fix: an open PR suppresses the reporter chase, which was oscillating

[johanzander]

Problem

Parking an item that has an open PR does not settle it — it oscillates.

Pass	Rule	Result
1	park — awaiting reporter, quiet 55d	move the card to Backlog
2	column derivation — an item with an open PR is In Review	move_card, put it back
3	park again	… forever

#162 did exactly this, and it was watched happening during a backlog pass: parked at 55 quiet days, immediately reported as a mis-placed card, restored, reported as needing a park again.

That is the "same conversation every tick" failure the board work (#636) exists to end — reappearing between two rules inside this script, rather than between the script and the maintainer.

Fix

.pr != null suppresses both park and nudge_reporter.

The rule is right on its own terms too, independent of the loop: an issue with a PR in flight is not a reporter chase, whatever its Awaiting field says. The work exists, and the wait that matters is the review — which the PR half of this pass already reports. Nudging the reporter of a 46-day-stale conflicted PR asks the wrong person about the wrong thing.

Evidence the test discriminates

Reverted the script to HEAD and re-ran the two new tests:

FAILED  test_an_open_pr_suppresses_the_chase_and_the_park
PASSED  test_the_chase_still_fires_without_a_pr
1 failed, 1 passed

• The first fails pre-fix and passes after — it asserts both that the chase is gone and that no move_card remains to oscillate against, at both the 14-day and 28-day thresholds.
• The second passes both before and after, deliberately. It is the guard that the suppression does not swallow the case the rule exists for: an issue with no PR still parks at 28 quiet days.

Test plan

• ./scripts/quality-check.sh green
• 45/45 in backend/tests/test_backlog_rhythm.py
• Observed live against the real board: 23 actions → 22, with no park/move_card pair left trading Add external_solar_mode battery setting for AC-coupled PV systems #162 back and forth.

Outcome-level coverage

Not applicable — no optimizer path touched. The pins above are the coverage.

Scope

scripts/backlog-rhythm.sh and its test file. No CHANGELOG entry: agent tooling, no user-visible effect.

🤖 Generated with Claude Code

[bess-agent]

@claude-bot review

[bess-manager-claude-bot]

Review: fix/park-vs-open-pr

Root cause matches the diagnosis. scripts/backlog-digest.sh:339 derives column as "In Review" whenever .pr != null. Before this fix, scripts/backlog-rhythm.sh's park/nudge rule ignored .pr entirely and fired park purely off awaiting == "reporter" + quiet days. Once park moved the card to Backlog, the next pass's board_status != .column check (scripts/backlog-rhythm.sh:146) emitted move_card back to "In Review", and the pass after that parked it again — the exact loop #162 hit. .pr != null then empty as the first branch of the if/elif chain at scripts/backlog-rhythm.sh:109 correctly suppresses both park and nudge_reporter in one guard, matching the PR description.

Independently reproduced, not just trusted. I extracted both the pre-fix (main) and fixed (origin/fix/park-vs-open-pr) versions of backlog-rhythm.sh and ran them directly against hand-built digest fixtures (bypassing pytest, since this is pure bash+jq):

• Matched-column case (column/board_status both "In Review", pr: 167, awaiting: "reporter", 28 quiet days): pre-fix → ["park"]; fixed → []. Confirms the fix.
• No-PR guard case (pr: null, same quiet days): fixed script still returns ["park"]. Confirms the suppression doesn't swallow the case the rule exists for.

This matches the PR's own "revert to HEAD, re-run" evidence and CI (all green, including Fast tests / Code quality).

Fitness of approach:

1. Best available fix, not just better-than-before — it directly encodes the real rule ("a PR in flight is not a reporter chase") rather than routing around the oscillation via e.g. an already-parked flag. No workaround shape (no new parameter/flag/second construction site) — straight condition added to the existing rule block it belongs in.
2. Holds for valid inputs generally: .pr is sourced only from gh pr list --state open (scripts/backlog-digest.sh:58), so it's null once a PR merges or closes — the suppression lifts naturally rather than persisting forever via a stale value.
3. Separation of concerns: the added branch lives inside the same "chase once, then park" jq expression it needs to gate — no responsibility leaked into an unrelated method.
4. Scope is minimal and matches the PR description exactly: 2 files (scripts/backlog-rhythm.sh +15/-1, backend/tests/test_backlog_rhythm.py +33), confirmed via gh pr view --json files. No scope creep.
5. This is clearly a local fix (one guard clause within an existing rule's existing contract), so the absence of an explicit "local vs structural" statement in the description is fine — not a structural-looking change that needed one.

Test quality: test_an_open_pr_suppresses_the_chase_and_the_park loops over both the park threshold (28d) and nudge threshold (14d) — matching the script's actual defaults (NUDGE_DAYS=14, PARK_DAYS=28 at scripts/backlog-rhythm.sh:39-40) — and asserts on the resulting action set (behavior), not implementation details. test_the_chase_still_fires_without_a_pr is the correct negative-space guard. Both would catch a regression; I confirmed this myself against the pre-fix script rather than relying on the PR's claim.

Rule violations: none found. No Optional[x], no hasattr/getattr fallbacks, no new classes, no exception-string matching, no API-layer or sensor-access code touched. Comments in the diff explain non-obvious WHY (the oscillation mechanism, tied to #162), consistent with the file's existing style.

No blockers. Approving.