fips205-source/tests/nist_acvp_vectors/extract_sha2_128s.py

174 lines
8.3 KiB
Python
Raw Normal View History

vectors: make the extraction re-derivable and the provenance claim literally true Round-7 review (both reviewers, independently) found that the ACVP provenance note said "the per-test private key `sk` dropped ... Nothing else altered" while the extraction had in fact also dropped `additionalRandomness` and `deferred` from all 42 tests, plus the top-level `isSample`. No field the tests consume was affected and no verdict changed — but the provenance block is the audit trail a third party diffs against, and as written it would have produced a false alarm or taught the next reviewer to wave differences through. Rather than reword the note, the transformation is now executable and pinned: tests/nist_acvp_vectors/extract_sha2_128s.py re-derives the file, verifies the upstream sha256 before doing anything, requires exactly 3 groups x 14 tests, removes exactly ONE field (`sk`) and asserts that invariant, carries every other per-test, group and top-level key through untouched, and writes canonical output. Run with no arguments it VERIFIES the committed file against a fresh extraction; --write regenerates. The file was regenerated with it, so "only `sk` removed" is now literally true and machine-checkable. Also from round 7, precision in the tests themselves: - exact assertions replace floors: `points == 108` (was >= 100), `with_ctx == 9` (was > 0), and prehash `3/4/7` (was `checked > 0`). With 7 of 14 prehash vectors skipped for unimplemented hash functions, a floor would have let real coverage fall from 3 to 1 while the total still summed to 14. - the randomized bridge's doc comment claimed corruption "across the WHOLE signature"; measured, the schedule hits 72 distinct positions in 11..=7779, never bytes 0-10 or 7780-7855. Corrected to the measured statement. Verified on stable Rust (rustc 1.95) as well as the pinned nightly: the bridge needs no nightly feature, so a third party can run all of it with cargo alone. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 11:01:13 +00:00
#!/usr/bin/env python3
"""Deterministically re-derive tests/nist_acvp_vectors/SLH-DSA-sigVer-FIPS205/
sha2_128s_extracted.json from the official NIST ACVP-Server vector set.
Why this file exists at all: the ACVP vector file vendored in this repository
contains no SLH-DSA-SHA2-128s sigVer group, so the parameter set the associated
verification work is about had no NIST known-answer verification coverage. The
three 128s groups are therefore taken from upstream.
Why this SCRIPT exists: external review (round 7) observed that a hand-made
extraction with a prose provenance note is not auditable the first version's
note said "only `sk` dropped" while three further fields had in fact been
removed. The transformation is now executable, pinned, and fails closed:
* the upstream file's sha256 must match SOURCE_SHA256 exactly;
* exactly EXPECTED_GROUPS groups must match the parameter set, each with
EXPECTED_TESTS_PER_GROUP tests;
* exactly one field, `sk`, is removed, and it must be present to be removed
(so a schema change is caught, not silently transformed);
* every other key is carried through untouched BY CONSTRUCTION, which a
reviewer verifies by reading `build()`, not by a self-check: no test inside a
transformer can detect a corrupted input, since the transformer is what
defines the output. The input is instead pinned by SOURCE_SHA256;
* verify mode re-derives and byte-compares the committed file, so a hand-edit
of the committed JSON IS caught;
* output is canonical (fixed indent, trailing newline).
NOT A GATE re-runnable EVIDENCE. Round-8 review noted the distinction: this
script needs network access, so nothing invokes it automatically (it is outside
`cargo test` and outside verification/check.sh). The committed JSON is still
trusted at review time; what this script provides is that a reviewer can
CHECK that trust cheaply and mechanically instead of taking a prose note's word.
A CI job running it in verify mode would close the remaining gap.
vectors: make the extraction re-derivable and the provenance claim literally true Round-7 review (both reviewers, independently) found that the ACVP provenance note said "the per-test private key `sk` dropped ... Nothing else altered" while the extraction had in fact also dropped `additionalRandomness` and `deferred` from all 42 tests, plus the top-level `isSample`. No field the tests consume was affected and no verdict changed — but the provenance block is the audit trail a third party diffs against, and as written it would have produced a false alarm or taught the next reviewer to wave differences through. Rather than reword the note, the transformation is now executable and pinned: tests/nist_acvp_vectors/extract_sha2_128s.py re-derives the file, verifies the upstream sha256 before doing anything, requires exactly 3 groups x 14 tests, removes exactly ONE field (`sk`) and asserts that invariant, carries every other per-test, group and top-level key through untouched, and writes canonical output. Run with no arguments it VERIFIES the committed file against a fresh extraction; --write regenerates. The file was regenerated with it, so "only `sk` removed" is now literally true and machine-checkable. Also from round 7, precision in the tests themselves: - exact assertions replace floors: `points == 108` (was >= 100), `with_ctx == 9` (was > 0), and prehash `3/4/7` (was `checked > 0`). With 7 of 14 prehash vectors skipped for unimplemented hash functions, a floor would have let real coverage fall from 3 to 1 while the total still summed to 14. - the randomized bridge's doc comment claimed corruption "across the WHOLE signature"; measured, the schedule hits 72 distinct positions in 11..=7779, never bytes 0-10 or 7780-7855. Corrected to the measured statement. Verified on stable Rust (rustc 1.95) as well as the pinned nightly: the bridge needs no nightly feature, so a third party can run all of it with cargo alone. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 11:01:13 +00:00
Usage:
python3 extract_sha2_128s.py # verify the committed file matches
python3 extract_sha2_128s.py --write # regenerate it
The upstream file is ~30 MB; it is fetched to a temporary path and not vendored.
