fips205-slhdsa-verified/verification/check-selftest.sh

156 lines
9.1 KiB
Bash
Raw Normal View History

post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
#!/usr/bin/env bash
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
# Adversarial self-test of the check.sh gates (the R3-5 tradition: an audit that
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
# cannot fail is theater). The audit runs INSIDE Lean (Proofs/Audit.lean); the
# model bytes are pinned in Phase 0; the certificate SET and each STATEMENT are
# bound by a manifest fingerprint. Every attack MUST make check.sh fail, via the
# intended gate:
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
#
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
# 1. DEAD FILE — a stray Proofs/*.lean not in the manifest.
# 2. SMUGGLED AXIOM — a listed cert whose real cone contains a disallowed
# axiom; exact-equality reports extra=[...].
# 3. DROPPED ORACLE — a listed cert whose expected cone claims an oracle its
# proof does not use; exact-equality reports missing=[...]
# (a subset checker would pass this).
# 4. VANISHED CERT — a manifest name that no longer resolves; the explicit
# existence check reports NOT FOUND (collectAxioms would
# otherwise return [] and pass).
# 5. UN-AUDITED FALSE — round-4 F1: an UN-manifested `theorem … : False := …`
# added to a certificate module. The full-module
# enumeration must flag its disallowed cone. (This is the
# attack that made the pre-round-4 button green over a
# repo proving False.)
# 6. GUTTED STATEMENT — round-4 F2: a certificate's statement replaced by a
# tautology of the SAME cone. The statement fingerprint
# must differ from the committed manifest value.
# 7. MODEL TAMPER — round-4 F3: a hand-edit of a gen/ model file. Phase 0
# must reject it before anything is compiled.
# 8. DELETED ROW — round-4 F1 (set half): a manifest row removed. Audit
# still compiles, but the printed MANIFEST-FINGERPRINT
# changes and check.sh's committed binding must reject it.
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
#
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
# Green here means the gate genuinely rejects all eight. Self-cleaning: every
# file a step tampers is saved and restored around that step.
set -uo pipefail
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
HERE="$(cd "$(dirname "$0")" && pwd)"
cd "$HERE"
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
TOUCHED=()
save() { for f in "$@"; do cp -f "$f" "$f.sfbak"; TOUCHED+=("$f"); done; }
restore() { for f in "${TOUCHED[@]}"; do [ -f "$f.sfbak" ] && mv -f "$f.sfbak" "$f"; done; TOUCHED=(); }
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
cleanup() {
restore
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
rm -f Proofs/Stray.lean Proofs/Stray.olean Proofs/EvilSpec.lean Proofs/EvilSpec.olean \
Proofs/Audit.olean Proofs/ChainSpec.olean *.sfbak Proofs/*.sfbak gen/SlhVerify/*.sfbak
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
}
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
trap cleanup EXIT
echo "check-selftest: attacking the gates"
echo "===================================="
# ── Attack 1: dead file ─────────────────────────────────────────────────────
echo "-- stray" > Proofs/Stray.lean
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
if ./check.sh > /tmp/sf1.out 2>&1; then echo "✗ ATTACK 1 SUCCEEDED (dead file stayed green)"; exit 1; fi
grep -q "DEAD FILE" /tmp/sf1.out || { echo "✗ ATTACK 1: failed, not via dead-file gate"; cat /tmp/sf1.out; exit 1; }
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
rm -f Proofs/Stray.lean Proofs/Stray.olean
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
echo "✓ attack 1 rejected (dead-file gate)"
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
# ── Attack 2: smuggled disallowed axiom in a real cone ──────────────────────
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
save check.sh Proofs/Audit.lean
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
cat > Proofs/EvilSpec.lean <<'EOF'
import Proofs.ChainSpec
axiom evil_ax : True
theorem evil_thm : True := evil_ax
EOF
python3 - <<'PY'
s = open("check.sh").read()
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
assert 'PROOFS=(\n' in s
open("check.sh","w").write(s.replace('PROOFS=(\n', 'PROOFS=(\n "EvilSpec"\n', 1))
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
a = open("Proofs/Audit.lean").read()
