mrwulf
2da0a79981
Round-3 verdicts: GPT-5.6 conditionally approves (blockers closed, one portability finding); the Claude reviewer's Socratic addendum produced F1*, the strongest finding of the series — deployed verify_consistency and mechanized ConsRec are NOT extensionally equal. Reproduced exactly (witness verify_consistency(1,3,R2,R3,P(2→3))=True vs ConsRec reject; 3,405 divergences n<60; strictly one-sided; power-of-two seeding mechanism confirmed in source). - KNOWN-GAPS gap 14: witness, mechanism, one-sidedness, and the pinned-pair side condition under which Theorem 3 transfers to the deployed verifier (pacta's pin-store flow supplies it by construction). No pacta code change; deployed behavior matches upstream RFC 9162 implementations. - fidelity: lied-size family — 73,573 boundary cases, 3,867 expected divergences PINNED, one-sided direction asserted per case. Banner rescoped: agreement over pinned families, not extensional equality. - Theorem3.lean: acceptCons_sound (F2) — soundness over the named acceptCons predicate, n₀=0 discharged from the non-prefix premise, size bound derived from acceptance via new consRec_some_le. Cones read from #print axioms; CONES/AxiomCheck/allowlist updated (218 → 222 constants, diff = the two theorems + two generated auxiliaries). - F3/GPT§7: verification/lean-toolchain pin + run_bare.sh (reviewer's standalone runner, plain public lean — verified green: 61 cones, 222 constants, gate green) + AENEAS_ENV override in check.sh and selftest_audit.sh. - F4: awk field-equality replaces regex-with-dots in Phase 3b. - F5: git-tracked .pyc removed (worse than reported — it was in the repo, not just the kit); __pycache__ gitignored; round-4 kit ships a corpus MANIFEST.sha256 + pinned commit (also GPT's governance condition). check.sh exit 0, ATTESTATION GREEN; selftest exit 0, 9/9 + control. Live LTL untouched (12 leaves, bcd15f9d…); attestation still gated on ePrint decision + author review + explicit operator order. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
5 KiB
Statement map: paper §6 ↔ Lean corpus
The kernel guarantees every proof below; what a reviewer must vet is the statements — that each Lean theorem says what the paper's item says. This map is the review surface. Paper = "The Lean Transparency Log" (https://ltl.zkdefi.org/paper), §6 and §10 (which scopes the mechanization to items i–v).
| paper item | Lean name | file | cone |
|---|---|---|---|
| §5.3 split point k (RFC 9162) | kbelow + kbelow_pos/lt, le_two_kbelow, kbelow_pow2 (2^j = k < n ≤ 2^{j+1} pins k uniquely) |
Basic | no hash axiom |
| §5.3 MTH | MTH |
Basic | sha256 |
| §5.3 Path | Path |
Completeness | sha256 |
| §5.3 Root (App. B) | Root (Option = rejection) |
Basic | sha256 |
| §5.3 inclusion accept | acceptIncl (= m < n ∧ Root … = some r); acceptIncl_complete, acceptIncl_sound route Thm 1/2 through it |
Basic, Completeness, Extract | sha256 (+choice) |
| Lemma 2 (general abstract form) | not mechanized as one theorem — proved as specializations (see KNOWN-GAPS gap 3); the row below and the Lemma-2 rows are those instances | — | — |
| §5.3 ConsRec | ConsRec (+ machine-checked base-refactor equivalences consRec_base_true_eq/false_eq) |
Basic, Refactor | sha256 |
| Lemma 1 (domain separation) | domsep |
Basic | axiom-free |
| Theorem 1 (inclusion completeness) | incl_complete |
Completeness | sha256 (+choice) |
| Lemma 2, width fact ("65-byte preimages") | Hash = length-32 subtype; hnode_preimage_inj |
gen, Basic | propext |
| Lemma 2, whole-tree instance | extractMTH + extractMTH_correct |
Descent | sha256 (+choice) |
| Lemma 2, ConsRec instance (Thm 3 steps 1–2) | consRecBinding |
Binding3 | sha256 (+choice) |
| Theorem 2 (inclusion soundness, explicit 𝓔) | extractIncl + extractIncl_correct |
Extract | sha256 (+choice) |
| Theorem 3 (consistency soundness, explicit 𝓔′) | extractCons + extractCons_correct; extractCons_correct_paper at the paper's exact quantifiers (n₀=0 discharged); acceptCons_sound routes it through the named acceptCons predicate (size bound derived from acceptance via consRec_some_le). Covers the MECHANIZED accept set; transfer to the deployed verifier is conditional on the pinned-pair side condition of gap 14 |
Theorem3 | sha256 (+choice) |
| Prop 1(1) (pin monotonicity + prefix) | pinAccept, pinAccept_monotone, pin_prefix_correct |
PinStore | sha256 (+choice) |
| Prop 1(2), Merkle share | fork_distinct (different roots ⇒ different content); transferability = signature layer, out of scope |
PinStore | sha256 |
| non-vacuity guards (anti-pigeonhole) | extractIncl_nonvacuous, extractMTH_nonvacuous, extractCons_nonvacuous, pin_prefix_nonvacuous |
Extract/Descent/Theorem3/PinStore | sha256 |
| definition fidelity vs deployed verifier | fidelity/ harness: MTH==merkle_root, Path==inclusion_proof, verifier agreement 230,271 inclusion + 230,016 consistency over the pinned case families — not extensional equality: the lied-size family (73,573 cases) pins the known one-sided divergence of gap 14 (3,867 expected, deployed-accepts-only, direction asserted) |
fidelity | (testing) |
Note on "assumption-free" (paper §10(i)): incl_complete's cone lists
LTLAcc.sha256, but the theorem assumes no property of it — it
merely mentions the opaque constant. Constant-dependence is not
property-assumption; the soundness theorems likewise carry sha256
without assuming collision resistance.
Design invariant of every soundness statement: the collision is the output
of a named extractor function and correctness is a claim about that
output. A bare ∃ x y, x ≠ y ∧ sha256 x = sha256 y is provable by
pigeonhole alone (sha256 maps an infinite domain into the finite 32-byte
type), so it carries no cryptographic content. What the guards certify
(precisely — round-2 M3): each named extractor does not return a
collision on at least one canonical honest input, which rules out the
degeneration where the conclusion is a globally inhabited bare collision
existential. They do NOT establish logical dependence on every listed
hypothesis, nor that no other classical argument could reach the
conclusion on some restricted domain.
Audit surface (enforced by verification/check.sh, exit 0 = green):
the FULL compiled environment of the corpus modules — 218 constants,
read from the Lean environment by Proofs/Inventory.lean (fully
qualified names, kinds, axiom cones) and pinned in
verification/inventory-allowlist.txt, diffed fail-closed both
directions on every run (round-3 replacement for the round-2 source-regex
gate, which GPT H1 showed was evadable). The 59 human-reviewed statement
cones above are additionally checked via #print axioms and
cross-checked against the inventory's independently computed cones.
verification/selftest_audit.sh attacks the gate with nine injection
cases (attributed/indented/private/instance declarations, a nested
namespace reusing an audited basename, a smuggled axiom, a deleted
declaration, and unmanifested Proofs/ and gen/ modules) — each must
fail the exact production gate.