reacher-z · reacher-z · Jul 26, 2026 · Jul 26, 2026
@@ -98,7 +98,7 @@ jobs:
       - run: uv run --project python --python ${{ matrix.python }} pytest
       - run: uv run --project python --python ${{ matrix.python }} python examples/quickstart/run.py
       - run: uv run --project python --python ${{ matrix.python }} ruff check python/src python/tests
-      - run: uv run --project python --python ${{ matrix.python }} mypy python/src
+      - run: uv run --project python --python ${{ matrix.python }} mypy --config-file python/pyproject.toml python/src
       - if: matrix.python == '3.13'
         run: uv build --project python
       - if: matrix.python == '3.13'

@@ -6,9 +6,25 @@ migration note.
 
 ## [Unreleased]
 
+### Added
+
+- Event-sourced durable start and resume APIs in TypeScript and Python. Runs bind
+  graph, original input, and caller-supplied implementation identity; node
+  outcomes commit before releasing dependants; committed successes are reused
+  after process loss.
+- Cross-language tagged Durable JSON for exact finite binary64 payload hashing,
+  stable activity/idempotency keys, terminal-result snapshots, and recovery
+  conformance fixtures.
+- Fail-closed interrupted-attempt handling: nodes declared with no or idempotent
+  side effects may retry within their original budgets, while omitted or
+  non-idempotent declarations surface `IN_DOUBT_SIDE_EFFECT`. Resuming a valid
+  terminal run returns its recorded result with no new event or executor call.
+
 ### In progress
 
-- Scheduler-integrated event emission, checkpoint recovery, replay, and fork.
+- Scheduler checkpoint acceleration, replay/fork, and distributed lease/fencing
+  providers. Recovery correctness currently comes from the complete event
+  stream; checkpoint files are not wired into the scheduler.
 - Streaming pipelines, conditional edge lowering, verifier panels, and bounded
   runtime loops.
 

@@ -28,7 +28,7 @@ Python uses uv:
 uv sync --project python --extra dev
 uv run --project python pytest python/tests
 uv run --project python ruff check python/src python/tests
-uv run --project python mypy python/src
+uv run --project python mypy --config-file python/pyproject.toml python/src
 ```
 
 Changes affecting the Graph IR, compiler diagnostics, event protocol, or

@@ -28,9 +28,11 @@ The project is an early alpha. The DAG compiler, ready-queue schedulers, safe pr
 initializer, machine-readable CLI, structured failure handling, retries,
 timeouts, budgets, settled barrier/router decisions, safe Mermaid/DOT rendering,
 pattern constructors, local event/checkpoint stores, read-only MCP server, and
-development progress scanner are executable today. Scheduler-integrated durable
-recovery and the broader v1 surface remain under active development; the
-repository does not silently mock unfinished capabilities.
+development progress scanner are executable today. Both native runtimes also
+provide event-sourced durable start/resume: committed successes are reused after
+process loss and unsafe ambiguous effects fail closed. Checkpoint acceleration,
+replay/fork, distributed leases, and the broader v1 surface remain under active
+development; the repository does not silently mock unfinished capabilities.
 
 > **Source-only alpha:** npm and PyPI packages are not published yet. Clone this
 > repository to try the current release candidate; registry publication remains
@@ -95,15 +97,17 @@ uv run --project python pytest python/tests
 | Diamond/verifier pattern constructors | Yes | Consumes the portable Graph IR |
 | Safe Mermaid/DOT visualization | Yes, through the CLI | Same portable Graph IR |
 | Local event/checkpoint stores | Yes | Yes |
-| Scheduler checkpoint/resume | In development | In development |
+| Event-sourced scheduler start/resume | Yes | Yes |
+| Scheduler checkpoint acceleration | Not yet | Not yet |
 | Read-only validation/planning MCP | Yes | Uses the same portable IR |
 | Streaming and scheduler-applied routers/verifier panels/loops | Target v1 | Target v1 |
 
 ## Design commitments
 
 - Explicit node and edge data contracts.
 - Parallel, pipeline, barrier, router, verifier, and bounded-loop topologies.
-- Durable checkpoints, resume, replay, and fork.
+- Durable event-sourced start/resume today; rebuildable checkpoint acceleration,
+  replay, and fork as explicit follow-up protocols.
 - Deterministic plumbing; models are reserved for judgment.
 - Provider-neutral adapters and deny-by-default capabilities.
 - Observable runs with portable events and traces.
@@ -123,7 +127,7 @@ corepack pnpm check:packages
 corepack pnpm check:packed-install
 corepack pnpm audit:prod
 uv run --project python ruff check python/src python/tests
-uv run --project python mypy python/src
+uv run --project python mypy --config-file python/pyproject.toml python/src
 uv build --project python
 python3 scripts/check-python-artifacts.py
 ```
@@ -139,6 +143,7 @@ python3 scripts/check-python-artifacts.py
 - [Support](SUPPORT.md)
 - [Runtime semantics](spec/runtime-semantics.md)
 - [Persistence semantics](spec/persistence-semantics.md)
+- [Durable recovery semantics](spec/durable-recovery-semantics.md)
 - [Primitive semantics](spec/primitives-semantics.md)
 - [21-day delivery plan](codex_plans/Graph-Engineering-21-Day-Master-Plan.md)
 

@@ -18,15 +18,19 @@ earned by executable cross-language tests.
 