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
assert 'import Proofs.ApexSpec' in a and ' [ (' in a
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
a = a.replace('import Proofs.ApexSpec', 'import Proofs.ApexSpec\nimport Proofs.EvilSpec', 1)
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
a = a.replace(' [ (', ' [ (`evil_thm, kernel3, 0),\n (', 1)
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
open("Proofs/Audit.lean","w").write(a)
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
PY
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
if ./check.sh > /tmp/sf2.out 2>&1; then echo "✗ ATTACK 2 SUCCEEDED (smuggled axiom)"; exit 1; fi
grep -q "evil_ax" /tmp/sf2.out || { echo "✗ ATTACK 2: rejected but evil_ax not named"; cat /tmp/sf2.out; exit 1; }
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
restore; rm -f Proofs/EvilSpec.lean Proofs/EvilSpec.olean Proofs/Audit.olean
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
echo "✓ attack 2 rejected (extra-axiom detection — evil_ax named)"
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
# ── Attack 3: dropped-oracle (subset would pass; exact must not) ─────────────
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
save Proofs/Audit.lean
review round 1: fix the fail-open audit gate + remove the overclaimed framing External review (both standing reviewers, 2026-07-24) returned DO NOT ATTEST. The eleven Lean theorems compile with genuinely clean cones (both reviewers independently reconstructed them), but two real defects were found and are fixed here. FIX 1 — the axiom audit was FAIL-OPEN (the critical blocker). check.sh Phase 3 grepped a single physical line of each `#print axioms` report; Lean WRAPS long cones across lines, so for ht/fors_outer/APEX the audit checked only `[propext,` and silently ignored the continuation lines — a disallowed axiom on line 2+ passed (the GPT reviewer demonstrated `review_evil_ax` passing). Since check.sh is the sole source of the word "proven", this is unacceptable. - New parser: FLATTEN the whole report (join newlines) BEFORE parsing, then extract each certificate's complete bracketed cone with a literal-string (regex-safe) scan and subset-check every axiom. Missing/empty report => FAIL CLOSED. The audit now prints the count of axioms actually audited per cert (apex: 8, previously 1). - check-selftest.sh gains ATTACK 3: a smuggled axiom bundled with the apex so its cone WRAPS with the evil axiom on a continuation line — the exact exploit. Verified: all three attacks now rejected, attack 3 via the axiom gate naming the continuation-line axiom. (Also fixed attack 2's leftover EvilSpec.lean tripping attack 3's dead-file gate.) FIX 2 — remove the overclaimed framing (refuted by both reviewers). Corrected in README, the ApexSpec header + apex docstring, and (separately) the control MANIFEST: - "composes all ten loop-fidelity certificates" — FALSE. The apex proof is a STRUCTURAL FACTORIZATION; it references NONE of the ten (grep: 0) and would remain provable if one were deleted. They are independent local-fidelity lemmas, not links in the apex proof. - "every loop is individually fidelity-certified" — FALSE. base_2b's inner accumulation loop is threaded opaquely and uncertified — and it determines the FORS indices / WOTS digits, so a defect there could change the recomputed root while all eleven theorems still hold. - "the deployed verifier" — the proved subject is verify_mono, a private #![allow(dead_code)] monomorphic facade NOT called by the public API; the bridge to the deployed generic verifier is the finite differential test, not a machine-checked refinement. - "verify-path pyramid complete" — replaced with "intermediate verification layer"; the apex is an ACCEPTANCE CHARACTERIZATION, not closed-form FIPS-205 correctness. Also: FunsExternal header noted the Take axiom "remains" (stale — deleted in de-plumbing round 2); corrected. check.sh green over all eleven certificates under the fixed fail-closed parser (exit 0, 8 axioms audited for the apex). Nothing about the theorems changed — they were and are sound; only the audit tool and the claims about them are fixed. NOT DONE (remaining reviewer blockers, tracked): reproducible extract tuple (pin commits, de-hard-code extract.sh) + Cargo.lock / toolchain pin. Attestation remains gated behind review round 2 + the operator halt + the appeal. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-24 14:55:10 +00:00
