feat: declare the journey map and keep results observable

[alphali]

Why

A round can already order its findings with journeyRef / step / role, documented here since 2026-08-12, and agents do use it. A replay against a real workspace (DQ-OLCRHL) returned 38 findings, 82% of them ordered into 8 journeys, with entry / action / outcome / failure all in play.

What no field carries is the journey's own name, goal and actor. Ordering says which findings belong together; nothing says what the journey is. Doable therefore titles each flow with one raw finding sentence and takes its goal from another.

What changes

journeys on the answer candidate. One entry per journeyRef you ordered: name, goal, actor, alternatePaths.

A declaration names journeys the findings already ordered — it cannot introduce one. normalizeJourneys rejects a journeyRef with no ordered findings, rejects duplicates, and puts every field through assertNoLocalProvenance, because a journey's wording crosses the boundary like any other prose.

journeyMapSkipReason declines when a map would add nothing. An empty map with a reason is a better answer than a guessed one.

Both are optional. A submission that only orders its findings validates exactly as it does today.

Two fixes alongside

The observable-result rule did not say observable to whom. The same replay produced verifications a tester cannot check — "the selected vote is submitted for the scene through the Live scene-vote request". That is the mechanism, not the result; the vote total, the credit balance and the disabled free tier were all established by the same findings and all visible on screen. 6 of that TRD's 37 verifications had this shape. The contract now says that where a behavior has a user-visible surface, the observable result is what that surface shows, and the pre-submit review restates such findings rather than only deleting ones that lack a result.

The watch checklist expected stop for consumed. A pre-create Round now hands its code to the follow-up Rounds of the TRD it just created, so the watch continues on the same DQ and the server answers wait. Only cancelled, or a Round bound to no session, stops. The Skill already follows next_action, so this is the checklist catching up to the server — item 10 in TESTING.md.

Testing

npm test — 2/2, including four new assertions: a valid declaration round-trips, a journeyRef with no ordered findings is rejected, a duplicate declaration is rejected, and local provenance in a journey goal is rejected.

node scripts/verify-release.mjs — doable-code-context@0.2.6, 3 Skills, Codex / Claude Code / Cursor.

Not covered: whether an agent actually produces journeys once this ships. The released plugin has no notion of the field, which is why the replay's journey_map came back NULL. That needs one round through an installed build.

Ships with

getdoable/trd — journey_map on the round response, journeys on RoundSubmission, consumption in create_source_extraction, and the code_context_round_revisions.journey_map column.

🤖 Generated with Claude Code