Repository navigation
feat(gate1): publish a real review round + the measurement protocol — gate 1 / PR 4 - #375
Merged
Merged
Conversation
… gate 1 / PR 4
mcp/gate1/publish_review.py plays the agent for one honest round on a project folder: save +
snapshot the baseline, record the brief (rev 1 + log), render the baseline excerpt with
export_av, then for each direction copy the project to an ISOLATED sibling folder, load the
copy, apply the edits, save, render the same excerpt, and snapshot the copy INTO the original's
work/versions (ADR-0062 §2 — the foreground document is never edited); write the candidate
records with provenance (the edits, the working copy) and evidence (export result, dropped
frames, measured RMS for the audition-only level match, "intent evaluator unavailable"); reload
the original and publish one review item + review_opened. Refuses to publish an excerpt whose
export dropped frames. The whole round on song-sketch: 28 s, 3 × 465 frames.
Control server + MCP (gate-1 slice; full families are gate 2):
- work_status — records dir, has_records, preferred_version, dirty, brief_rev, open reviews.
- work_snapshot_version {label, source, parent, brief_rev, into} — a version of the LOADED folder,
optionally into another project's records; refuses an unsaved document (reads disk).
- work::snapshot_version_into() behind it (+ test: the copy's own work/ is never content, nothing
lands in the copy).
Fixes found on the way:
- save_project over MCP never cleared the dirty flag / autosave slot (the GUI save does): an agent's
save left the edited-dot on and the quit prompt armed. Now mark_saved() + autosave::clear().
- Leaving Review appended review_visited into ANY folder project, creating work/ where none existed
(it wrote into examples/song-sketch). Now only when work/ already exists.
docs/roadmap/gate1-measurement.md is the protocol for the gate itself: the creator returns cold,
two minutes, no transcript, eight silent observations, the ADR-0064 probes, and a Result section
to fill in. The plan doc's PR 4 section reflects what shipped.
Verified live: the published round appears in Review (badge 1) with three lockstep tiles, level
match enabled from the measured RMS, brief + protections shown. Full headless ctest green (91);
MCP name + argument parity green; layering guard green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
jeffcrouse
enabled auto-merge (squash)
September 19, 2026 14:21
Contributor
Production gate (core): ✅ PASS
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fourth and last code PR of gate 1 (plan:
docs/roadmap/continuous-creative-workflow-gate1.md). Real media, honestly produced, plus the protocol for the measurement the gate actually asks for.mcp/gate1/publish_review.py— one honest roundsave + snapshot the baseline → record the brief → render the baseline excerpt → for each direction: copy the project to an isolated sibling, load the copy, apply the edits, save, render the same excerpt,
work_snapshot_version … into=<original>(ADR-0062 §2: the foreground document is never edited) → candidate records with provenance (edits, working copy) and evidence (export result, dropped frames, measured RMS for the audition-only level match, "intent evaluator unavailable") → reload the original → publish one review item. Refuses an excerpt whose export dropped frames. Nothing inwork/is hand-written. On song-sketch: 28 s for the round, 3 × 465 frames.Control server + MCP (gate-1 slice; full families are gate 2 / ADR-0063)
work_status— records dir,has_records,preferred_version,dirty, brief rev, open reviews.work_snapshot_version {label, source, parent, brief_rev, into}— refuses an unsaved document (it reads disk).work::snapshot_version_into()+ test.Fixes found on the way
save_projectover MCP never cleared the dirty flag / autosave slot (the GUI save does) — an agent's save left the edited-dot on and the quit prompt armed.review_visitedinto any folder project, creatingwork/where none existed (it did so inexamples/song-sketch). Now only whenwork/already exists.docs/roadmap/gate1-measurement.md— the protocolCreator returns cold, two minutes, no transcript; eight silent observations; the ADR-0064 probes (no agent connected, undo/redo truthfulness, feedback on an older preview, Neither, back to Create, save prompt); a Result section to fill. Pass = found it, could say what changed, chose, left a bar-specific comment, all inside 2:00 without leaving Review.
Verified live
The published round appears in Review (badge
1) with three lockstep tiles, level match enabled from the measured RMS, brief + protections shown.Test plan
test_work_records(+snapshot_version_into: 114 checks) · full headless ctest green (91)vividbuildspublish_review.pyend to end on a scratch copy of song-sketch (28 s, 0 dropped frames)gate1-measurement.md+ ADR-0064's acceptance section as a follow-up🤖 Generated with Claude Code