python3 - <<'PY'
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
import re
a = open("Proofs/Audit.lean").read()
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
new, n = re.subn(r'(`fips205\.to_int_loop_eq,\s*)kernel3(,\s*\d+\))', r'\1kernel3 ++ [oracleF]\2', a)
assert n == 1, f"patched {n}"
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
open("Proofs/Audit.lean","w").write(new)
PY
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
if ./check.sh > /tmp/sf3.out 2>&1; then echo "✗ ATTACK 3 SUCCEEDED (dropped oracle, subset hole)"; exit 1; fi
grep -q "missing=\[verify_mono.oracle.f\]" /tmp/sf3.out || { echo "✗ ATTACK 3: rejected but missing oracle not named"; cat /tmp/sf3.out; exit 1; }
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
restore; rm -f Proofs/Audit.olean
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
echo "✓ attack 3 rejected (missing-oracle detection — exact cone, not subset)"
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
# ── Attack 4: vanished certificate ──────────────────────────────────────────
save Proofs/Audit.lean
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
python3 - <<'PY'
a = open("Proofs/Audit.lean").read()
assert a.count('`fips205.chain_free_loop_eq') >= 1
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
open("Proofs/Audit.lean","w").write(a.replace('`fips205.chain_free_loop_eq,', '`fips205.chain_free_loop_eq_VANISHED,', 1))
PY
if ./check.sh > /tmp/sf4.out 2>&1; then echo "✗ ATTACK 4 SUCCEEDED (vanished cert)"; exit 1; fi
grep -q "NOT FOUND" /tmp/sf4.out || { echo "✗ ATTACK 4: rejected but not via existence check"; cat /tmp/sf4.out; exit 1; }
restore; rm -f Proofs/Audit.olean
echo "✓ attack 4 rejected (existence check — vanished cert cannot pass as 0-axiom)"
# ── Attack 5: un-manifested theorem proving False (round-4 F1, the big one) ──
save Proofs/ChainSpec.lean
printf '\n-- SELFTEST ATTACK 5 (round-4 F1): an un-manifested cert proving False.\naxiom cheat : ∀ (P : Prop), P\ntheorem repo_proves_false : False := cheat _\n' >> Proofs/ChainSpec.lean
if ./check.sh > /tmp/sf5.out 2>&1; then echo "✗ ATTACK 5 SUCCEEDED: check.sh GREEN over a repo proving False!"; exit 1; fi
grep -q "repo_proves_false" /tmp/sf5.out || { echo "✗ ATTACK 5: rejected but not via the module enumeration"; cat /tmp/sf5.out; exit 1; }
restore; rm -f Proofs/ChainSpec.olean Proofs/Audit.olean
echo "✓ attack 5 rejected (module enumeration — an un-manifested False theorem cannot pass)"
# ── Attack 6: gutted statement, cone preserved (round-4 F2) ─────────────────
# Gut a LEAF certificate (to_byte_loop_eq — nothing depends on it, so Phase 2
# still compiles and the fingerprint gate is what must bite). Replace its
# statement+proof with a kernel-3 tautology: same cone, different type.
save Proofs/InputPrepSpec.lean
python3 - <<'PY'
s = open("Proofs/InputPrepSpec.lean").read()
i = s.index("theorem to_byte_loop_eq (n : Std.U32) (k : Nat) :")
j = s.index("theorem hbody_cs", i) # the NEXT declaration (to_byte_loop_eq is a leaf)
gut = "theorem to_byte_loop_eq : (∀ p : Prop, p ¬p) := Classical.em\n\n"
open("Proofs/InputPrepSpec.lean","w").write(s[:i] + gut + s[j:])
PY
if ./check.sh > /tmp/sf6.out 2>&1; then echo "✗ ATTACK 6 SUCCEEDED: gutted statement passed (cone preserved)"; exit 1; fi
grep -qi "fingerprint" /tmp/sf6.out || { echo "✗ ATTACK 6: rejected but not via the statement fingerprint"; cat /tmp/sf6.out; exit 1; }
restore; rm -f Proofs/InputPrepSpec.olean Proofs/Audit.olean
echo "✓ attack 6 rejected (statement fingerprint — a gutted statement of the same cone cannot pass)"
# ── Attack 7: hand-edited model file (round-4 F3) ───────────────────────────
save gen/SlhVerify/Funs.lean
printf '\n-- SELFTEST ATTACK 7 (round-4 F3): hand-edited model.\n' >> gen/SlhVerify/Funs.lean
if ./check.sh > /tmp/sf7.out 2>&1; then echo "✗ ATTACK 7 SUCCEEDED: hand-edited model passed"; exit 1; fi
grep -q "MODEL INTEGRITY FAILED" /tmp/sf7.out || { echo "✗ ATTACK 7: rejected but not via Phase 0"; cat /tmp/sf7.out; exit 1; }
restore
echo "✓ attack 7 rejected (Phase 0 model-byte integrity — a hand-edited model cannot compile)"
# ── Attack 8: deleted manifest row (round-4 F1 set half) ────────────────────
save Proofs/Audit.lean
python3 - <<'PY'
import re
a = open("Proofs/Audit.lean").read()
new, n = re.subn(r'\n *\(`fips205\.to_int_loop_eq,[^\n]*\),', '', a)
