Execution Log: a closed poll-only work item leaves the board by itself

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

Phase transitions

Phase	Entered	Reviewed/approved by	Notes
phase-selection	2026-09-09	—	Tier 3 (human-approves-pr; below humanSignOffMinTier: 4): a second candidate set for a question the poller already asks, under a window and a cap; one optional ledger field; no config key, schema, grant or route. Brainstorming skipped: the ticket names the options and ranks them. No authorized the-loop execute reaches this cloud session, so the selection is recorded here and every phase is walked
requirements-definition	2026-09-09		requirements.md — three requirements, four abuse cases
design	2026-09-09		design.md — the schedule, the set, the question; decision-115
test-planning	2026-09-09		testing-plan.md — fourteen rows, six applicable
tasks-breakdown	2026-09-09		tasks.md — four tasks
implementation	2026-09-09		On claude/github-issue-332-z3xktv — tasks 1–3
verification	2026-09-09		evidence/verification.md — rows T1, T3, T9, T11, T13; evidence/security-review.md — four abuse cases, four closed
needs-review	2026-09-09		PR raised; awaiting the owner (tier 3: human-approves-pr)
complete

Pull requests

PR	Scope / tasks	Status
#335	tasks 1–4: the whole work item	open

Progress entries

2026-09-09 — implemented, verified, ready for review

• Phase: implementation → verification → needs-review
• Did: tasks 1–4, red first (the fourteen new tests were run against the unchanged source and fail there — the red block in evidence/verification.md). Two constants beside _SEEN_COMMENTS_CAP; PollState.absent_since / note_closure_check and the closureCheckedAt field; PollSummary.ledger_checks on poll.cycle; _closure_candidates became _candidate_sets (tracked, ledger-only), the question became _ask_closure shared by both sets, and _ledger_candidates puts the second set on its schedule — listed, stamped and not-yet-due records out, longest-absent first — under the cap; a non-closure dates the record. The docs (state.md, concepts.md), the capability doc's rule and history row, the two catalog descriptions, and decision-115.
• Checkpoint/tests: make check — see evidence/verification.md. Existing assertions changed: one (test_a_poll_only_record_is_not_reconciled pinned issue-329's R3.2 and is now the pair test_a_poll_only_record_seen_this_window_is_not_asked / test_a_closed_ledger_only_item_is_forgotten_and_never_asked_again).
• Self-review: three passes over the diff. Pass one: _ledger_candidates re-parsed the now string it had just been handed and asserted on the result — it now takes the datetime cutoff computed once by the caller; the due-list sort would have compared WorkItemRef objects on a (impossible) tie — sorted by key now; a string tuple[...] annotation where the module uses typing — Tuple. Confirmed the design's claim that a hot reload updates config.interval_seconds in place (_maybe_reload). Pass two: the integration scenario asserted the record's key set and tripped on forget's documented "poll": null tombstone — the assertion now reads populated sections; the testing plan's trace named two tests as planned rather than as written — aligned. Pass three: nothing new. Observations, out of scope: uv.lock re-resolves on every uv run because the 13.7.1 bump did not update it (not committed here); the full suite still leaves a fixture record under .the-loop/portable/ in the checkout (the pre-existing leak issue-331's log noted) — removed before committing, not fixed here. One more, fixed here because this PR's own CI would trip on it: test_one_repository_with_issues_disabled_does_not_blind_the_others (issue-315) asserted the session record right after waiting for the tmux spawn, and the dispatcher registers after it spawns, on its own thread — 3 of 20 runs failed on the unchanged source; the assertion now waits for the record as it waited for the spawn.
• Next: the owner's review.
• Blockers: none.

2026-09-09 — spec chain drafted

• Phase: requirements-definition → tasks-breakdown
• Did: read _reconcile_closures / _closure_candidates / _process_item, PollState (finalize, baseline_comments, forget, flush), PollSummary and poll_once in poller/poller.py; Dispatcher.handle's closed branch, _tracks and _record_closure in webhook/dispatcher.py (the branch runs inline, so the stamp precedes the poller's forget); the issue-329 spec chain and decision-113; the issue-315 re-probe constant. Confirmed the ticket's account: a poll-only record is never asked, so its closure is never recorded. Chose the ticket's first option and wrote the four artifacts and the decision.
• Checkpoint/tests: baseline — test_poller.py green (203).
• Next: task 1, red first.
• Blockers: none.

