dalek-ed25519-verified/verification/selftest-correspondence.sh

170 lines
8.4 KiB
Bash
Raw Normal View History

P2-c: classify and pin the extraction boundary Aeneas emits a *_Template.lean naming everything the extracted code needs from outside itself — the extraction's own statement of its boundary. extract.sh has always said, in prose, "after regenerating, diff the template against the hand-written file". Prose is not a gate, and the diff cannot be one: the two files legitimately differ in almost every line, holes and Aeneas comments against real definitions and modeling policy. MEASURING FIRST CHANGED WHAT THIS ITEM SHOULD BE. The TODO offered two options — enforce the diff, or pin both files — and the answer turned out to be neither. Both files were ALREADY byte-pinned by Phase 0b. And two further things stand here: the generated Funs.lean imports the model and CALLS these externals, so the Lean compiler enforces their TYPES wherever the extracted code uses them; and the per-certificate exact cones catch any external that becomes, or stops being, an assumption anything depends on. What none of those three sees is the CLASSIFICATION: for each name the extraction asks for, whether this repository answers with an ASSUMPTION or with a PROOF. That is the tier-A/B claim the documents make in prose — the curve calls and the three curve types resolve to proven definitions rather than axioms, because gen/CurveField/Funs.lean opens `namespace curve25519_dalek` and so defines the very names Aeneas asks for. Nothing checked it. A regeneration that renamed one, or a model that quietly answered one with an axiom instead, would have left the documents claiming a proof where the repository had an assumption. Phase 0d recomputes the classification with model-correspondence.py (namespace-aware, so a definition inside a namespace counts under its full name) and requires equality with the committed MODEL-CORRESPONDENCE.txt. UNRESOLVED — the extraction asking for something nothing here provides — is a hard failure. dalek 43 MODEL 8 PROVEN 3 EXTRA anza 38 MODEL 0 PROVEN 4 EXTRA (no CurveSig crate) risc0 36 MODEL 8 PROVEN 4 EXTRA betrusted 35 MODEL 8 PROVEN 4 EXTRA selftest-correspondence.sh, five cases, negative-tested by disabling the comparison. The case that matters is 2: a PROVEN external answered by an axiom instead. No name changes anywhere, every byte pin still matches, and it compiles, because the signature is unchanged — before Phase 0d nothing in the button could tell. Trap recorded for whoever extends it: case 3 first deleted the PROVEN rows, which was VACUOUS on anza, since anza has none — it removed nothing, the table still matched, and the case passed while testing nothing. It now deletes the first row whatever its verdict AND asserts the file changed. extract.sh now points at the gate instead of asking a human to look. Certified by a full sweep: both buttons, all four forks, purged trees, machine otherwise idle. 8/8 green.
2026-07-31 15:53:31 +00:00
#!/usr/bin/env bash
# ─────────────────────────────────────────────────────────────────────────────
# selftest-correspondence.sh — attacks check.sh Phase 0d.
#
# Phase 0d asserts HOW each external the extraction asks for is answered: with
# an assumption in the hand-written model, or with a proof already in the
# extracted corpus. The second class is the one the documents make a claim
# about — the curve calls and curve types are said to resolve to the proven
# model's own definitions rather than to axioms — and that claim was prose
# until this phase existed.
#
# 0 control: the committed table matches the files
# 1 the extraction asks for something NOTHING provides -> UNRESOLVED
# 2 a PROVEN external answered by an axiom in the model instead. This is the
# attack that matters: a proof silently downgraded to an assumption, in a
# name whose spelling does not change anywhere else.
# 3 a row deleted from the committed table -> drift
# 4 a row's verdict edited in the committed table -> drift
#
# No Lean: Phase 0d is pure text over gen/. Seconds, not minutes.
# ─────────────────────────────────────────────────────────────────────────────
set -uo pipefail
HERE="$(cd "$(dirname "$0")" && pwd)"
FAILURES=0
STASH="$(mktemp -d)"
cleanup() {
[ -f "$STASH/corr" ] && cp "$STASH/corr" "$HERE/MODEL-CORRESPONDENCE.txt"
[ -f "$STASH/model" ] && cp "$STASH/model" "$HERE/$MODEL_REL"
[ -f "$STASH/tmpl" ] && cp "$STASH/tmpl" "$HERE/$TMPL_REL"
rm -rf "$STASH"
