# Single Adaptive Refinement

<!-- auto-oncall-optimization-record
{
  "schemaVersion": "auto-oncall-optimization-record-v1",
  "optimizationId": "single-adaptive-refinement-2026-08-12",
  "problems": [
    {
      "problemId": "adaptive-refinement-deepens-bounded-root-cause-until-timeout",
      "summary": "Wetech scenarios consumed serialized provider-wait budget while using third and fourth rounds to deepen an already bounded diagnosis, producing misleading generic partial failures.",
      "status": "optimized_verified",
      "issueKeys": ["APID-29259", "APID-29253"],
      "sourceRefs": [],
      "signalTextIncludesAny": ["refinement timeout", "adaptive batch", "scenario deadline"],
      "observedThrough": "2026-08-12",
      "effectiveAt": "2026-08-12T06:49:36Z",
      "verifiedAt": "2026-08-12T07:45:00Z",
      "attemptedAction": "Allow only the initial batch plus one load-bearing adaptive refinement, preserve exact refinement failure codes, and report partial evidence as degraded rather than fully successful or wholly missing.",
      "evidenceUrl": "https://github.com/tidbcloud/aisupport/tree/main/auto-oncall-services/docs/opts/2026-08-12-single-adaptive-refinement.md"
    }
  ]
}
-->

- Owner: auto-oncall-services
- Sources: APID-29259 and APID-29253
- Status: optimized_verified

Adaptive diagnosis is limited to one refinement after the initial batch;
same-round plan repair and transport retries remain available. Exact
later-round failures stay attached to prior validated evidence. Deployment and
APID-29259 replay `669ba399-75a4-4818-bb54-739e5df24efc` completed with only
rounds 1 and 2 across three scenarios. No round-3 or round-4 plan, provider
batch, or Scenario Diagnostic turn was created. Partial provider failures were
reported as `degraded` with exact failure codes instead of generic total loss.