Follow published research findings into the parent document

[MaggieAppleton]

Completed research and investigations now queue a Planner review in the parent document. The review compares captured evidence with current prose and human decisions, revises provisional recommendations when warranted, links the evidence, and explains the next useful move.

Publication atomically creates one durable receipt per captured result. Reviews use the existing Planner owner and wait through unavailable ownership, remote disconnection, or an active build. Result cards show progress and give writers Retry/Dismiss controls. Interrupted work requires an explicit retry; reconnects refresh its status without replaying model work.

This applies to newly published ordinary results. Existing prototype-spike settlement and consent for further research or local execution remain in place. Historical results are not backfilled.

Verification

• Final head e5df142d4366efa5bb1a9a39f9736066992e219d: all 24 GitHub checks passed, including all unit shards, real PostgreSQL contracts, 12 browser shards, design/gallery, production container, CLI and Windows CLI.
• The downloaded browser report matches that head: 800 integration and 41 design/gallery cases passed; 11 skipped. Two existing navigation cases passed on their first retry: a fixed-delay Back/Forward animation sample and an intermittent parent-rename last-visited-document mismatch. Independent review found no introduced regression; the intermittent last-visited-document mismatch remains unresolved outside this feature.
• Locally, all 17 combined follow-through/preview browser cases passed without retries; all 499 Chat tests passed. The canonical-navigation focus regression failed 3/3 under CPU slowdown before the fix and passed 3/3 afterward. Workspace types, formatting/design checks, explicit lint, and independent exact-head review pass.
• Coverage includes receipt publication/fencing/restart, current-owner admission, interrupted recovery, writer controls and reader authorization, actual socket reconnect, one-time delivery, and contradictory evidence applied through real document tools while preserving human prose.
• The behavioral demo uses a scripted Planner; it is not a live-model quality evaluation.

Before and after

The captured finding challenges caching every query. The review narrows the provisional recommendation, preserves the human retention constraint, and links its explanation to the evidence child.