}
# Derive the victims from this repo rather than naming them: the forks do not
# share a gen/ layout (anza has no CurveSig crate at all, so it has no PROVEN
# rows and case 2 does not apply there).
TMPL_REL=$(cd "$HERE" && ls gen/*/FunsExternal_Template.lean | head -1)
MODEL_REL="${TMPL_REL/_Template/}"
PROVEN_ROW=$(grep -m1 '|PROVEN$' "$HERE/MODEL-CORRESPONDENCE.txt" || true)
trap cleanup EXIT INT TERM
cp "$HERE/MODEL-CORRESPONDENCE.txt" "$STASH/corr"
cp "$HERE/$MODEL_REL" "$STASH/model"
cp "$HERE/$TMPL_REL" "$STASH/tmpl"
# Phase 0d lifted from the shipping button.
{ echo 'set -euo pipefail'
echo "HERE=\"$HERE\""
awk '/^# ── Phase 0d/{f=1} f&&/^# ── (Phase 1|Phases end)/{exit} f{print}' "$HERE/check.sh"
} > "$STASH/p0d.sh"
for want in 'Phase 0d' 'MODEL CORRESPONDENCE' 'model-correspondence.py'; do
grep -qF "$want" "$STASH/p0d.sh" || {
echo "FATAL: the lifted driver has no '$want' — check.sh's phase markers moved."; exit 1; }
done
expect() { # expect <label> <want-rc> <want-substring>
local label="$1" want_rc="$2" want_txt="$3" out rc
out=$(bash "$STASH/p0d.sh" 2>&1); rc=$?
if [ "$rc" -ne "$want_rc" ]; then
echo "$label: exit $rc, expected $want_rc"; tail -6 <<<"$out" | sed 's/^/ /'
FAILURES=$((FAILURES+1)); return
fi
if ! grep -qF "$want_txt" <<<"$out"; then
echo "$label: exit code right, diagnostic wrong (rejected for the wrong reason)"
echo " wanted: $want_txt"; tail -6 <<<"$out" | sed 's/^/ /'
FAILURES=$((FAILURES+1)); return
fi
echo "$label"
}
echo "=== selftest-correspondence: attacking check.sh Phase 0d ==="
echo " template: $TMPL_REL"
expect "control: the committed table matches the files" 0 "answered by PROVEN definitions"
# ── 1. the extraction asks for something nothing provides ──────────────────
printf '\naxiom selftest_unprovided_external : Nat\n' >> "$HERE/$TMPL_REL"
expect "case 1: an external nothing provides" 1 "MODEL CORRESPONDENCE FAILED"
cp "$STASH/tmpl" "$HERE/$TMPL_REL"
# ── 2. a proof silently downgraded to an assumption ────────────────────────
# Answer a PROVEN external with an axiom in the model. The name does not change
# anywhere; only the way it is answered does. Nothing else in the button sees
# this: the byte pins still match their files, the compiler is content because
# the signature is unchanged, and no certificate's cone moves unless something
# happens to depend on it.
if [ -n "$PROVEN_ROW" ]; then
PROVEN_NAME=$(cut -d'|' -f2 <<<"$PROVEN_ROW")
PROVEN_TMPL=$(cut -d'|' -f1 <<<"$PROVEN_ROW")
VICTIM_MODEL="gen/${PROVEN_TMPL}.lean"
cp "$HERE/$VICTIM_MODEL" "$STASH/model2"
printf '\naxiom %s : Nat\n' "$PROVEN_NAME" >> "$HERE/$VICTIM_MODEL"
expect "case 2: a PROVEN external downgraded to an assumption" 1 "MODEL CORRESPONDENCE DRIFT"
cp "$STASH/model2" "$HERE/$VICTIM_MODEL"
else
echo " · case 2 skipped: this fork's extraction has no PROVEN externals"
