# Supplied-evidence finalization reserve

<!-- auto-oncall-optimization-record
{
  "schemaVersion": "auto-oncall-optimization-record-v1",
  "optimizationId": "supplied-evidence-finalization-reserve-2026-09-02",
  "problems": [
    {
      "problemId": "supplied-evidence-consumes-downstream-synthesis-time",
      "summary": "A provider-none supplied-evidence run could finish evidence acquisition without an independently allocated Scenario Diagnostic finalization stage, leaving Router reconciliation and Final Response unable to complete.",
      "status": "candidate",
      "applicability": "Diagnostic Scenario runs that use ticket-supplied evidence with evidenceProvider=none.",
      "issueKeys": ["APID-29394"],
      "sourceRefs": [
        "https://tidb.atlassian.net/browse/APID-29394#problem:current-supplied-evidence-finalization-budget"
      ],
      "signalTextIncludesAny": [
        "supplied evidence finalization budget",
        "diagnostic_router_reconciliation_failed",
        "APID-29394"
      ],
      "observedThrough": "2026-09-02",
      "effectiveAt": null,
      "verifiedAt": null,
      "attemptedAction": "Allocate supplied-evidence acquisition and Scenario Diagnostic finalization as separate bounded stages while retaining the existing Router reconciliation and Final Response reserves.",
      "evidenceUrl": "https://tidb.atlassian.net/browse/APID-29394"
    }
  ]
}
-->

- Owner: Auto-oncall diagnostic runner
- Source: APID-29394 selected single-ticket finding
  `current-supplied-evidence-finalization-budget`
- Implementation status: local repair and deterministic regression test
- Effective status: `candidate`

## Bounded problem

The supplied-evidence Evidence Agent completed, but its Scenario Diagnostic
finalization shared the acquisition stage's expiring signal. The fresh replay
then failed finalization, Router reconciliation, and Final Response in sequence
and returned the generic deterministic fallback instead of consuming the
materialized facts.

## Intended change

Reserve the supplied-evidence Scenario Diagnostic finalization before starting
acquisition, allocate acquisition and finalization as separate stages from the
same absolute pipeline budget, and preserve the existing independent Router and
Final Response reserves. Provider-backed scenarios and all semantic, safety,
publication, and customer-write boundaries remain unchanged.

## Verification

The runner regression verifies that supplied-evidence acquisition, Scenario
Diagnostic finalization, Router reconciliation, and Final Response each receive
a positive timeout and that the completed result is not the generic Router
fallback. Deployment and APID-29394 post-change replay are not part of this
repair Agent's authority.

## Unresolved gaps

The exact Agent9 error beneath the historical finalization failure was not
retained. A staging replay is still required to prove that the bounded runtime
criterion passes with the live backend.

## Next validation action

After the orchestrator merges and deploys the change, run the APID-29394
read-only replay and verify normal Scenario Diagnostic, Router, and Final
Response completion before any authorized Jira refresh.