Verification results

Only when this work item declared test-planning away. It did not: results live in testing-plan.md.

What was verified	Command	Outcome	Evidence
—	—	—	see testing-plan.md

Design critic review

Not selected for this work item.

Round	Critic (<harness>/<model>)	Outcome	Findings → disposition	Link

Review cycles

Cycle	Type (self/critic/security)	Reviewer	Outcome	Link
1	self (diff)	the-loop (this session)	new findings — a re-parse-and-assert, a tie-unsafe sort, a stray annotation style: fixed	this log
2	self (tests + docs)	the-loop (this session)	new findings — a tombstone-blind assertion in the scenario, two planned-vs-written test names in the plan: fixed	this log
3	self (diff)	the-loop (this session)	zero (converged)	this log
—	critic	—	unavailable — reviews.critics is empty in this repository's config; does not count toward criticReviewCount	—
4	security	the-loop checklist	pass — four abuse cases, four closed	evidence/security-review.md

Security review (gate)

• Mechanism: the-loop checklist (security.review.mechanism: auto; no security-review skill is invocable from this session's plugin set)
• Outcome: pass — evidence/security-review.md
• Human sign-off: n/a (tier 3, below security.review.humanSignOffMinTier: 4)

Final validation evidence

Requirement	Proof
R1.1, R1.2	test_a_closed_ledger_only_item_is_forgotten_and_never_asked_again, test_a_poll_only_record_seen_this_window_is_not_asked, test_a_poll_only_record_checked_this_window_is_not_asked, test_a_poll_only_record_with_an_unparsable_timestamp_is_due, test_a_listed_poll_only_record_is_not_asked, test_a_stamped_poll_only_record_is_not_asked
R1.3, R1.7	test_ledger_only_records_are_asked_longest_absent_first_up_to_the_cap, test_unowned_and_degraded_ledger_only_records_do_not_spend_the_cap; the issue-159 / issue-315 reconciliation tests unchanged
R1.4	test_a_closed_ledger_only_item_is_stamped_after_the_window (stamped ended, source: poll, ledger forgotten, record kept)
R1.5	test_a_still_open_ledger_only_item_is_dated_not_closed, test_an_unanswerable_ledger_only_item_is_dated_not_retried_next_cycle, test_poll_state_absent_since_is_the_later_timestamp, test_poll_state_note_closure_check_writes_through
R1.6	test_the_cycle_counts_ledger_checks
R2.1–R2.3	test_a_poll_only_record_beside_a_session_record_is_tracked_not_ledger_only, test_a_record_with_poll_beside_another_section_is_asked_without_a_window ×3; every issue-329 reconciliation test unchanged and green
R3.1–R3.3	docs/cli/state.md, docs/cli/concepts.md, docs/capabilities/webhook-triggers.md, decision-115, the two catalog entries (test_eventlog.py, make lint)
A1–A4	evidence/security-review.md

Capability docs

Capability doc	What changed	History row
webhook-triggers.md	the issue-329 reconciliation bullet no longer says a poll-only record is not asked; a new bullet states the lazy rule — the window on the ledger's two stamps, the cap, longest-absent first, the unchanged close path on a closure, the date-only write on a non-closure, ledger_checks	issue-332 row added at the top
control-plane.md	unchanged — the board renders the record a closure leaves (ended only) exactly as issue-329 taught it; no new element, group or rule	none
cli.md	unchanged — no command, option or host behaviour changed	none

Documentation

Document	What changed
docs/cli/state.md	the poll table gains closureCheckedAt; a paragraph on the lazy rule for a poll-only record; the delete note says what removing the date does
docs/cli/concepts.md	How a session ends: one sentence on the item the poller only ever listed
docs/decisions/decision-115.md + index row	time on the ledger rather than a counter; a constant window and cap rather than a key; the unanswerable case deferred
README.md, skills/the-loop/**, docs/config/cli/polling-options.md	unchanged — none describes closure reconciliation, and no option was added (the window and cap are constants, decision-115 D4)