assert n == 1, f"removed {n}"
open("Proofs/Audit.lean","w").write(new)
review round 1: fix the fail-open audit gate + remove the overclaimed framing External review (both standing reviewers, 2026-07-24) returned DO NOT ATTEST. The eleven Lean theorems compile with genuinely clean cones (both reviewers independently reconstructed them), but two real defects were found and are fixed here. FIX 1 — the axiom audit was FAIL-OPEN (the critical blocker). check.sh Phase 3 grepped a single physical line of each `#print axioms` report; Lean WRAPS long cones across lines, so for ht/fors_outer/APEX the audit checked only `[propext,` and silently ignored the continuation lines — a disallowed axiom on line 2+ passed (the GPT reviewer demonstrated `review_evil_ax` passing). Since check.sh is the sole source of the word "proven", this is unacceptable. - New parser: FLATTEN the whole report (join newlines) BEFORE parsing, then extract each certificate's complete bracketed cone with a literal-string (regex-safe) scan and subset-check every axiom. Missing/empty report => FAIL CLOSED. The audit now prints the count of axioms actually audited per cert (apex: 8, previously 1). - check-selftest.sh gains ATTACK 3: a smuggled axiom bundled with the apex so its cone WRAPS with the evil axiom on a continuation line — the exact exploit. Verified: all three attacks now rejected, attack 3 via the axiom gate naming the continuation-line axiom. (Also fixed attack 2's leftover EvilSpec.lean tripping attack 3's dead-file gate.) FIX 2 — remove the overclaimed framing (refuted by both reviewers). Corrected in README, the ApexSpec header + apex docstring, and (separately) the control MANIFEST: - "composes all ten loop-fidelity certificates" — FALSE. The apex proof is a STRUCTURAL FACTORIZATION; it references NONE of the ten (grep: 0) and would remain provable if one were deleted. They are independent local-fidelity lemmas, not links in the apex proof. - "every loop is individually fidelity-certified" — FALSE. base_2b's inner accumulation loop is threaded opaquely and uncertified — and it determines the FORS indices / WOTS digits, so a defect there could change the recomputed root while all eleven theorems still hold. - "the deployed verifier" — the proved subject is verify_mono, a private #![allow(dead_code)] monomorphic facade NOT called by the public API; the bridge to the deployed generic verifier is the finite differential test, not a machine-checked refinement. - "verify-path pyramid complete" — replaced with "intermediate verification layer"; the apex is an ACCEPTANCE CHARACTERIZATION, not closed-form FIPS-205 correctness. Also: FunsExternal header noted the Take axiom "remains" (stale — deleted in de-plumbing round 2); corrected. check.sh green over all eleven certificates under the fixed fail-closed parser (exit 0, 8 axioms audited for the apex). Nothing about the theorems changed — they were and are sound; only the audit tool and the claims about them are fixed. NOT DONE (remaining reviewer blockers, tracked): reproducible extract tuple (pin commits, de-hard-code extract.sh) + Cargo.lock / toolchain pin. Attestation remains gated behind review round 2 + the operator halt + the appeal. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-24 14:55:10 +00:00
PY
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
if ./check.sh > /tmp/sf8.out 2>&1; then echo "✗ ATTACK 8 SUCCEEDED: a deleted manifest row passed (set unbound)"; exit 1; fi
grep -q "manifest fingerprint mismatch" /tmp/sf8.out || { echo "✗ ATTACK 8: rejected but not via the manifest-fingerprint binding"; cat /tmp/sf8.out; exit 1; }
review round 2: in-Lean exact-cone audit + reproducibility + doc honesty Addresses the round-2 reviewer punch-list. No theorem statement, proof term, or fold definition changed; the eleven cones are unchanged (independent collectAxioms dump in verification/RECORDED-RUN.md). AUDIT GATE (both reviewers, the critical one) - Retire the bash #print-axioms text parser (fail-open on empty/truncated reports, and only a SUBSET check). Replace with verification/Proofs/Audit.lean: reads each certificate's cone from the kernel via collectAxioms and asserts EXACT set equality against its expected boundary. Extra axiom, dropped oracle, renamed/deleted cert, or an axiom/opaque sham each throw -> non-zero Lean exit. No text to misparse; nothing