Specs

App-merge P3 — parity / kill duplicates

Status SPEC on main (#217).

All sections

Learn

  1. Getting started
  2. The outcome ladder
  3. The handoff
  4. Asks
  5. Forgotten context
  6. Suspect commits
  7. MCP setup
  8. Benchmarks
  9. The loop
  10. The governor

Reference

  1. CLI63
  2. HTTP API16
  3. MCP Tools19

Specs

  1. Agent bus
  2. App-merge P3 — parity / kill duplicates
  3. App-merge
  4. Archie bench
  5. Archie — the memory agents do not have
  6. Asks: a questions-to-answers index
  7. Beliefs
  8. Cache economics — what returning cold actually costs
  9. CLI grammar, completions, and the cockpit
  10. Codex human turns
  11. Compaction diff
  12. Compaction, and what it costs
  13. Context composition — what fills the window before the work does
  14. Convergence
  15. Desktop App — feasibility spec
  16. The done gate
  17. Dropped commitments
  18. Efficiency receipts
  19. The extension surface
  20. Fleet view
  21. Governor
  22. Handoff
  23. home latency: web deck vs terminal, measured
  24. home
  25. Insights proxy naming (session user_messages ≠ human_turns)
  26. Local search and the small-model angle
  27. Loop
  28. Loose ends
  29. Market map: observability that opens PRs
  30. MCP server
  31. Machine memory
  32. The questions worth answering
  33. The AgentWorth and SpacePilot loop
  34. Suspect commits
  35. Token accounting
  36. Trajectory scrubber
  37. Verified outcome rate
  38. Wake

Research

  1. All memos3

Changelog

  1. Every releasev0.1.28

Parent product call (do not re-open)

From docs/specs/app-merge.md / #212:

  1. Archive stays a phase inside the home deck through P3 (a / /home/s/<id>).
  2. Do not delete apps/dashboard.
  3. Do not flip the serve root (/ → home deck).
  4. P4 (flip root / delete dashboard) and P5 stay blocked until P3 dogfood + explicit COS relaunch of P4.

This SPEC is only the P3 parity cut. Closed early-delete attempt: #182.

Goal

Bring the dashboard-only rail surfaces into the home deck so the archive phase is no longer "sessions view only," then remove safe duplicate codepaths. After P3 dogfood, the deck is the place a person lives for Overview / Coverage / Archaeology / Exports / palette and the archive phase — while apps/dashboard remains in the tree until P4.

In scope (COS lock)

1. Port these panes into the home deck

Surface Dashboard home today Deck target
Overview apps/dashboard/src/shell/OverviewPane.tsx (pulls VerdictBoard, FleetStrip, cache-cliff) A deck-reachable Overview pane (rail or equivalent phase / route — pick one navigation model in the implement PR and document it in apps/home/DESIGN.md / archive header)
Coverage apps/dashboard/src/shell/CoveragePane.tsx + components/CoverageMatrix.tsx Same
Archaeology apps/dashboard/src/shell/ArchaeologyPane.tsx + components/ArchaeologyPanel.tsx Same
Exports apps/dashboard/src/shell/ExportsPane.tsx (+ embedded ExportModal) Same
Command palette apps/dashboard/src/shell/CommandPalette.tsx Deck-global palette; commands must reach the new deck panes and existing archive / insights routes

P2 left an explicit gap — apps/home/src/archive/Archive.tsx and InspectorPane.tsx state that rail / overview / coverage / archaeology / exports / palette stay on the dashboard until P3. This SPEC clears that debt. Empty-inspector copy that says "no OverviewPane until P3" must be updated when Overview lands.

API clients already partially mirrored: apps/home/src/archive/api.ts already knows fetchArchaeology / fetchCoverageMatrix types. Prefer wiring those rather than inventing new endpoints.

2. Shared-package dedupe

After the panes work in the deck, collapse duplicated modules into a shared package (likely under packages/ui or a small new packages/ sibling — implement PR chooses based on existing import graph; do not create a parallel design system). Goal: one implementation imported by deck (and, until P4, optionally still by dashboard), not two drifting copies.

3. Safe delete after cutover — identical twins only

Measured 2026-09-30 (MD5):

File apps/dashboard/src/shell/ apps/home/src/archive/shell/ Identical?
TrajectoryScrubber.tsx yes yes Yes (same MD5)
OutcomeLadder.tsx yes yes Yes (same MD5)

Allowed after deck cutover + shared import: delete one of the twin copies (or both, once the shared package owns them) for only these two files, once the deck (and dashboard, while it still embeds them) import the shared module and tests pass.

