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.
Add an executable abuse suite that drives the real server with synthetic hostile bytes while a
counting stub worker records every call that reaches the model boundary, plus the verdict rule that
turns those observations into a release decision.
- src/release-security-gate.js: nearest-rank latency percentiles, per-case pass/fail derivation that
names the field that broke, and the rule that any failing P0 case blocks RC approval while a
required class with no verified evidence blocks it too. Reports are redacted and fail closed.
- scripts/security_attack_suite.mjs: oversized body, HTML polyglot as JPEG, undecodable bytes,
missing consent, non-conforming worker answer, byte-identical replay, concurrent burst, captured
log scan, and an out-of-surface deletion case counted only against a real test file. Writes the
report, prints the verdict, exits 1 when blocked.
- docs/RC-SECURITY-EVIDENCE.md: procedure, rule, and the current finding stated plainly.
- Demo: a real run of the suite on camera, printing the verdict it derived instead of a flattering
one; CI regenerates the same evidence on every change.
The suite currently returns VERDICT: BLOCKED: bytes are trusted from the data-URL label, so an HTML
polyglot still reaches the worker and only fails afterwards on schema. That is a finding against
#16, not a passing check, and it is why the gate treats a failing case as a block.
The app keeps the journal in the person's browser, so the server has no diary to lose. What an
operator can lose is the corpus: the append-only manifest.jsonl with its provenance records and its
withdrawal tombstones. scripts/operator_backup.py archives only records and names every path it
refuses (model weights, transient inference images, secrets, logs, symlinks, oversize) with the
reason, writes one SHA-256 per stored member, and produces a reproducible archive.
restore re-verifies every digest, member name, and record count before it writes a byte, and refuses
absolute, traversal, link, unlisted, and non-empty-target cases, so a tampered archive or a lost
tombstone fails loudly. docs/PRODUCTION-RUNBOOK.md covers topology, network boundaries (including
the app's real 0.0.0.0 default and the staging HOST=127.0.0.1 template), the four secrets, retention,
and both rollback paths; its documented commands are parsed against the real CLIs by its test.
The release demo runs the real backup, verify, and restore against a synthetic corpus built by the
real ingest and revoke commands and shows what was stored, what was refused, and what came back.
The corpus had an append-only manifest, a reviewer tool, and an evaluation harness, but nothing a
stranger could read: what is in the corpus, what a label means, and how leakage is prevented lived in
code and in reviewers' heads.
scripts/dataset_card.py generates docs/DATASET_CARD.md from manifest.jsonl. Every count is read back
from real records — active samples, contributors, per-split contributor and sample counts, class
balance over active samples only — and every fact the pipeline does not collect (devices, contexts,
licenses) is printed as pending rather than estimated. The generator refuses to render a card for a
corpus where one contributor straddles two splits or where a recorded split is not the one the
subject key derives, and it never publishes a subject key or a withdrawn sample's labels. --check
fails when the committed page is not the generator's own output, so the card cannot drift into a
hand-written document.
docs/LABELING_GUIDE.md is the reviewer's rulebook: visible form, broad visible color, and photo
quality; visible color is not blood; nothing is diagnosed; a model suggestion is provisional input to
a human decision. Exclusions are named, and the split policy is stated where it is enforced.
The demo's new DATASET CARD scene builds a throwaway corpus with the real ingest command from
committed synthetic fixtures, publishes the card, and shows the refusal a planted contributor earns,
then re-checks the committed page. Its captions are release gates in build_release.py, which also runs
tests/dataset-card.test.py before a release exists.
tests/dataset-card.test.py covers the counts, the pending facts, the withdrawal receipt, both
refusals, determinism, the empty-corpus case, the committed-page identity, and both documents'
prohibitions. Every fixture is synthetic.
Closes#27
The corpus manifest records what was ingested and whether the person still allows it; it does not
record whether the labels are right. This adds the labeling control the data-flow threat model
named: scripts/review_corpus.py, a command-line tool with no app screen.
- The review queue shows the contributor's proposal and carries no model field at all: blindness is
structural, because build_queue() never receives predictions.
- reveal() refuses a contribution with no committed review, so the model answer cannot steer the
human who labels it. Revealing appends nothing.
- A review is one append-only line carrying the proposal it judged; a correction must change it. An
adjudication is a second reviewer's line that names the record it resolves; the review it
overrules survives untouched.
- A withdrawn contribution cannot be queued, reviewed, revealed, adjudicated, or trained on, and the
refusal says it was withdrawn rather than that it is already reviewed.
- trainable_records() returns only active samples with a human decision that was not an exclusion,
so unreviewed proposals are never training data.