fails open. check.sh Phase 3 now just compiles it (and still requires the explicit PASSED line). - check-selftest.sh rewritten to attack the new gate: dead-file, smuggled extra axiom (named), dropped-oracle (subset would pass, exact must not), and a vanished certificate (the collectAxioms-returns-[] trap). All four rejected. REPRODUCIBILITY (GPT B1.4 / B1.5) - extract.sh refuses a wrong-commit or dirty source tree (fail-closed), takes an optional source-path arg, and pins the source commit. - verification/PROVENANCE.json: single machine-readable pin set (source + charon + aeneas commits/channel + lean + ocaml) with generated-file sha256. - Re-running extract.sh reproduces gen/SlhVerify/{Types,Funs}.lean byte-identically (companion fips205-source commit adds Cargo.lock + rust-toolchain.toml; verified not to perturb the model). DOC HONESTY (both reviewers) - README: fix the self-contradiction (apex "not yet proven" trailer vs the proven apex), the false "oracles kept OUTSIDE every cone" (they are INSIDE, by design), "deployed monomorphic path" and "semantics-identical for every parameter set" overclaims, "only two lines changed", stale snapshot head; retitle the stale future-tense "what will be claimed" section. - TRUSTED-BASE: drop "nothing proven yet"; add base_2b-inner and deployment- bridge non-claims explicitly; current pin. - ChainSpec header: "deployed monomorphic path" -> private verify_mono facade (comment only). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-24 17:13:55 +00:00
restore; rm -f Proofs/Audit.olean
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
echo "✓ attack 8 rejected (manifest fingerprint — a silently-dropped cert cannot pass)"
review round 1: fix the fail-open audit gate + remove the overclaimed framing External review (both standing reviewers, 2026-07-24) returned DO NOT ATTEST. The eleven Lean theorems compile with genuinely clean cones (both reviewers independently reconstructed them), but two real defects were found and are fixed here. FIX 1 — the axiom audit was FAIL-OPEN (the critical blocker). check.sh Phase 3 grepped a single physical line of each `#print axioms` report; Lean WRAPS long cones across lines, so for ht/fors_outer/APEX the audit checked only `[propext,` and silently ignored the continuation lines — a disallowed axiom on line 2+ passed (the GPT reviewer demonstrated `review_evil_ax` passing). Since check.sh is the sole source of the word "proven", this is unacceptable. - New parser: FLATTEN the whole report (join newlines) BEFORE parsing, then extract each certificate's complete bracketed cone with a literal-string (regex-safe) scan and subset-check every axiom. Missing/empty report => FAIL CLOSED. The audit now prints the count of axioms actually audited per cert (apex: 8, previously 1). - check-selftest.sh gains ATTACK 3: a smuggled axiom bundled with the apex so its cone WRAPS with the evil axiom on a continuation line — the exact exploit. Verified: all three attacks now rejected, attack 3 via the axiom gate naming the continuation-line axiom. (Also fixed attack 2's leftover EvilSpec.lean tripping attack 3's dead-file gate.) FIX 2 — remove the overclaimed framing (refuted by both reviewers). Corrected in README, the ApexSpec header + apex docstring, and (separately) the control MANIFEST: - "composes all ten loop-fidelity certificates" — FALSE. The apex proof is a STRUCTURAL FACTORIZATION; it references NONE of the ten (grep: 0) and would remain provable if one were deleted. They are independent local-fidelity lemmas, not links in the apex proof. - "every loop is individually fidelity-certified" — FALSE. base_2b's inner accumulation loop is threaded opaquely and uncertified — and it determines the FORS indices / WOTS digits, so a defect there could change the recomputed root while all eleven theorems still hold. - "the deployed verifier" — the proved subject is verify_mono, a private #![allow(dead_code)] monomorphic facade NOT called by the public API; the bridge to the deployed generic verifier is the finite differential test, not a machine-checked refinement. - "verify-path pyramid complete" — replaced with "intermediate verification layer"; the apex is an ACCEPTANCE CHARACTERIZATION, not closed-form FIPS-205 correctness. Also: FunsExternal header noted the Take axiom "remains" (stale — deleted in de-plumbing round 2); corrected. check.sh green over all eleven certificates under the fixed fail-closed parser (exit 0, 8 axioms audited for the apex). Nothing about the theorems changed — they were and are sound; only the audit tool and the claims about them are fixed. NOT DONE (remaining reviewer blockers, tracked): reproducible extract tuple (pin commits, de-hard-code extract.sh) + Cargo.lock / toolchain pin. Attestation remains gated behind review round 2 + the operator halt + the appeal. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-24 14:55:10 +00:00
