docs: close out terminal service-tier provenance repair

openspec/changes/fix-terminal-service-tier-provenance/tasks.md → openspec/changes/archive/2026-09-14-fix-terminal-service-tier-provenance/tasks.md

@@ -8,10 +8,10 @@
 
 - [x] 2.1 Prove conflicting, missing, and failed-terminal cases through response and request-log regressions.
 - [x] 2.2 Run focused tests, required repository validation, and independent review on the settled tree.
-- [ ] 2.3 Synchronize the normative specs and operational context, then validate and archive this change.
+- [x] 2.3 Synchronize the normative specs and operational context, then validate and archive this change.
 
 ## 3. Release
 
-- [ ] 3.1 Publish a PR targeting ericjuta/codex-lb main and satisfy current-head CI and review gates.
-- [ ] 3.2 Merge and deploy the verified revision without changing runtime settings.
-- [ ] 3.3 Verify deployment identity and readiness, then inspect aggregate normal-traffic terminal-tier evidence without model benchmarks or payload capture.
+- [x] 3.1 Publish a PR targeting ericjuta/codex-lb main and satisfy current-head CI and review gates.
+- [x] 3.2 Merge and deploy the verified revision without changing runtime settings.
+- [x] 3.3 Verify deployment identity and readiness, then inspect aggregate normal-traffic terminal-tier evidence without model benchmarks or payload capture.

openspec/specs/responses-api-compat/ops.md

@@ -130,10 +130,16 @@ Expected useful output:
 
 Interpretation:
 
-- `service_tier = "fast"`:
-  upstream entitlement exists for this account; continue to Step 3.
+Use successful-terminal `service_tier` as reported metadata, not proof of account entitlement or actual scheduling.
+
+- `service_tier = "priority"` or historical `"fast"`:
+  the response reports that tier; this alone does not prove Fast scheduling.
 - `service_tier = "default"`:
-  upstream completed normally, but this account did not receive `fast`.
+  the response reports default; this alone does not prove lack of Fast entitlement.
+- `service_tier = "auto"` or missing:
+  the tier is inconclusive; do not infer priority or default.
+- Failed or incomplete request:
+  leave the actual tier unset; only successful-terminal metadata can report it.
 - `result = "error"` with `Unsupported service_tier: fast`:
   the probe is wrong; remove raw `service_tier` from the JSON payload.
 
@@ -211,14 +217,18 @@ Dashboard shortcut:
 
 ## Result Matrix
 
+Compare the account, model, request settings, transport, and successful-terminal metadata through approved, redacted diagnostics. Do not retain account identifiers, credentials, prompts, or response bodies in reports. Unmatched runs cannot isolate proxy effects.
+
 - Direct upstream probe = `default`, `codex-lb` run = `default`:
-  proxy is behaving correctly; the account/upstream path is not yielding `fast`.
-- Direct upstream probe = `fast`, `codex-lb` run = `default`:
-  this is a real proxy regression; inspect websocket proxying and account selection.
-- Direct upstream probe = `fast`, `codex-lb` run = `fast`:
-  end-to-end support is confirmed.
+  matching metadata alone proves neither proxy correctness nor lack of Fast entitlement.
+- Direct upstream probe = `priority` or historical `fast`, `codex-lb` run = `default`:
+  reported tiers differ; compare matched runs and account selection before attributing a proxy regression.
+- Both runs report `priority` or historical `fast`:
+  both paths report priority-related metadata, not proof of Fast scheduling or end-to-end correctness.
+- Either successful terminal reports `auto` or omits the tier:
+  the tier remains inconclusive; preserve that uncertainty.
 - CLI run falls back to HTTP/SSE:
-  websocket transport regression in the proxy path.
+  fallback alone is not a transport regression; inspect handshake errors, client configuration, and upstream/proxy logs.
 
 ## Optional Deep-Dive: Capture Native `Codex CLI` Websocket Payload