fi
# ── 3/4. the committed table itself ────────────────────────────────────────
# Delete the FIRST row, whatever its verdict. An earlier draft deleted the
# PROVEN rows, which was vacuous on anza — that fork's extraction has none, so
# nothing was removed, the table still matched, and the case passed by testing
# nothing. Pick a row every fork is guaranteed to have.
sed '0,/|/{/|/d}' "$STASH/corr" > "$HERE/MODEL-CORRESPONDENCE.txt"
if ! diff -q "$STASH/corr" "$HERE/MODEL-CORRESPONDENCE.txt" >/dev/null; then
expect "case 3: a row deleted from the committed table" 1 "MODEL CORRESPONDENCE DRIFT"
else
echo " ✗ case 3: the table was not actually modified — the case is vacuous"
FAILURES=$((FAILURES+1))
fi
cp "$STASH/corr" "$HERE/MODEL-CORRESPONDENCE.txt"
sed -i '0,/|MODEL$/s/|MODEL$/|PROVEN/' "$HERE/MODEL-CORRESPONDENCE.txt"
expect "case 4: a verdict edited in the committed table" 1 "MODEL CORRESPONDENCE DRIFT"
cp "$STASH/corr" "$HERE/MODEL-CORRESPONDENCE.txt"
Round-7 F1: make model/template correspondence SEMANTIC, and fail closed THE DEFECT, as found. GPT-5.6 showed the textual classifier could report PROVEN for a name Lean resolves to an axiom: a definition appearing only inside a `/- -/` comment was read as real. Repairing that exposed something larger and already live. Aeneas wraps long declarations: axiom curve25519_dalek.edwards.EdwardsPoint.Insts.CoreOpsArithNegEdwardsPoint.neg The old scanner required keyword and name on one physical line, so it matched nothing there and SILENTLY DROPPED the declaration — no MODEL row, no PROVEN row, no failure. Nine to ten externals per fork had no row at all, and the tier-A/B `neg` was missing from every committed table while the trust documents claimed that class was machine-checked. Four pinned tables were wrong, in four repositories, with the buttons green over them. No attacker was required; Aeneas's own formatting did it. The lesson is not "write a better regex". The scanner was FAIL-OPEN: input it could not parse produced silence instead of a stop. A gate that drops what it cannot read is worse than no gate, because the green covers a gap that is invisible in the diff. THE REPAIR, in three layers that each do only what they honestly can. · DISCOVERY stays textual, because the template cannot be imported — it declares the same names as the model and they would clash. It now strips NESTED `/- -/` blocks, reads a name that wraps to a later line, tracks namespace/section/end, and FAILS CLOSED: an unparseable keyword exits 2 naming file and line. · Phase 0d keeps the cheap pre-compile comparison against the table. · Phase 2d is new and authoritative. After compilation it generates a temp Lean driver (the Phase 2b idiom, so no permanent module and no manifest churn), imports every non-template gen module, and asks env.find? what each external IS: kind and originating module. Model module -> MODEL; non-axiom in a generated module -> PROVEN; an axiom outside the model, or a name that does not resolve, is a failure. Lean's verdict must equal the committed table's, and the answer count must equal the question count so a truncated resolver cannot pass. THE TABLES were regenerated and verified three ways before being committed: an independently written second scanner agreed on 228/228 rows across the four forks; all 47 changed rows carry file:line evidence; and the PROVEN row was put to Lean directly — kind = def, module = CurveField.Funs. dalek 54 -> 64 rows anza 42 -> 51 risc0 48 -> 57 betrusted 47 -> 56 Also corrected: `from_residual` was recorded EXTRA in every fork; the template does demand it, so it is MODEL. anza carried a spurious `hash|EXTRA` row from a mis-parse. NEGATIVE-TESTED. With Phase 0d neutered so Phase 2d was the gate under test, a PROVEN row edited to MODEL produces: SEMANTIC DRIFT: ...neg — table says MODEL, Lean says PROVEN selftest-correspondence.sh grew 4 cases -> 7: GPT's comment counterexample, the wrapped declaration that was actually live, and a fail-closed check. WHAT THIS DOES NOT CLOSE. Phase 2d resolves names in the environment built from gen/. It establishes what a name IS — assumption or proof, and where — not that the extraction faithfully represents the Rust. That remains the Charon/Aeneas trust assumption and no gate here can close it. Certified: both buttons, all four forks, purged trees. 8/8 green, with 62/48/53/52 externals resolved by Lean and every verdict matching.
2026-08-02 00:24:15 +00:00
# ── 5/6. THE ROUND-7 FINDINGS, so they cannot regress ──────────────────────
# Both were real. Case 5 is GPT-5.6's constructive counterexample: a definition
# that exists ONLY inside a block comment was read as a real declaration, so the
# scanner reported PROVEN for a name Lean resolves to an axiom. Case 6 is the
# one that was live in four committed tables: Aeneas wraps long declarations,
# the old scanner required keyword and name on one physical line, and so it
# SILENTLY DROPPED them — nine to ten externals per fork had no row at all.
#
# Case 6 is the more important of the two. A gate that drops what it cannot
# read is worse than no gate: it prints green across a gap that is invisible in
# the diff. The scanner must now FAIL rather than skip.
CX=$(mktemp -d)
mkdir -p "$CX/gen/Forged"
printf 'axiom Forged.value : Nat\n' > "$CX/gen/Forged/FunsExternal_Template.lean"
printf 'axiom\n Forged.value : Nat\n' > "$CX/gen/Forged/FunsExternal.lean"
printf '/-\nnamespace Forged\ndef value : Nat := 0\nend Forged\n-/\n' > "$CX/gen/Forged/Funs.lean"
OUT=$(python3 "$HERE/model-correspondence.py" "$CX" 2>&1)
if grep -q 'Forged.value|MODEL' <<<"$OUT"; then
echo " ✓ case 5: a definition inside a block comment is not read as a declaration"
else
echo " ✗ case 5: comment-only definition mis-read — scanner says:"; sed 's/^/ /' <<<"$OUT"
FAILURES=$((FAILURES+1))
fi
printf 'axiom\n Forged.wrapped\n :\n Nat\n' >> "$CX/gen/Forged/FunsExternal_Template.lean"
OUT=$(python3 "$HERE/model-correspondence.py" "$CX" 2>&1); RC=$?