"""
import argparse, hashlib, json, os, sys, tempfile, urllib.request
SOURCE_URL = (
"https://raw.githubusercontent.com/usnistgov/ACVP-Server/master/"
"gen-val/json-files/SLH-DSA-sigVer-FIPS205/internalProjection.json"
)
SOURCE_SHA256 = "a013fc2104f4ed4799d96d51141f65b965969b2cf10646626a021b6d456ce792"
PARAMETER_SET = "SLH-DSA-SHA2-128s"
EXPECTED_GROUPS = 3
EXPECTED_TESTS_PER_GROUP = 14
DROP_FIELDS = frozenset({"sk"}) # the ONLY per-test removal
OUT = os.path.join(os.path.dirname(os.path.abspath(__file__)),
"SLH-DSA-sigVer-FIPS205", "sha2_128s_extracted.json")
PROVENANCE = {
"what": f"{PARAMETER_SET} sigVer test groups, extracted verbatim from the "
"official NIST ACVP-Server vector set.",
"why": "The vector file vendored upstream in this directory contains NO "
f"{PARAMETER_SET} sigVer group (only 192s/256f/SHAKE variants), so the "
"one parameter set the verification work targets had zero NIST "
"known-answer verification coverage.",
"source_url": SOURCE_URL,
"source_sha256": SOURCE_SHA256,
"retrieved_utc": "2026-07-28",
"extraction": "Produced by tests/nist_acvp_vectors/extract_sha2_128s.py, which "
"verifies the upstream sha256, selects every testGroup whose "
f"parameterSet == {PARAMETER_SET}, and removes exactly ONE "
"per-test field: `sk` (the private key, not needed to verify a "
"signature). Every other per-test field and all group and "
"top-level metadata are carried through unchanged by construction. "
"The guarantees that can actually fail are: the pinned upstream "
"SOURCE_SHA256; the requirement that `sk` be present to be dropped; "
"the expected group and per-group test counts; and verify mode, which "
"re-derives and byte-compares this committed file.",
vectors: make the extraction re-derivable and the provenance claim literally true Round-7 review (both reviewers, independently) found that the ACVP provenance note said "the per-test private key `sk` dropped ... Nothing else altered" while the extraction had in fact also dropped `additionalRandomness` and `deferred` from all 42 tests, plus the top-level `isSample`. No field the tests consume was affected and no verdict changed — but the provenance block is the audit trail a third party diffs against, and as written it would have produced a false alarm or taught the next reviewer to wave differences through. Rather than reword the note, the transformation is now executable and pinned: tests/nist_acvp_vectors/extract_sha2_128s.py re-derives the file, verifies the upstream sha256 before doing anything, requires exactly 3 groups x 14 tests, removes exactly ONE field (`sk`) and asserts that invariant, carries every other per-test, group and top-level key through untouched, and writes canonical output. Run with no arguments it VERIFIES the committed file against a fresh extraction; --write regenerates. The file was regenerated with it, so "only `sk` removed" is now literally true and machine-checkable. Also from round 7, precision in the tests themselves: - exact assertions replace floors: `points == 108` (was >= 100), `with_ctx == 9` (was > 0), and prehash `3/4/7` (was `checked > 0`). With 7 of 14 prehash vectors skipped for unimplemented hash functions, a floor would have let real coverage fall from 3 to 1 while the total still summed to 14. - the randomized bridge's doc comment claimed corruption "across the WHOLE signature"; measured, the schedule hits 72 distinct positions in 11..=7779, never bytes 0-10 or 7780-7855. Corrected to the measured statement. Verified on stable Rust (rustc 1.95) as well as the pinned nightly: the bridge needs no nightly feature, so a third party can run all of it with cargo alone. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 11:01:13 +00:00
"note": "Test DATA only. Expected outcomes are NIST's `testPassed` field; "
"`reason` records why a negative case must be rejected.",
"regenerate": "python3 tests/nist_acvp_vectors/extract_sha2_128s.py --write",
}
def fetch() -> dict:
with tempfile.NamedTemporaryFile(delete=False, suffix=".json") as fh:
tmp = fh.name
try:
urllib.request.urlretrieve(SOURCE_URL, tmp)
raw = open(tmp, "rb").read()
finally:
os.unlink(tmp)
got = hashlib.sha256(raw).hexdigest()
if got != SOURCE_SHA256:
sys.exit(f"FATAL: upstream sha256 {got} != pinned {SOURCE_SHA256}.\n"
"The NIST file changed. Review the delta and update the pin "
"deliberately; do not regenerate blindly.")