post-flip drill over the chain certificate: HELD; audit gates now self-tested The window under audit claimed the campaign's first certificate, so this drill was maximally adversarial. Everything of substance HELD: - three-way model fidelity EXACT: extracted chain_free_loop.body == chainFoldN step == the Rust origin, operation-for-operation including address threading - button green fresh; axiom sweep over ALL 8 declarations minimal (pure lemmas = kernel-3; oracle-touching = kernel-3 + oracle.f only) - non-vacuity PROVEN: the concrete 1-step consequence (one address-set + one hash call) derives from the certificate by rfl - commit body of cfd50bb intact (the one flagged fragment was a bad drill grep pattern, not an artifact); worktree clean; heads synced NEW, from the drill (R3-5 tradition): verification/check-selftest.sh - permanent adversarial self-test of the check.sh gates. Attack 1 (dead Proofs file) and attack 2 (certificate with a smuggled axiom) must both make check.sh fail; both verified rejected, selftest green, self-cleaning. An audit that cannot fail is theater; this one demonstrably can. Two notes for the record: (a) bind_congr is the generic Bind-class congruence from core/Mathlib, not Aeneas.Std.Primitives (memory corrected); (b) the certificate covers chain_free_loop - the thin chain_free wrapper (bound computation + massert + clone) gets its trivial composition lemma in the wots layer, where it is consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 12:34:42 +00:00
echo
review round 4: bind the cert set, statements, and model bytes (F1/F2/F3) The third reviewer demonstrated that the round-2 in-Lean exact-cone audit, though sound for LISTED certs, left three fail-opens OUTSIDE the cone check — and made check.sh print ALL GREEN over a repo proving False. All closed; no theorem, proof, or fold changed (the 11 cones are unchanged). F1 — the audited SET was unbound. Audit.lean now (a) enumerates EVERY theorem defined in the eight certificate modules and requires each cone ⊆ boundary, so an un-manifested `theorem _ : False := cheat _` fails regardless of naming (this is the exact exploit the reviewer used); and (b) prints a MANIFEST fingerprint over the whole committed manifest, which check.sh binds to — so deleting/swapping a cert row fails outside Lean too. F2 — only cones were bound, not statements. Each cert now also carries the structural fingerprint (Expr.hash) of its elaborated type; a statement gutted to a tautology of the same cone changes the fingerprint and fails. F3 — the gen/ model bytes were unbound. New check.sh Phase 0 sha256-pins all four gen/SlhVerify/*.lean (incl. the two hand-maintained *External files, now hashed in PROVENANCE.json) BEFORE compiling; a hand-edited model fails first. F4/F5 — docs. README cone diagram now roots honestly at slh_verify_internal and states the pure/prehash domain-separator byte, the ctx>255 check, M' assembly, and deserialization are ABOVE the root and uncovered (new TRUSTED-BASE item 10). The false "rules out a wrong ADRS field" claim is corrected in README + ChainSpec (a transliteration makes the field visible, not excluded). check-selftest.sh: eight attacks, all rejected (dead file; extra axiom; dropped oracle; vanished cert; un-manifested False theorem; gutted statement; hand-edited model; deleted manifest row). Full transcript + green check.sh in verification/RECORDED-RUN.md. Standing limit unchanged and disclosed: an audit cannot defend against an author who edits the manifest AND check.sh AND the proofs together; the consumer defense is the pinned commit reviewed at the pin. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 17:47:39 +00:00
echo "SELFTEST GREEN: the gate rejects dead files, extra axioms, dropped oracles,"
echo "vanished certs, un-manifested False theorems, gutted statements, hand-edited"
echo "models, and deleted manifest rows."