docs(gap): record current exact-head remediation state

docs/product-technical-gap-baseline.md

@@ -1,5 +1,40 @@
 # Product and Technical Gap Baseline
 
+## 2026-08-28 13:27 KST exact-head PR remediation slice
+
+Protected `main` is `8d5924f8`. PR #879 is now at `2b1829a8`; its exact tree
+passed **2516 tests** after fixing ambiguous classified passthrough failures so
+status-less connection errors cannot replay a possibly accepted request. Its
+Full unit and contract suite passed, but Strix failed closed on the provider
+outage described below. PR #868 is now at exact head `3f575a64` after rejecting
+whitespace-only embeddings models; its targeted 27-test proof and interrogate
+100% passed, while hosted Checks are re-running. A local full run passed 2546
+tests and exposed only the optional MCP test's environment API mismatch
+(`mcp==1.23.3` exports no `Client`), which is not in the project lock files.
+PR #857 is at externally advanced head `b2aedb7b` with all other checks passing
+but the same Strix provider outage. PR #901 is now at externally advanced head
+`d1bd3626ddb04a7b14e43aebf60827ac50ef8d17` and its code checks pass, while Noema, OpenCode, and Strix remain
+failed or unavailable gates. PR #902 is a stacked change on #857 at exact head
+`eee5323f`; its merge state is `UNSTABLE`, Devin review is pending, and its own
+body requires #857 to land first. PR #903 adds the governed CEFR
+language-observation slice at exact
+head `f456393e`; its local exact tree passed 2492 tests with 100% interrogate
+coverage, and its ordinary code/security/coverage checks plus Devin passed.
+Noema and Strix failed before producing a structured review because their
+sidecar catalog loader still expects a mapping while the generated catalog is
+list-shaped; OpenCode failed closed because no current-head reviewer verdict
+was posted. Full unit and Atheris subsequently passed on the same exact head.
+All seven open PRs remain subject to protected review and hosted gates; no PR
+is claimed as protected-main delivery. The docs-only recording PR #900 remains
+open.
+
+The earlier local merge attempt for #857 at `0aec4a54` was aborted without a
+push; no stale branch evidence is promoted to protected-main evidence. The
+docs-only recording PR #900 remains open and is not claimed as protected-main
+delivery.
+The existing hourly OpenCode maintenance workflow remains the single
+scheduled loop (`23 * * * *` UTC); no duplicate scheduler was added.
+
 ## 2026-08-27 20:10 KST main trace-rpds regression slice
 
 Protected `main` briefly carried a merge-order regression from PR #891 merged