docs(adr): ADR 0010's "pending A2 (#350)" points at a closed issue

CHANGELOG.md

@@ -68,7 +68,7 @@
 ### Architecture
 
 - **ADR 0013** — persist the `beta1m` fact, derive the session window (expert-panel design: Kleppmann/Hickey/Helland/Bailis).
-- **ADR 0010 rewritten by #350** — per-conversation coreHash ownership replaces per-turn early-exit.
+- **ADR 0010 rewritten by #350** — per-conversation coreHash ownership replaces the per-turn early-exit in the **batch** path (`wfInferLanes`) only; the live path and turn list keep the per-turn form.
 
 ## 2.0.0
 

CLAUDE.md

@@ -35,7 +35,7 @@ These constraints are enforced by structural encapsulation where possible, and g
 - **Use `_wfIsMainLane(lane)` for main/orchestrator detection, never `!lane.spawnParent`** (`spawnParent` is always `null`, in every lane object, everywhere) — @docs/decisions/0007-wf-is-main-lane-not-spawn-parent.md
 - **Temporal overlap overrides agentKey for lane placement** — the main lane is strictly serial (no temporally-overlapping turns); same-convId overlap turns (jitter/re-send/rewind) ride main's `overlapEntries` as event markers, never a separate lane (#364); different-convId overlaps still create `parallel-*`/`agent-*` lanes; null-convId non-main lanes retain strict no-overlap split. `_wfSeqRetroMove` must rebuild `mainConvIds` and migrate stale markers after every retro-move — @docs/decisions/0008-temporal-overlap-overrides-agent-key.md
 - **Sequential-interleave classification goes through the shared seq tracker in both files** (`wfCreateSeqTracker` in `workflow-timeline.js`; instances in `wfInferLanes`, `wfAddEntry`, and `entry-rendering.js` `addEntry`), and the tracker never consults `isCompacted` (fan-out first-turns carry a false flag) — @docs/decisions/0009-sequential-interleave-conv-bracketing.md
-- **coreHash lane ownership is per-CONVERSATION, never per-turn** — in `wfInferLanes` a main-agentKey turn is a subagent iff NONE of its conversation's *plurality-tied* coreHashes (`_wfComputeConvIdentity`, `convMaxHashes` = the hashes sharing the max count) is in `mainCoreHashSet` (the seed conversation's dominant only — routing is per-conversation, so a mixed `/model` conversation already rides main as a unit; adding a main conversation's *secondary* coreHashes over-merges a separate subagent sharing that hash — real c6e1ddaa). A lone seed/blip turn carrying main's coreHash is a minority so it cannot flip its conversation into main (add88512); a genuine plurality tie resolves toward main-membership so a `/model`-switching compaction successor is not ejected. Seed = the earliest-starting main-agentKey CONVERSATION that has a coreHash (not the earliest coreHash-bearing turn — else a subagent could seize the seed when main opens coreHash-less). This REWRITES ADR 0010's per-turn early-exit for the batch path; the live `wfAddEntry` + turn-list still use the per-turn form pending A2 (#350) — @docs/decisions/0010-corehash-identity-routing.md
+- **coreHash lane ownership is per-CONVERSATION, never per-turn** — in `wfInferLanes` a main-agentKey turn is a subagent iff NONE of its conversation's *plurality-tied* coreHashes (`_wfComputeConvIdentity`, `convMaxHashes` = the hashes sharing the max count) is in `mainCoreHashSet` (the seed conversation's dominant only — routing is per-conversation, so a mixed `/model` conversation already rides main as a unit; adding a main conversation's *secondary* coreHashes over-merges a separate subagent sharing that hash — real c6e1ddaa). A lone seed/blip turn carrying main's coreHash is a minority so it cannot flip its conversation into main (add88512); a genuine plurality tie resolves toward main-membership so a `/model`-switching compaction successor is not ejected. Seed = the earliest-starting main-agentKey CONVERSATION that has a coreHash (not the earliest coreHash-bearing turn — else a subagent could seize the seed when main opens coreHash-less). This REWRITES ADR 0010's per-turn early-exit for the batch path; the live `wfAddEntry` + turn-list keep the per-turn form as an accepted bounded divergence (no open issue tracks a mirror) — @docs/decisions/0010-corehash-identity-routing.md
 - **context% denominator: display folds a per-session window, classification never does** — display sites (session card, its window label, timeline minimap, swimlane) divide by `sessionCtxWindow(sid)` (or the `_wfWinByTurn` lane fold): 1M if any main turn has `beta1m===true` (authoritative, `SUPPORTS_1M`-gated at write time), else the largest `maxContext` fossil (heals legacy), else 200K. Classification (`isCompacted`, per-turn `severity`, lane placement) keeps dividing by raw per-turn `maxContext` — never the fold (feeding a future-dependent latch into a prefix-local classifier reintroduces the reverted live≠batch divergence). The fold is stateless-at-render (never a stored `sess.maxWindow` → can't go stale); `beta1m` is persisted add-only so it's rebuildable. Never cross the two — @docs/decisions/0013-beta1m-persist-session-window-derive.md
 - **persisted weather carries a derivation revision; all three writers stamp it through `_assignWeather`** — the load-time probe tests field EXISTENCE and `reconcile` compares COUNTS, so neither notices a change in HOW weather is derived; a semantics change without a `WEATHER_REV` bump leaves every cold session card rendering the deleted rule with all checks green. An unstamped writer (restore's step-6 `setWeather` overwrites the rebuild's output) turns the probe into a permanent rebuild loop — @docs/decisions/0013-beta1m-persist-session-window-derive.md
 - **cost-worker lifecycle: imported mode is side-effect free, executed mode exits by drain** — R1 no handle may ref the loop past the final stdout write; R2 no `process.exit()` on success; R3 one result protocol (stdout JSON); R4 adding an env-derived root obliges updating `isolatedEnv()` in every test that forks the worker — @docs/decisions/0015-cost-worker-lifecycle-drain-exit.md

