feat(prose): a discussion holds for outstanding research — and the wait gate is one fetch

[leeovery]

What

The prose half of the research-wait change, stacked on #1091. A discussion re-entering over outstanding research is told so and keeps its reconcile flag until the research has landed; once it has, the landed file is read in full into the discussion before the session opens. Both conclude flows fetch the single wait-gate surface, and the experiment-wait-gate alias goes.

Changes

• reconcile-advisory.md, research branch — reads the research item's status first. completed: the existing advisory, the research file read in full, the flag cleared. in-progress / triaged: a different non-blocking advisory (research outstanding, in flight or parked; decisions may rest on ground it re-examines; this work cannot conclude until it lands; the menu's research row is the way in) and the flag held. The header sentence now states the one exception.
• discussion-session.md §G and topic-completion.md — the awaiting_experiments read plus conditional render become one render wait-gate fetch: empty proceeds, sections emit and STOP into the existing pause/keep branches. The pause signpost is kind-neutral; the gate names what is owed.
• Alias dropped — experiment-wait-gate removed from the engine, the usage banner, commands.md, CLAUDE.md, the simulation, and the four prose cases whose invariants named it (now render wait-gate, the awaiting_experiments read no longer asserted where the prose no longer performs it).
• CLAUDE.md — the Experiment paragraph names wait-gate; Staleness hops records the in-progress hop out of research and the advisory's held branch.

Cases

• discussion-holds-for-outstanding-research (new, claims world): resume over a parked research stub → the outstanding advisory, flag held → wrap-up meets the wait gate (parked) → pause. Invariants pin the status read, the gate fetch, and the absence of any flag delete, completion, or closing gate.
• discussion-resumes-on-landed-research (new, assertion state): resume after the research completed → the landed advisory, the research read in full, one manifest delete — ordered after the status read and before the resume gate — and no wait gate anywhere.
• Both fixtures are recipe-built on the epic mainline: the stub is born by a real research-side topic triage delivery from a sibling research session, the landed variant absorbs the concern and completes.

Verification

npm test 2783 pass · npm run typecheck clean · npm run test:cli green. node tests/prose/run.cjs select --diff main lists 54 cases; the six this slice touches directly are the two new ones plus discussion-spawns-an-experiment-and-continues, experiment-abandoned-returns-open, research-resumes-on-the-verdict, research-spawns-an-experiment.

Observation, not changed

Waits are work-type agnostic engine-side, but the advisory's "research row" is an epic-menu construct; a linear work unit with research reopened beneath an in-flight discussion would have no route back to the research. No prose path produces that state today (revisit reopens research only after the discussion has concluded), so it is noted rather than built.

🤖 Generated with Claude Code