Execution Log: nesting work items and their PRs in the sidebar

Append-only log of progress for the user's visibility.

Phase transitions

Phase	Entered	Reviewed/approved by	Notes
phase-selection	2026-08-26	—	Tier 3 (human-approves-pr): one screen's structure, no contract change. Brainstorming skipped — the issue states the target shape outright
requirements-definition	2026-08-26		requirements.md — five requirements, three abuse cases, scope boundary drawn at collapse/expand
design	2026-08-26		design.md — render the existing sessionTree; move the viewed trace onto the route; three CSS rules
test-planning	2026-08-26		testing-plan.md — eleven rows, six of them applicable
tasks-breakdown	2026-08-26		tasks.md — seven tasks
implementation	2026-08-26		On claude/github-issue-300-ab9r0d
verification	2026-08-26		evidence/verification.md — 171/171, lint and build clean, screenshots captured
needs-review	2026-08-26		PR raised; awaiting the owner
complete

What was delivered

The sidebar the issue-298 redesign shipped is flat by construction, while the board's data is not: a work item's outer loop spawns a pdlc-pr-loop per pull request, each with its own session and transcript. Those sessions had no row on this surface at all — they were reachable only as unlabelled tabs inside an item's canvas.

• Two-level sidebar. Each work item's PR sessions render as nested rows beneath it (dot · #216, or loop-docs#47 when the PR is in another repository · age), inside a <ul> named for the item so the nesting is in the accessibility tree and not only in the indent. Items with no PR sessions are unchanged; loops with no outer/inner split (pdlc-adhoc-loop, pdlc-contribution-loop, pdlc-review-loop) render treeless.
• Rendered from the join that already existed. sessionTree in src/api/model.ts was written for the pre-298 Sessions screen and has been dead production code since that screen retired — its only remaining caller was its own test. It already knew which items are treeless and which endpoints hang off an item, so it gained two fields for the row (label, lastActivity) instead of a second, divergent derivation being written beside it.
• The hash became the single source of truth for the viewed trace. WorkItemDetail held it in useState seeded once from a prop — sound while the pane's own tabs were the only way to change traces, and a latent bug the moment the sidebar can select a PR too (clicking a PR of the already-open item would have changed nothing). It is now a prop derived from the route, the trace tabs are links onto the same hashes the sidebar rows use, and the four consumers that read the old state — the transcript fetch, the caption's session, the event-fallback filter and the chat bar's target — read one resolved value.
• A ref the shown item does not own falls back to that item's own session rather than requesting a transcript for another item's (abuse case A1). The service's fail-closed path resolution (issue-209) is still the enforcing boundary; this keeps the page from knocking on the door.
• Three row states, because a selected PR would otherwise leave the sidebar reading as "nothing is open" while the canvas plainly shows an item: current for the selected row at either level, and a lighter owner tint on the work item whose PR is selected.

Presentation and client-side routing only: no endpoint, request shape or response shape is touched, and src/api/client.ts, src/api/types.ts and src/demo/ are unchanged.

Verification

Full results in evidence/verification.md: 171/171 tests (168 before), oxlint --type-aware clean at the version the lockfile pins, tsc --noEmit + production build green, markdownlint clean on every doc touched, and Chromium screenshots of the nested sidebar and of the canvas following a PR row — design.uiArtifacts.screenshotEvidence requires them for a UI change.

Documentation

• docs/capabilities/control-plane.md — the two-surfaces bullet now says the sidebar nests, a new bullet states the two-level rule (labels, treeless loops, the hash as the one source of truth for the viewed trace), and a history row records the change.
• ui/README.md — the screen description, the sidebar's row in the "where the screens get their data" table, and the routing note.
• No other user-facing doc describes the sidebar's structure.

Decisions and open questions

No decision record was raised: nothing here changes a contract, a boundary or a convention, and the two judgement calls are recorded where they belong — the alternatives table in design.md (why not a count chip, why not selected-item-only nesting, why not a collapse toggle) and the Out of scope section of requirements.md.

One question for the owner at the review gate: collapse/expand was deliberately left out. Every board observed carries 1–3 PRs per item, so the list stays short and a disclosure that always starts open and is never closed is a control that pays no rent. If real boards grow deeper than that, the toggle is a small follow-up on this shape.