| name | coq-honesty |
| description | Coq proof editing rules for Trinity S3AI — honesty tags, Admitted conventions, and proof structure |
| trigger | {"paths":["proofs/**/*.v","derivations/**/*.v"]} |
Coq Honesty Rules
Adding or Editing Coq Proofs
Honesty Tags (mandatory)
Every theorem that relies on a multi-atom phi/π/e formula or an unproven assumption must carry a tag:
| Tag | Meaning | When to use |
|---|
[phenomenological_fit] | Numerical coincidence with SM parameter | Formula matches PDG value but lacks physical derivation |
[NUMERICAL_FIT] | Explicit numerical fit (legacy) | Same as above, older notation |
[PHYSICAL_AXIOM] | Physically motivated assumption | Step needs physics justification, not just math |
[MATH_TODO] | Mathematical gap | Proof step is blocked by missing lemma |
[LIBRARY_GAP] | Missing Coq library | Blocked by unformalized external result |
[OPEN_PROBLEM] | Honestly open | Deliberately left open with citation |
Admitted Convention
- Never leave
Admitted. without a tag in committed code.
- If a theorem is genuinely open, tag it with
[OPEN_PROBLEM] and add a citation or cross-reference to docs/analysis/admitted_log.md.
- If a theorem is blocked by a missing lemma, tag it with
[MATH_TODO] or [LIBRARY_GAP] and name the blocker.
proofs/trinity/ targets 0 real Admitted — every Qed. must be a real proof.
proofs/clifford_cl8/ allows load-bearing Axioms only with citations (e.g., Lounesto 2001, Atiyah-Bott-Shapiro 1964).
Running the Honest Counter
After editing Coq files, run:
python3 scripts/count_admitted_honest.py
This strips all comments and strings before counting, so historical Admitted mentions in comments do not affect the tally.
Claim Status Mapping
| Coq Artifact | Claim Status |
|---|
Qed. on a theorem | verified |
Qed. on a refuted theorem | verified (impossibility result) |
Admitted. with [phenomenological_fit] tag | empirical_fit |
Admitted. with [OPEN_PROBLEM] tag | open_conjecture |
Axiom / Parameter with citation | verified (axiomatic base) |
Axiom / Parameter without citation | unverified — add citation or tag |
Boundary Theorems
BT-1 through BT-4 are impossibility results (proven in proofs/trinity/BoundaryTheorems.v). They prove that certain direct constructions fail. When citing them:
- Do not call them "dead ends" — they are guideposts.
- Always reference the specific BT number and the obstruction it proves.
Limits of Formal Proof (honest scope)
- Coq
Qed. on a mathematical theorem does not guarantee correctness of a generated implementation (RTL, C, etc.).
- Formal proofs and implementation are orthogonal layers: proofs verify the spec, implementation must be verified separately (functional sweep, simulation, testing).
- Example: 84 Coq proofs of φ-mathematics did not catch systematic RTL bugs in GF multipliers (out-of-bounds product width, wrong normalization bit).
- Rule of thumb: if there is a code generator or handwritten implementation, add a functional verification layer (sweep against exact reference) in addition to formal proofs.