return json.loads(raw)
def build(full: dict) -> dict:
groups = []
for g in full["testGroups"]:
if g.get("parameterSet") != PARAMETER_SET:
continue
tests = []
for t in g["tests"]:
for f in DROP_FIELDS:
if f not in t:
sys.exit(f"FATAL: tcId {t.get('tcId')} has no field {f!r} to drop")
# Fields are carried through BY CONSTRUCTION: `kept` is `t` minus
# DROP_FIELDS, so every retained key holds the identical object.
#
# Round-8 review found a tautological `assert` here — it compared
# `kept` against the comprehension that had just built it, so it could
# never fire. The first attempt to repair it (comparing kept[k] to
# t[k]) was tautological for the same reason, and that is the lesson
# worth recording: NO check inside this function can detect a
# corrupted input, because this function is what defines the output
# from that input. Faithfulness here is a property of the two lines
# below, which a reviewer reads; it is not something the script can
# test about itself.
#
# What actually protects the result, and can fail:
# * SOURCE_SHA256 — the input is pinned, so upstream cannot drift
# or be substituted without an explicit, reviewed pin change;
# * the `f not in t` presence check above — `sk` must exist to be
# dropped, so a schema change is caught rather than silently
# producing a different transformation;
# * EXPECTED_GROUPS / EXPECTED_TESTS_PER_GROUP below;
# * verify mode, which re-derives and byte-compares the committed
# file, so a hand-edit of the committed JSON is caught.
vectors: make the extraction re-derivable and the provenance claim literally true Round-7 review (both reviewers, independently) found that the ACVP provenance note said "the per-test private key `sk` dropped ... Nothing else altered" while the extraction had in fact also dropped `additionalRandomness` and `deferred` from all 42 tests, plus the top-level `isSample`. No field the tests consume was affected and no verdict changed — but the provenance block is the audit trail a third party diffs against, and as written it would have produced a false alarm or taught the next reviewer to wave differences through. Rather than reword the note, the transformation is now executable and pinned: tests/nist_acvp_vectors/extract_sha2_128s.py re-derives the file, verifies the upstream sha256 before doing anything, requires exactly 3 groups x 14 tests, removes exactly ONE field (`sk`) and asserts that invariant, carries every other per-test, group and top-level key through untouched, and writes canonical output. Run with no arguments it VERIFIES the committed file against a fresh extraction; --write regenerates. The file was regenerated with it, so "only `sk` removed" is now literally true and machine-checkable. Also from round 7, precision in the tests themselves: - exact assertions replace floors: `points == 108` (was >= 100), `with_ctx == 9` (was > 0), and prehash `3/4/7` (was `checked > 0`). With 7 of 14 prehash vectors skipped for unimplemented hash functions, a floor would have let real coverage fall from 3 to 1 while the total still summed to 14. - the randomized bridge's doc comment claimed corruption "across the WHOLE signature"; measured, the schedule hits 72 distinct positions in 11..=7779, never bytes 0-10 or 7780-7855. Corrected to the measured statement. Verified on stable Rust (rustc 1.95) as well as the pinned nightly: the bridge needs no nightly feature, so a third party can run all of it with cargo alone. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 11:01:13 +00:00
kept = {k: v for k, v in t.items() if k not in DROP_FIELDS}
tests.append(kept)
if len(tests) != EXPECTED_TESTS_PER_GROUP:
sys.exit(f"FATAL: group {g['tgId']} has {len(tests)} tests, "
f"expected {EXPECTED_TESTS_PER_GROUP}")
ng = {k: v for k, v in g.items() if k != "tests"}
ng["tests"] = tests
groups.append(ng)
if len(groups) != EXPECTED_GROUPS:
sys.exit(f"FATAL: found {len(groups)} {PARAMETER_SET} groups, expected {EXPECTED_GROUPS}")
out = {"_provenance": PROVENANCE}
for k, v in full.items(): # all top-level metadata, verbatim
if k != "testGroups":
out[k] = v
out["testGroups"] = groups
return out
def render(obj: dict) -> str:
return json.dumps(obj, indent=1) + "\n"
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--write", action="store_true", help="regenerate the committed file")
args = ap.parse_args()
text = render(build(fetch()))
if args.write:
open(OUT, "w").write(text)
print(f"wrote {OUT} ({len(text)} bytes)")
return 0
if not os.path.exists(OUT):
print(f"MISSING: {OUT}"); return 1
cur = open(OUT).read()
if cur == text:
print(f"OK: {OUT} matches a fresh extraction from the pinned upstream file")
return 0
print(f"MISMATCH: {OUT} differs from a fresh extraction — re-run with --write "
"and review the diff")
return 1
if __name__ == "__main__":
sys.exit(main())