fix(orb): stop a conclusion-derived verdict from overwriting a recorded close

[JSONbored]

What

Stops a gate verdict derived from the check conclusion from overwriting the verdict the bot actually acted on. Advances #8825 — this is the recording half; deciding how policy closes should be scored is the remaining part of that issue.

Why

Two callers write gate_decision, and both key the same deterministic row id (gate:<source>:<project>#<pr>@<sha>) with ON CONFLICT DO UPDATE — so the last writer wins:

• processors.ts:3278 records the actual disposition the bot acted on (action: disposition.actionClass).
• processors.ts:~10621 records a verdict derived from the gate conclusion alone, where success → merge (nativeGateActionFromConclusion).

On a PR the bot closed for a downstream reason (CI failure, contributor cap, policy), the conclusion-only writer runs last and clobbers the real close with merge. Concretely, #5861:

time	event	value
20:23:31	agent.action.close	"CI is failing (validate, validate-tests (5))"
20:23:42	pr_outcome	closed
20:23:44	gate_decision	merge / success ← 13s AFTER the close

Fleet calibration reads the latest gate_decision as the gate's prediction, so every such row is scored as "the gate said merge and it ended up closed" — a false positive that never happened. The gate decided close, and the close was correct.

Measured on the live self-host across the merge-verdict-but-closed-outcome class (n=210): 84 were bot-closed and 59 carry a verdict timestamped after the close action it contradicts. This biases published accuracy downward — the opposite direction from the reversal under-counting in #8823, which is why no accuracy figure is currently trustworthy in either direction.

How

The DO UPDATE is guarded: it skips when the incoming write is conclusion-derived (input.action absent) and the stored decision is already close. A close is a terminal action that already happened; no later conclusion can un-close it, so the older row wins.

Deliberately narrow — the guard fires only for the exact contradiction observed:

• an explicit action still replaces a stored close (a genuine later disposition change is still recorded);
• non-close rows keep latest-finalize-wins unchanged;
• no schema change, no migration, no behavior change to the gate itself.

Verification

• TSC clean; 33 parity-wire tests and 398 across the downstream queue/outcomes suites, all green.
• Three new regression tests: conclusion-derived never overwrites a close; an explicit action still does; non-close rows still update.
• Changed-line coverage: 0 uncovered statements/branches.

Note on the historical rows

This stops new contradictions. The 59 already-recorded ones still misreport; quarantining or correcting them stays tracked on #8825 alongside the policy-close scoring decision.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.28%. Comparing base (4d6584d) to head (f85006d).
⚠️ Report is 2 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8826      +/-   ##
==========================================
- Coverage   93.83%   92.28%   -1.55%     
==========================================
  Files         800      800              
  Lines       79832    79833       +1     
  Branches    24197    24198       +1     
==========================================
- Hits        74909    73677    -1232     
- Misses       3558     5092    +1534     
+ Partials     1365     1064     -301     

Flag	Coverage Δ
backend	92.96% <100.00%> (-2.16%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/review/parity-wire.ts	96.55% <100.00%> (+0.12%)	⬆️

... and 3 files with indirect coverage changes