Cover private goal continuations through shared scenarios

Draft
VA
18h ago
+1381-448

Draft after post-#955 validation. #955 is merged and signed local integrations are prepared (600:597da9578, 700:c0c4fea5f), but not pushed. The complete39-case parent goal family still has37 passes and2 unchanged5s deadline failures (default-at/default-before,5.102s). Isolated profiles use~4.9 CPU-seconds/5.1s with continuous settles and growing per-turn cost. Remaining co-perf-h history/table/publication prerequisites, including observer-prefix PR #985, must land before re-review. Full e2e and fresh Windows checks stopped at the family failure; neither is claimed green. Workloads and guards are preserved.

Problem

The private-goal cases duplicated the LocalHarness runner and used an obsolete output-volume override. The shared runner could not preserve their completed project-GOAL decoy or independently assert project and private final files.

Design

Stack on #600 and express all six original write histories as small data cases using the shared goal-local driver. Run production request-file and continuation admission, task-status snapshot/follow and scheduler wire frames. Preserve original policies, task/request counts, exact final private bytes and the unchanged completed project goal. Verify both volumes before and after cold reopen. The schema rejects a seeded project file without a final expectation; corruption/omission controls exercise the same oracle.

Deletions

Remove the duplicate private catalog/runner and output-volume override. Relative to parent #600:12 files,+81/-16 text lines,net+65,plus the existing1,251-byte fuzz seed. Stock goal-library regressions remain. The authority retry fix is inherited once from #600, not duplicated here.

Decisions

Signed-merge the reviewed parent fixes; do not rebase published history or add another model fake. Keep all six scenarios, their original workloads and five-second guards. No new production API, lock, store, policy or protocol. Keep #600/#700 as the active review pair; #614/#649 remain draft and set aside under WIP2.

#955 owns incremental closure proof and co-perf-h owns remaining segment/table/object-publication costs. No overlapping changes are included. #955 is merged. Its production change is included in the signed local integration, but the complete parent goal gate still fails both default-32 deadlines. Remaining co-perf-h prerequisites include #985; the private-only gate cannot establish a green full target.

Tests run

Signed head5716d2441e52b20d9341e12287714d6ec68074da, one bb1 gate: Q:/lanes/bs.ps1 python3 -c (Get-Content -Raw Q:/lanes/gc-e2e-private-ready-gate.py), job1792284651726-1928739,terminal0. Native build132.277s and Windows compile95.691s are excluded from runtime.

  • Six goal codec/boundary properties, checked-in fuzz corpus and oracle controls PASS3.497s.
  • All six private LocalHarness scenarios PASS1.548s, including completed project decoys and cold-reopen file assertions.
  • Windows GNU Harness lib/tests check PASS95.691s. Canonical six features retained; parent #600's identical filesystem code has its separate green GNU proof.
  • Shared inherited retry fix: real-lock Loom3 PASS0.0507s (two actors,preemption2,unchanged branch bounds,broken-key control), slot properties2,authority regressions8 and native lost-ack1 pass. See #600 for the full independent non-goal results and both passing/failing parallel-load observations.

Artifact /home/var/targets/gc-e2e-da7560f9/e2e-private-ready-i8myhlca holds the binary and per-group logs. Native command: cargo test --locked -p acyclic-fs -p acyclic-harness --lib --test e2e --features acyclic-harness/filesystem-local,acyclic-harness/native-execution,acyclic-harness/mcp-http,acyclic-harness/grpc,acyclic-harness/terminal,acyclic-harness/search-http --no-run; run the copied binary with goals_tests::, then E2E_FILTER=/scenarios/goals/private/ and scenarios --exact. Cross-check: cargo check --locked -p acyclic-harness --lib --tests with the same feature closure and --target x86_64-pc-windows-gnu. Fuzz engine targets are Linux-only; scenario codec/target/corpus are unchanged from their prior pinned-nightly proof. No whole-suite elapsed assertion, raised guard or reduced coverage.