Tests are written first against every refusal (tests/corpus-review.test.py, 21 cases); the release
builder runs the suite and gates four on-camera captions, and the demo now runs the real queue ->
review -> reveal -> adjudicate -> withdraw sequence against the committed synthetic sample and draws
the evidence card from that output. docs/REVIEWER-TOOL.md is the contract; no model was measured.
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.
Promote the ingest script into an auditable corpus pipeline. Each sample now
records consent (version + timestamp), source kind and raw-input digest, both
hashes, the derivation pipeline, review status, split, and lifecycle state, with
the caller's path never written down.
Withdrawal is a new `--revoke <contribution-id> --reason` operation: it deletes
the derived image and any raw pre-derivation copy, then appends a tombstone that
addresses each removed artifact by contribution id and digest while dropping
labels, subject key, and paths. History is append-only, so the original sample
record survives the withdrawal it documents.
Illegal moves are refused instead of silently applied: revoked -> active,
revoking an unknown or already-revoked contribution, an unknown state, a
duplicate contribution id, and a second label set for the same derived image.
The duplicate guard runs before any byte is written, so a refused ingest cannot
strand an image in the active dataset.
Tests written first and watched fail (tests/corpus-manifest.test.py): provenance,
tombstone + byte removal, active-record filtering, and the transition guards.
The transition helper itself passed on first write and is reported as a
hold-out check rather than a RED cycle. Wired into the quality workflow and
check:syntax; contract documented in docs/CORPUS-MANIFEST.md.
No browser surface changed, so record_release_demo.mjs is untouched and claims
nothing new: the nearest honest visible effect is the withdrawn-receipt
tombstone already recorded for #24.
A backup is the person's own record and has to keep working as the AI fields evolve.
- Export stamps schema version 2 and carries the confirmed logs plus the training
contribution receipts, including withdrawn tombstones; both are written through an
allowlist so unknown keys and credentials never leave the device.
- Import accepts version 1 and 2. A version 1 file is upgraded, the notice names the
version it came from, and entries that predate provenance stay null instead of being
given a source they never recorded.
- A file from a newer Timmy is refused with an update instruction, malformed, oversized
and over-limit files fail closed, and a refused import leaves the ledger unchanged.
- A log saved from a confirmed or corrected AI suggestion records its provenance
(source, suggested visible fields, model, confirmedAt) and the entry sheet states it.
- Service worker shell cache v5 -> v6 so installed clients actually receive the change.
Evidence: tests/portability.test.js (7 unit cases), tests/portability.acceptance.mjs
(real browser: v1 upgrade, future refusal with unchanged ledger, provenance recorded,
export v2, round trip), release demo scenes with on-camera caption proof, and the
release builder gate portable_backup_restore.
Refs #35
A journal save now records no training contribution. A contribution is a second
consent: the privacy page states consent version training-consent-v1 and the
receipt (contribution id, included fields, never-included fields) before the
confirm control is enabled, the receipt is queued on this device only and copies
no photo bytes, note text, or symptom answers, and withdrawal keeps a tombstone
naming every derived artifact by contribution id. Delete-all clears the receipts
with the ledger.
Closes#24
The demo now picks a committed text fixture that only claims to be a photo, asserts on camera that
nothing was posted to /api/analyze, and keeps manual entry in the same scene. The builder requires
that caption to have painted and records image_ingress_refusal in the manifest; the threat model
names each implemented control, its module, and the tests that enforce it.
- docs/PHOTO-DATAFLOW-THREAT-MODEL.md: browser, API ingress, model worker,
transient buffers, training opt-in store, logs, exports, and deletion,
with trust boundaries and a numbered threat register
- every named abuse case (replay, oversized input, metadata leakage,
poisoning, unauthorized training, operator access) carries a mitigation,
an owner, a tracked follow-up ticket, and the artifact that will prove it
- tests/threat-model.test.js enforces diagram coverage, owner/follow-up
presence against ROADMAP ticket numbers, and the non-diagnostic boundary
with synthetic-only fixtures
- link the model from the accepted product decision record
Logging must never depend on AI availability, and an unsaved entry must not
disappear when the user navigates away mid-log.
- add src/vision-readiness.js so the photo flow consumes one readiness contract
- when the worker is offline/disabled/unknown, offer no AI action and no
analysis-consent prompt, and keep manual Bristol entry one action away
- add src/draft.js and make the draft own the photo so a chosen photo and
entered context survive a recoverable close or navigation, with a visible
kept-draft summary and an explicit start over
- drafts stay in session memory only: a photo awaiting consent is never
persisted, and reload starts clean
- service worker now refuses to cache image bodies in addition to analysis
requests and no-store responses
- tests: 2 new unit files, 3 service-worker privacy assertions (1 hold-out),
1 browser acceptance file, release-demo and CI/package assertions
- release demo now visibly replays the vision-offline pass and the retained
draft with synthetic data and an in-script assertion
timmy
Timmy