Brings #82 up to today's `main` (1bbcc10) and resolves the two conflicts as ordered
unions: `package.json` `test`/`check:syntax` and the release builder's syntax allow-list
and evidence keys. Both sides only appended, so nothing was dropped.
Verifying the merge against main's RC attack suite (#40) found two real defects, each
fixed test-first:
- An oversized body destroyed the socket mid-upload, so the caller saw `fetch failed`
instead of the 413 it had already earned, and `security_attack_suite.mjs` could not
complete at all. New test in `tests/analyze-ingress.acceptance.test.js` failed with
`fetch failed` before the fix; `src/http-body.js` now keeps draining without buffering
and refuses on `end`, so the ceiling still holds and the refusal is readable.
- #16's per-caller budget turned the suite's own load phase into 8/8 429s, which would
have published refusals as analysis latency. The load phase now runs against a second
permissive target and the suite refuses to publish a load figure if any sample was
refused. `buildReport` also publishes `evidenceVerified`, so the verdict is
re-derivable from the report — asserted by #40's acceptance test, which only trips once
the verdict is GO.
Evidence on the merged tree: `npm test` 277/277, `test:security-attack` 8/8, the RC attack
suite returns VERDICT: GO with both P0 ingress cases passing (main is BLOCKED on exactly
those two), 14 browser acceptance suites green, and the release demo records end to end
(8:13, 97 captions) including the on-camera refusal scene.
Confidence was the only thing standing between a model answer and a
prefilled journal entry, so a confident reading of a blurry photo could
still be offered as fact. Quality is now part of the policy: `poor`
abstains however confident the reply looked, with its own fixed copy.
The confidence floor, accepted quality levels, and refusal copy move into
src/abstention-policy.js so the parser, the corpus, and the browser cannot
drift into three explanations of the same refusal. parseVisionResponse
takes an optional policy, which is how the tests prove the parser carries
no hidden limits of its own.
evaluatePrefillGate turns the #29 calibration report into one release
decision: GO plus a validation-selected threshold scored on the held-out
test split, or prefill is blocked with each reason named. A missing report
is a block, never a default allow. scripts/abstention_policy_receipt.mjs
exposes that as a receipt (exit 0 / 3 / 2).
The committed synthetic predictions currently report BLOCKED and the
receipt says so. No candidate exists until #31, and the builder does not
hard-gate on the receipt yet; the on-camera quality gate is the evidence
this release carries.
The corpus gains a case that replays a 0.94-confidence answer about the
synthetic motion-blur fixture, and the release demo asserts in the browser
that nothing is prefilled before captioning it.
Verified: 226 unit tests, calibration harness (19), corpus manifest (8),
vision-refusal, photo-first, and provisional acceptance, check:syntax,
check:diff, and a full real-browser demo recording whose caption receipt
confirms the new scene painted on screen.
The ROADMAP gates (macro-F1 >= 0.80, precision@coverage >= 0.90, ECE <= 0.05) were prose with no
tool behind them. scripts/calibration_report.py now computes them from prediction records and
refuses input it cannot honestly score:
- a split is re-derived from the pseudonymous subject key with the ingestion rule, so a corpus
cannot re-label its own splits; a contributor spanning two splits is refused before scoring
- the threshold is selected on validation only, and selecting on test or train is refused
- coverage, precision-at-coverage, macro-F1, per-class precision/recall, confusion, and expected
calibration error are computed on accepted suggestions, with classes that abstained away named
in excludedClasses instead of being scored as a silent zero
- exit codes carry the verdict: 0 GO, 3 BLOCKED, 2 refused
The release demo records the real harness against committed synthetic fixtures: a blocked
three-candidate comparison, a fixture engineered to clear every gate, and a leaked corpus refused.
Every number on the card is read back from the report the run wrote, and the captions state that
no model was measured, because a fixture is not evidence about accuracy.
The harness suite is wired into CI, check:syntax, and the release builder's own gates.
The declared data-URL MIME was a claim the boundary believed. Every image decision now comes from
the bytes: the container is sniffed, walked to its own end marker, refused when metadata segments
remain, and refused above a 4096px edge or 16MP. The request path adds a per-caller sliding-window
budget (429 + Retry-After) and a body deadline (408), so a flood, a stalled body, or a decode bomb
fails before a provider call.
Log hygiene is part of the contract: refusal copy is fixed prose, the bucket key is a salted hash
kept in memory only, and the acceptance test asserts no image bytes reach the process log.
The blank-first ordering let one blank field short-circuit the whole reply, so a
response carrying an impossible value came back as `unusable_visual_fields`:
the user was told the photo was unreadable when the model had in fact returned
something Timmy must never apply.
Blank fields still abstain and still prefill nothing. A filled-in field is a
claim, and a claim outside the contract now fails closed even when the other
field came back blank.
The corpus earns its label too. `out-of-range-fields` drops from confidence 1.4
to 0.9 so the case reaches the field-range check instead of dying at the
confidence gate, and a new `blank-plus-out-of-range` case covers the class the
old ordering let through. The release demo replays that case on camera and the
fail-closed 503 budget moves from one to two, so the scene cannot be dropped
silently.
Closes#73
The worker contract was only proven by one happy-path fixture and a handful of
inline assertions, so nothing continuously checked that a reply Timmy cannot use
is refused instead of guessed at.
Corpus: a CC0 synthetic corpus (stool, non-stool, blur, glare, empty bowl,
malformed response) generated deterministically by scripts/build_vision_corpus.py.
Every fixture records its sha256, provenance, licence, case class, the provider
reply it replays, and the outcome that reply must produce. loadVisionCorpus
refuses a fixture whose bytes no longer match the manifest, so the corpus cannot
be edited into a different test. No real medical imagery, camera, or third-party
content is involved.
Contract fixes the corpus exposed:
- A schema-valid reply with a null form and colour is now an abstention
(abstainedFor: unusable_visual_fields) instead of a hard provider failure. The
declared response schema permits those nulls, so treating the reply as an error
was wrong; it now abstains calmly and still prefill nothing.
- Fail-closed replies carry a machine-readable code (invalid_json, invalid_fields)
instead of prose matching, and visionFailureResponse() maps them to plain,
user-safe copy with no provider text, image bytes, or credentials echoed. A
running server now returns that copy for unparsable and out-of-range replies.
Evidence: unit corpus contract and hash integrity, corpus-wide conformance
through the real analysis path, the same corpus through a running server.mjs with
a stub worker (200 abstain vs 503 fail-closed, no leaked bytes), and a real
browser acceptance run for the refusal UI. The release demo replays three corpus
cases through the real code path so the video shows the status the server would
actually send, with the on-camera caption gated by the release builder.
timmy