant_gate",
   "status": "PROMOTED",
   "version": "1.1.0",
   "wilson_95": [
    0.9036,
    1.0
   ]
  }
 ],
 "schema": "cain.e117.skills.v1",
 "skills": [
  {
   "baseline": "B_single",
   "baseline_acc": 0.0,
   "candidate": "F_full",
   "candidate_acc": 1.0,
   "grants_authority": false,
   "id": "C1_compositional_synthesis",
   "p": 0.00048828,
   "promoted": true,
   "provenance": "E117 sealed OOD",
   "revoked": false
  },
  {
   "baseline": "D_compose",
   "baseline_acc": 0.0,
   "candidate": "F_full",
   "candidate_acc": 1.0,
   "grants_authority": false,
   "id": "C2_precondition_discovery",
   "p": 0.03125,
   "promoted": true,
   "provenance": "E117 sealed OOD",
   "revoked": false
  },
  {
   "baseline": "E_compose_pre",
   "baseline_acc": 0.5714,
   "candidate": "F_full",
   "candidate_acc": 1.0,
   "grants_authority": false,
   "id": "C3_evidence_revision_abstention",
   "p": 0.03125,
   "promoted": true,
   "provenance": "E117 sealed OOD",
   "revoked": false
  }
 ]
}
