[zzet]

The edit-and-probe harness now covers explicit Go, TypeScript and Python fixtures for append, body/caller/signature changes, intrafile/crossfile calls, rename and symbol deletion through applicable file/write/symbol/batch adapters. Changed-fact probes distinguish positive evidence from exact scoped absence, preserve visibility before ACK, and require complete nontruncated relation/search evidence. Unsupported combinations refuse explicitly.

Go relocation fixes a real declaration-separator splice defect and observes existing writes/reconciliation without changing durability or scheduling. Shared RPC A, each original file's write window, and actual direct/read-assisted publication windows remain separate. Exact coalesced execution views count once with both request causes retained; no-op work cannot donate phases. Immediate own-ACK readbacks precede status polling, and a completed move restores both files through one public batch.

The first live relocation diagnostic at 4b3f5b3ef failed honestly: inline suggestions did not match the absence parser, and the target's actual publication window was unavailable. That raw result remains failed. The changed-source R2 at 893289137 passed one native functional sample: all eight oracles and four readbacks, original-file ownership/accounting, committed two-file restore and normal shutdown. It observed direct publication for both files; read-assisted/coalesced support has deterministic real-executor controls, not live R2 coverage. Shared A of 46.117 ms and direct windows of 18.868/20.503 ms partitioned with zero residual; these are descriptive one-sample measurements, not performance acceptance.

Validation: 194 focused bench/MCP race records at 4b5f492d1, including actual reader-before-direct and coalesced coordinator executions; 722 whole-bench race records and optional-only schema updates at 893289137. Final head 41d825744 passes all prescribed pinned pre-push checks: uncapped golangci-lint 2.14 (zero issues), 561 affected semantic/capability race records with the required modernc exclusions, vet, whole-branch formatting/diff checks, safe CGO/workflow guards, module tidy/verification and clean status. The final delta from measured 893289137 is equivalent lint/format-only correction; R2 remains attributed to its actual artifact/source. Exact validation is recorded in the locally retained FINAL-PREPUSH-RECEIPT.json; that receipt is not attached to this PR.

This remains an incomplete acceptance PR. Native lifecycle, output-mode and traffic routing implementation/coverage, reference 100 per traffic configuration, the broader applicable matrix and paired no-regression remain open. Historical failed measurements keep their exact source/result qualifications; no unsupported cell is waived and no benchmark corpus is committed.

Refs #1005; keep issue open.