docs/decisions/0010-corehash-identity-routing.md

@@ -2,7 +2,8 @@
 
 - Status: Accepted — **batch routing rule REWRITTEN by #350 (2026-07-24), see
   "## Rewritten by #350" below**; the live path (`wfAddEntry`) + turn-list
-  (`entry-rendering.js`) still use the original per-turn early-exit pending A2.
+  (`entry-rendering.js`) keep the original per-turn early-exit as an ACCEPTED
+  bounded divergence — see "A1/A2 boundary" below; no open issue tracks a mirror.
 - Date: 2026-07-16
 - Related: #258 / #257 / #350 / ADR 0005 / ADR 0008
 
@@ -227,8 +228,9 @@ fix any *completed* session on cold-load+rebuild, which is how the dashboard
 renders `4b15c248`. `wfBuildState` re-derives `mainCoreHash`/`mainConvIds` from
 the now-correctly-composed final main lane, so the live `wfAddEntry` and
 turn-list `entry-rendering.js` paths inherit a correct main identity for new
-turns; their per-turn early-exit is left in place. A2 (#350) mirrors the
+turns; their per-turn early-exit is left in place. Mirroring the
 per-conversation rule + dominant-coreHash flip/trunk rebuild into those two
-sites. Until A2, the only residual divergence is a *live* seed turn arriving on
-an actively-viewed session — bounded and self-correcting on the next full
-rebuild. See `docs/solutions/same-convid-lane-classification.md`.
+sites — once tracked as "A2" — was never implemented, and #350 is closed, so
+nothing tracks it. The residual divergence is therefore ACCEPTED, not pending:
+a *live* seed turn arriving on an actively-viewed session — bounded and
+self-correcting on the next full rebuild. See `docs/solutions/same-convid-lane-classification.md`.

public/workflow-timeline.js

@@ -1097,7 +1097,8 @@ function wfAddEntry(entry) {
   // derived from the (now correctly composed) final main lane — so a fresh
   // cold-load/rebuild of a completed session is already correct, and the only
   // residual divergence is a *live* seed turn on an actively-viewed session.
-  // A2 mirrors the per-conversation rule + dominant-coreHash flip/rebuild here.
+  // Mirroring the per-conversation rule here (once "A2") was never implemented
+  // and #350 is closed — this per-turn form is the accepted bounded divergence.
   if (entry.agentKey && WF_MAIN_AGENT_KEYS[entry.agentKey] &&
       wfState.mainCoreHash && entry.coreHash && entry.coreHash !== wfState.mainCoreHash &&
       entry.convId && wfState.mainConvIds && !wfState.mainConvIds.has(entry.convId)) {