Jev conversation-to-document workflow

[MaggieAppleton]

Ordinary chat can create sourced decision cards and shared options. People explicitly Save a choice; a scoped Planner job then writes that saved decision into the document. Any writer can add multiple options to an open card. Existing saved custom answers remain readable.

Review status: all ten incoming PRs (#242–#251) are merged into this prototype. All 14 replacement PRs are published and independently reviewed; their complete Git tree exactly matches this integrated PR. All 14 PRs are ready for review with passing current-head validation, browser, and container checks. Keep this integration PR in draft as the complete-feature reference; review the replacement stack below.

Review and merge order

Order	PR	Review boundary
1	#252	Bounded state, identities and source ranges
2	#253	Event replay and restoration
3	#255	Decision records and shared drafts
4	#256	Exact question context
5	#257	Interpretation policy and complete option lists
6	#258	Jev adapter and checked judgments
7	#259	Durable follow-up work
8	#260	Storage and research recovery records
9	#261	Document projections
10	#262	Decision services and fenced commits
11	#263	Conversation processor and runtime
12	#264	Research execution and placement recovery
13	#265	Chat, decision controls and feature activation
14	#266	System verification and review guides

Each PR is based on the preceding layer. Review bottom to top and keep the stack synchronized with gh stack as layers land. The service, runtime and UI layers remain substantial; their descriptions separate production changes from test/support volume. Nothing in this stack has been merged into main.

What changes for users

• Chat moves to the left; desktop panes remain side by side down to 500px.
• Jev interprets ordinary discussion using exact, attributed excerpts. Open cards, a title/goal, refinements, and advisory option suggestions may appear before Save.
• Shared decision cards support repeated option additions, Save, Reopen, and Discard. Saving writes prose grounded in the current selected choice and rationale.
• Evidence links connect cards, chat, and document prose. People can attach missed excerpts and inspect analysis through explicit debugging buttons.
• Optional research offers require human acceptance before work starts, then use Chopin’s existing child-document workflow.

Explicit @chopin editing, collaborative documents, and ordinary research remain existing Chopin foundations. Conversation analysis remains opt-in with CONVERSATION_PLAN=on.

Current interface

Real Chromium capture with controlled Jev/Planner fixtures: Chat on the left, exact source highlighting, a shared option selected, and explicit human Save. The document opening was seeded; this demonstrates the interface, not live model quality.

Scope and cleanup

Against pinned main 3e57e40e, the cleaned tree adds 94,627 lines across 862 changed files: 25,534 application/package lines, 68,200 tests and test infrastructure, 761 documentation, and 132 other lines. These are textual additions, including comments and whitespace. The classification now counts 592 lines of confirmed test-only drivers and helpers with test infrastructure, rather than application code.

The cleanup removes sentence-specific recovery rules, the unused unkeyed option route, artificial callback/read-order tests, vacuous assertions, and obsolete documentation. It retains the keyed shared-option route and meaningful durability, permission, recovery, and multi-user tests. It also fixes full-history processing cost, reserved option capacity, duplicate option IDs clearing selections, partial option-list admission, stale marker geometry after card collapse, and recovery after exhausted research-link lookups. The last fix retries an existing read and cannot start more research. Relative to the branch with all ten incoming fixes, cleanup is +1,823/−4,604 lines (net −2,781).

Verification

• Final tree b46f3d76 (identical to integrated head a4ad84ef): 3,551 unit tests, with three PostgreSQL-only skips; 109 component Chromium tests; 331 application Chromium tests in GitHub CI; 79 sealed Jev Chromium tests. Browser pass runs report no retries. Types, repository checks, and build pass; all three integrated-PR GitHub checks are green.
• 72 real PostgreSQL contract/lifecycle tests passed on the unchanged server implementation. The extracted layers additionally pass their focused checks; the complete local stack and this PR share Git tree 66dd073c962973790e2d099ea9a7317f789298e6.
• A local application run had one mobile comment-sheet drag failure (330/331). That unchanged case passed in isolation and in the complete GitHub run; its intermittent cause is unconfirmed. No production guard or assertion was weakened to obtain the passing runs.
• Controlled provider fixtures exercise actual authorization, persistence, sockets, and UI. Live model quality across paraphrases remains unverified. All 14 replacement PRs pass all three current-head CI jobs. Final stack CI also ran 14 design and 331 application browser cases without retries or flaky classification.

UI reconciliation · Feature quality handoff

Scope and review map

The scope image explains the user-facing capabilities; the review map shows the 14-layer order and separates application changes from test/support volume.