[bwachman]

Note on base

This branch is stacked on the cl-10432 branch, which sits on top of the stack that provides the worker-wait / parked-question machinery. The diff is the CL-10437 change only: the queued-input strip on top of cl-10432. The base must stay cl-10432-codify-outcome-aware-escalation-policy-for-blocked-worker so the PR shows only the single CL-10437 commit.

Summary

Shows accepted-but-unread queued operator input in the TUI so the operator can see that the dispatcher has accepted input that has not been read yet. When a queued delivery is accepted (an accepted final result arrives), the ledger entry appears as a QUEUED INPUT row and stays visible until the director reads the message, which is signaled by the existing upstream message.run.started event (Option A read signal). Worker wakes never enter the ledger. The strip occupies a one-row geometry zone, coexists with the WORKER WAITING zone, and collapses and restores cleanly.

Verification

After the rebase onto main and the review fixes, CI passes on 9cfed0e; the results below are from the original verification.

Check	Result
Focused suite (bun test src/tui/queued-input.test.ts src/tui/runtime-bridge.test.ts src/tui/queued-delivery-hop.test.ts src/tui/queued-input-chrome.test.ts src/tui/worker-wait-chrome.test.ts src/tui/geometry.test.ts src/tui/geometry/resolve.test.ts src/tui/stream-event-map.test.ts)	206 pass / 0 fail
Typecheck	clean
Regression	none
Read-signal blocker (Option A)	resolved: ledger exits on the existing upstream message.run.started event (not message.received), so the row paints until the director reads the message
Worker wakes in ledger	none (wake-origin fix)
End-to-end mid-run steer probe	row visible until the director reads the steer
v2 review findings	all resolved (v2 blocker + Should-fix 1 wake-origin + Should-fix 2 runner-level/failure-discard tests + Should-fix 3 docs/TUI.md + nits; build-info fixup is not in this branch since cl-10200 does not carry it)

AC to code and test mapping

• QUEUED INPUT row visible while accepted-but-unread operator input exists: src/tui/queued-input.ts + queued-input-chrome.test.ts
• Ledger entry only after an accepted final result, exits on message.run.started (Option A read signal): src/tui/runtime-bridge.ts + stream-event-map.ts + queued-input-chrome.test.ts
• messageId identity/dedup, FIFO, out-of-order consume; no content fallback: src/tui/queued-input.ts
• Worker wakes never enter the ledger (wake-origin fix): src/tui/runner/submit.ts + delivery-queue.ts + queued-input-chrome.test.ts
• Lifecycle clears (clearQueuedDelivery/dispose/teardown), stale accepts inert: src/tui/runtime-bridge.ts + tests
• Geometry one-row zone, coexistence with WORKER WAITING, collapse/restore: src/tui/geometry/zones.ts + geometry/resolve.ts + geometry/resolve.test.ts
• docs/TUI.md updated for the strip and stacking: docs/TUI.md

Deferred / Notes

• Production message.received correlation confirmed in probe (deferred; the read signal uses message.run.started).
• Durable persistence is out of scope: the strip is display-only, no persistence across restart.

Refs CL-10437

[linear-code]

CL-10437

[TheGreatAxios]

Ledger lifecycle verified against the reactor — accepted-result entry, message.run.started read signal, wake-origin exclusion, and lifecycle clears all confirmed (206/0 focused suite). One item: the body is stale — base is cl-10432 (not cl-10200), the PR is one squashed commit (body says 5), focused suite reports 206 (body says 203). Refresh the body.

[bwachman]

The PR body is refreshed: base is now cl-10432 (was cl-10200), the PR is one commit (was 5), and the focused suite count is 206 pass / 0 fail (was 203), re-run at this head.