2.2 KiB
Sprint 178 Plan: Real Replay, Transaction Resume/Rollback, and GA Evidence
Context
docs/constructive_editing_runtime_plan.md Phase D requires replay and
transaction tooling to validate real determinism/recovery evidence.
Goals
- Replay real run traces and compare deterministic outcomes
- Resume/rollback constructive transactions against actual state transitions
- Produce GA evidence packets from real replay/transaction runs
Steps
Step 1929: Real replay suite wiring to runtime traces (10 tests)
Primary files:
editor/src/mcp/RegisterSprint153Tools.h- replay diagnostics/runtime trace readers
Tests (10): replay launch paths, trace ingestion integrity, deterministic outcome comparison, mismatch reporting, and schema-stable replay result packets.
Step 1930: Real transaction resume/rollback execution (10 tests)
Primary files:
editor/src/mcp/RegisterSprint154Tools.heditor/src/graduation/ConstructiveTransactionSchema.h
Tests (10): resume from checkpoint, rollback correctness, mixed-stage recovery, state integrity checks, deterministic rollback ordering, and failure contracts.
Step 1931: Real GA evidence packet generation (10 tests)
Primary files:
editor/src/mcp/RegisterSprint155Tools.heditor/src/graduation/ConstructiveReleaseCertificationPacketModel.heditor/src/graduation/MCPReplayDiagnosticsArtifact.h
Tests (10): evidence packet completeness, replay linkage, recovery metrics population, determinism score fields, and publication schema validation.
Step 1932: End-to-end replay + rollback scenario suite (8 tests)
Add scenarios that force divergence then verify deterministic recovery path.
Tests (8): scenario reproducibility, successful recovery evidence, blocked-case reason quality, and replay/rollback report compatibility.
Step 1933: Sprint 178 Integration Summary (8 tests)
Add editor/src/Sprint178IntegrationSummary.h.
Record:
steps_completed=5replay_suite_uses_real_runtime_traces=truetransaction_resume_rollback_is_stateful_and_real=truega_evidence_derived_from_real_runs=truesuccess=true
Architecture Gate
- GA evidence must be derived from real replay/rollback execution
- Determinism claims must be supported by reproducible trace comparisons