4. Keep-winners (COS written decisions)

Recorded keep-winners for P3. Do not reopen without a new COS call.

Pair Decision Notes
TrackScrubber vs TrajectoryScrubber BOTH KEEP apps/home/src/deck/TrackScrubber.tsx stays (deck course track). packages/shell/TrajectoryScrubber.tsx stays (archive/dashboard trajectory). Do not delete or unify.
VerdictBoard twins Extract → packages/shell Dashboard + home archive copies were byte-identical. Authoritative module: packages/shell/VerdictBoard.tsx. App twins deleted after rewire to @shell/VerdictBoard. OutcomeLadder already lives in packages/shell — keep; no LadderFunnel in repo.
FleetStrip vs Ambient FleetStrip → packages/shell; Ambient KEEP apps/home/src/deck/Ambient.tsx stays — do not delete or merge into FleetStrip. Shared strip: packages/shell/FleetStrip.tsx with required onOpenSession (dashboard call sites pass navigate/open). App FleetStrip twins deleted after rewire.

Earlier wording treated VerdictBoard / FleetStrip as "do not delete without a keep-winner." Those keep-winners are now written above; the extract+delete of the app twins (not TrackScrubber / Ambient) is in scope.

Success criteria

  1. From the home deck (no legacy dashboard window required for daily use of these surfaces), a user can open Overview, Coverage, Archaeology, Exports, and the command palette, and each pane loads real API data (same contracts the dashboard uses today).
  2. Archive phase still works (a / /home/s/<id>); product call #212 unchanged.
  3. Dual codepaths for the ported panes are either deleted or reduced to a single shared package import (dashboard may keep a thin re-export until P4).
  4. TrajectoryScrubber + OutcomeLadder exist in at most one authoritative module after cutover (shared package or single app path).
  5. Keep-winners in §4 are honoured: TrackScrubber + Ambient remain in place; VerdictBoard / FleetStrip app twins may be deleted only after shared @shell import.
  6. apps/dashboard remains in the tree and still builds; serve root unchanged.
  7. Docs: this SPEC + docs/specs/app-merge.md phase table updated to "SPEC opened" / implement PR links; Archive.tsx header comment no longer says "until P3" once implement lands.
  8. Ask-before-main: no merge without COS / Saurabh.

Non-goals

  • P4: delete apps/dashboard, flip / to the home deck.
  • P5: voice (#144), promote-to-default, Show HN, dsh.
  • Visual redesign of the deck or a new design system.
  • Pi / OpenAI adapter work, live-tail (#213), honesty matrix (#215) — other trains.
  • Deleting TrackScrubber or Ambient "because P3." (keep-winners in §4).
  • Merging implement PRs to main without approval.

Suggested implement stack (after this SPEC PR)

Prefer quality over speed. Suggested order for follow-up PRs (not this one):

Step Branch theme Notes
A Shared extract of identical TrajectoryScrubber + OutcomeLadder Lowest risk; proves package wiring
B Port Coverage + Archaeology (API already typed in archive) Self-contained panes
C Port Exports + Command palette Palette must know new targets
D Port Overview (VerdictBoard / FleetStrip — share per §4 keep-winners) Highest product surface
E Dedupe remaining dual imports; update Archive / DESIGN comments; dogfood checklist Stop before P4

Each step is its own PR when possible. Do not batch a giant delete with the first port.

Dogfood checklist (before anyone asks for P4)

  • Deck-only session: Overview / Coverage / Archaeology / Exports / palette all used once on a real index
  • Archive phase still opens a session and scrubs trajectory
  • Dashboard still serves for anyone who bookmarks it (until P4)
  • No orphan imports / broken rust-embed asset build for either app

References

  • docs/specs/app-merge.md — locked product call (#212)
  • docs/specs/home.md — deck product shape; app-merge note
  • apps/home/src/archive/Archive.tsx — "until P3" comment to clear on implement
  • apps/dashboard/src/shell/{OverviewPane,CoveragePane,ArchaeologyPane,ExportsPane,CommandPalette,TrajectoryScrubber,OutcomeLadder,FleetStrip}.tsx
  • apps/home/src/deck/TrackScrubber.tsx — keep; not a twin of TrajectoryScrubber
  • Closed early delete: #182

docs/specs/app-merge-p3-parity.md on GitHub

Point it at your own machine.

npx -y agentworth@latest scan