-
SUPERSEDED — Timmy sleek + Hermes review release — 2026-08-20.1
Pre-ReleaseAll checks were successfulQuality gates / quality (push) Successful in 1m32sreleased this
2026-08-20 16:28:17 +00:00 | 16 commits to main since this releaseSUPERSEDED — DO NOT DEPLOY. Independent post-merge review found an urgent-symptom interception gap in the Hermes chat boundary. Use the forthcoming 2026-08-20.2 hotfix release instead.
Original source SHA-256: 40f18b0699a76791a31484b97423136c3ea75140a4136136369e180dcd9622cc
Original video SHA-256: ae5fd45ab6d73e078d9fbafcb6544762741b3c64cdd4be9183c898c7936b5cdb -
Timmy daily review 2026-08-20
Pre-ReleaseAll checks were successfulQuality gates / quality (push) Successful in 1m26sreleased this
2026-08-20 13:18:33 +00:00 | 18 commits to main since this releaseReview candidate
- Exact commit:
930f156590b9a9d01c275d49f14dd0108b85b6a3 - Completed: issue #9 / PR #46 — product boundary and release authority decision
- Tangible change: Timmy now has a tested durable decision record limiting AI to visible Bristol form, broad color, and image quality; it prohibits diagnosis and nonvisual symptom inference, fixes routine release authority, and names the three human gates.
- Demonstrated features: the real mobile app replays a clearly labeled synthetic photo-first path with visible touch feedback, consent before analysis, bounded AI scope, user correction/confirmation, and the clinical/privacy, beta-consent, and RC-approval gates shown on screen.
- Model/profile:
SmolVLM2-2.2B-Instruct/selfhostdemo profile using a deterministic synthetic provider response. - Gates: 27/27 unit/security passed; mobile green path passed; photo-first acceptance passed; dependency audit found 0 high vulnerabilities; syntax, diff, secret, forbidden-artifact, H.264 720x1280 probe, complete MP4 decode, contact-sheet review, and full-resolution changed-feature frame review passed.
- Source SHA-256:
53ad02631562c89b876ee2e63bc119a471fc75c72beb9648c350c9459048b833 - Demo-video SHA-256:
5d4b6cc9e9111de6469e3cbe6a6ce23482a74dcd30d745ef3e35a1f3a3b9b182 - Known accuracy limitation: open-weight VLM integration acceptance is proven; Bristol-form accuracy is not clinically validated. Timmy provides visual assistance, never diagnosis, and user confirmation remains mandatory.
- Exact commit:
-
Timmy daily review 2026-08-19.2
Pre-ReleaseAll checks were successfulQuality gates / quality (push) Successful in 1m18sreleased this
2026-08-19 12:10:19 +00:00 | 20 commits to main since this releaseReview candidate
- Exact commit:
8ab1e85bd1a7c9c0b42e1f1f04ddea60f6479c5b - Completed: issue #8 / PR #45 — reproducible CI quality gates
- Tangible change: Gitea Actions now gates pull requests and main on reproducible install, 26 unit/security tests, both real-browser mobile flows, high-severity dependency audit, syntax, and changed-range diff hygiene.
- Demonstrated feature: the real mobile app replays a clearly labeled synthetic photo-first path, shows touch interactions and consent, suggests only visible Bristol form and broad color, preserves the non-diagnostic warning, and requires user review/confirmation before persistence.
- Operational boundary: the CI system is operational; the demo's self-hosted provider response is deterministic synthetic fixture data, not a live clinical model evaluation.
- Model/profile:
SmolVLM2-2.2B-Instruct/selfhostdemo profile. - Gates: 26/26 unit/security passed; mobile green path passed; photo-first acceptance passed; dependency audit found 0 high vulnerabilities; syntax, diff, secret, forbidden-artifact, H.264 720x1280 probe, and complete MP4 decode passed.
- Source SHA-256:
d1cc85eba453c0e153071c3aec27cb7b62d108b6ce8b16706bc59f6e0295b955 - Demo-video SHA-256:
f18c618fc2faf0ba9cfa2553e6f1591a3abc13fddb2fed11ddc8bd63c8fb156a - Known accuracy limitation: open-weight VLM integration acceptance is proven; Bristol-form accuracy is not clinically validated. This is visual assistance, never diagnosis, and user confirmation remains mandatory.
- Exact commit:
-
Timmy daily review — 2026-08-19 video edition Pre-Release
released this
2026-08-19 03:38:07 +00:00 | 23 commits to main since this releaseEvery Timmy review release now requires a real-app feature demonstration.
- Demonstrates photo-first self-hosted readiness, explicit consent, constrained Bristol/color suggestion, and user confirmation
- 22/22 unit/security tests passed
- Both mobile acceptance flows passed
- MP4 fully decoded and probed at H.264 720×1280 yuv420p
- Source SHA-256:
00f55e2443fd7c38c0663e2e9a95ee659364da3fdbb3899ee73f06e5224c4b72 - Video SHA-256:
d0d431dde41e353b15eed04b3b6c66e581bf76d44f04626fc1777a9ab6b3091a
The recording uses a synthetic fixture and deterministic provider response. Open-weight stool-image acceptance is proven; clinical Bristol accuracy is not.
-
Timmy daily review — 2026-08-19 Pre-Release
released this
2026-08-19 03:05:01 +00:00 | 24 commits to main since this releaseGated review build from
ce84ea577c01.- 22/22 unit and security tests passed
- Mobile journal and photo-first acceptance passed
- Dependency audit found 0 vulnerabilities
- Secret and forbidden-artifact gates passed
- SHA-256:
287395ffdb063d02b6588aa26fb516f0d6254a6cd36a59322fbd606edaa25a56
Known limitation: the self-hosted VLM accepts real stool imagery, but Bristol accuracy is not clinically validated. Suggestions require user confirmation.