if [ "$RC" -ne 0 ] && grep -q 'UNRESOLVED\|Forged.wrapped' <<<"$OUT"; then
echo " ✓ case 6: a declaration whose name wraps to the next line is SEEN, not dropped"
else
echo " ✗ case 6: wrapped declaration dropped or mis-handled (rc=$RC):"; sed 's/^/ /' <<<"$OUT"
FAILURES=$((FAILURES+1))
fi
printf 'axiom\n' > "$CX/gen/Forged/FunsExternal_Template.lean"
OUT=$(python3 "$HERE/model-correspondence.py" "$CX" 2>&1); RC=$?
if [ "$RC" -eq 2 ] && grep -q 'fails closed' <<<"$OUT"; then
echo " ✓ case 7: an unparseable declaration stops the scanner (exit 2), never silence"
else
echo " ✗ case 7: unparseable declaration did not fail closed (rc=$RC)"
FAILURES=$((FAILURES+1))
fi
rm -rf "$CX"
P2-c: classify and pin the extraction boundary Aeneas emits a *_Template.lean naming everything the extracted code needs from outside itself — the extraction's own statement of its boundary. extract.sh has always said, in prose, "after regenerating, diff the template against the hand-written file". Prose is not a gate, and the diff cannot be one: the two files legitimately differ in almost every line, holes and Aeneas comments against real definitions and modeling policy. MEASURING FIRST CHANGED WHAT THIS ITEM SHOULD BE. The TODO offered two options — enforce the diff, or pin both files — and the answer turned out to be neither. Both files were ALREADY byte-pinned by Phase 0b. And two further things stand here: the generated Funs.lean imports the model and CALLS these externals, so the Lean compiler enforces their TYPES wherever the extracted code uses them; and the per-certificate exact cones catch any external that becomes, or stops being, an assumption anything depends on. What none of those three sees is the CLASSIFICATION: for each name the extraction asks for, whether this repository answers with an ASSUMPTION or with a PROOF. That is the tier-A/B claim the documents make in prose — the curve calls and the three curve types resolve to proven definitions rather than axioms, because gen/CurveField/Funs.lean opens `namespace curve25519_dalek` and so defines the very names Aeneas asks for. Nothing checked it. A regeneration that renamed one, or a model that quietly answered one with an axiom instead, would have left the documents claiming a proof where the repository had an assumption. Phase 0d recomputes the classification with model-correspondence.py (namespace-aware, so a definition inside a namespace counts under its full name) and requires equality with the committed MODEL-CORRESPONDENCE.txt. UNRESOLVED — the extraction asking for something nothing here provides — is a hard failure. dalek 43 MODEL 8 PROVEN 3 EXTRA anza 38 MODEL 0 PROVEN 4 EXTRA (no CurveSig crate) risc0 36 MODEL 8 PROVEN 4 EXTRA betrusted 35 MODEL 8 PROVEN 4 EXTRA selftest-correspondence.sh, five cases, negative-tested by disabling the comparison. The case that matters is 2: a PROVEN external answered by an axiom instead. No name changes anywhere, every byte pin still matches, and it compiles, because the signature is unchanged — before Phase 0d nothing in the button could tell. Trap recorded for whoever extends it: case 3 first deleted the PROVEN rows, which was VACUOUS on anza, since anza has none — it removed nothing, the table still matched, and the case passed while testing nothing. It now deletes the first row whatever its verdict AND asserts the file changed. extract.sh now points at the gate instead of asking a human to look. Certified by a full sweep: both buttons, all four forks, purged trees, machine otherwise idle. 8/8 green.
2026-07-31 15:53:31 +00:00
expect "restored: the table matches again" 0 "answered by PROVEN definitions"
echo ""
if [ "$FAILURES" -eq 0 ]; then
echo "SELFTEST PASSED — an external cannot change how it is answered, and a"
echo "proof cannot be downgraded to an assumption, without failing the button."
exit 0
fi
echo "SELFTEST FAILED: $FAILURES case(s) did not behave as claimed."
exit 1