 ## Alpha 2 — durable graph patterns
 
-- Scheduler event emission and checkpoint-safe node commit.
-- Crash recovery that reuses completed node results.
-- Replay and fork with immutable lineage.
-- Pipeline buffers/backpressure and explicit barrier policies.
-- Deterministic routers, verifier verdicts, quorum/unknown outcomes, reflection,
+- [x] Scheduler event emission with commit-before-release node outcomes and
+  tagged Durable JSON payloads.
+- [x] Event-sourced crash continuation that reuses committed successes, preserves
+  attempt budgets, and fails closed for ambiguous unsafe effects.
+- [ ] Scheduler checkpoint acceleration; correctness already rebuilds from the
+  authoritative event history.
+- [ ] Replay and fork with immutable lineage.
+- [ ] Pipeline buffers/backpressure and explicit barrier policies.
+- [ ] Deterministic routers, verifier verdicts, quorum/unknown outcomes, reflection,
   and bounded loop-until-dry primitives.
-- Deterministic mock, OpenAI, Anthropic, Gemini, OpenAI-compatible, HTTP, shell,
+- [ ] Deterministic mock, OpenAI, Anthropic, Gemini, OpenAI-compatible, HTTP, shell,
   and MCP adapters behind explicit capability declarations.
-- Ten executable TypeScript/Python patterns with failure and recovery fixtures.
+- [ ] Ten executable TypeScript/Python patterns with failure and recovery fixtures.
 
 ## Beta — operations and external validation
 

@@ -147,3 +147,107 @@ Canonical IR -> language models/builders -> compiler -> deterministic scheduler
   code-review/security graphs, ready-queue benchmarking, and honest adopter stories.
 - Verified anonymously that the repository, release, discussions, issue chooser,
   private vulnerability report form, and raw README each return HTTP 200.
+
+## 12:42 UTC durable-recovery design checkpoint
+
+- Confirmed that both CI and CodeQL passed again for the public-release log
+  commit `d242b65`; the protected public alpha is green before feature work.
+- Created branch `feat/durable-recovery` and ran three parallel read-only audits
+  over the TypeScript scheduler, Python scheduler, persistence stores, recovery
+  state machine, crash windows, and shared conformance requirements.
+- Froze the first scheduler-integrated recovery contract around an authoritative
+  CAS event history, commit-before-dependent-release, immutable graph/input/
+  implementation bindings, stable activity keys, terminal idempotence, and
+  conservative handling of unknown external effects. Checkpoints remain
+  rebuildable acceleration rather than a second source of truth.
+- Added a shared diamond recovery case in which one committed branch is reused,
+  one interrupted safe branch advances to attempt two, the merge receives both
+  outputs, total attempt budgets survive restart, and a second terminal resume
+  produces zero events and zero executor calls.
+- Started parallel native implementation lanes and a separate exact Tagged
+  Durable JSON vector/schema lane. Tagged values preserve finite binary64 data
+  across TypeScript/Python while staying inside checkpoint safe-integer rules.
+- Registry publication remains deferred. This feature will be committed as
+  `reacher-z <mtrxcop@gmail.com>` without co-author trailers only after focused,
+  cross-language, full-suite, packaging, and security validation passes.
+
+## 13:41 UTC durable-recovery adversarial checkpoint
+
+- Independently reran the Python lane after its first implementation: 419 tests,
+  Ruff, strict mypy, and diff hygiene passed. Core and persistence remain green
+  at 36 and 27 tests, and all 14 Tagged Durable JSON vectors still validate.
+- Two independent read-only audits then found actionable protocol defects before
+  commit: zero-attempt node outcomes existed only in a terminal snapshot;
+  Python could trust a shallowly frozen compiled graph; normal retry admission
+  ignored the global attempt budget; and malformed retry timestamps could append
+  `RunResumed` before failing validation.
+- Froze `NodeSettledWithoutAttempt` as an explicit event-derived outcome for
+  missing-executor, binding, upstream, budget, and pre-dispatch cancellation
+  paths. Both schedulers must commit it before releasing dependants, and terminal
+  folding must derive every node result and both orders from history.
+- Hardened the TypeScript scheduler so a journal failure is not held hostage by
+  an executor that ignores cancellation, and public `runGraph` now runtime-
+  whitelists options so JavaScript callers cannot inject internal journals or
+  recovery seeds. A dedicated scheduler lane is adding atomic global retry
+  reservations and settlement hooks with concurrency tests.
+- Added safe-integer limits to the shared Graph IR schema and TypeScript shape
+  validation, expanded the portable event vocabulary, and started a
+  cross-language durable report. Its Python side already proves committed-left
+  reuse, interrupted-right attempt advancement, stable activity keys, merge
+  input parity, four total attempts, and terminal zero-event/zero-executor
+  idempotence.
+- The thirty-minute timer remains active. A manual scan after registering the
+  conformance lane reports 30 tasks: 29 healthy and one waiting only on the two
+  active native implementations; no stale, blocked, warning, or integration-risk
+  item was reported.
+
+## 14:49 UTC durable-recovery completion checkpoint
+
+- Completed native event-sourced `start` and `resume` APIs in TypeScript and
+  Python. The append-only CAS event stream is authoritative; graph, input, and
+  implementation identities are bound at creation; attempt claims commit before
+  executor work; outcomes commit before dependants are released; and a valid
+  terminal resume performs zero writes and zero executor calls.
+- Added exact Tagged Durable JSON for finite binary64 values, stable activity
+  keys, strict real-calendar RFC 3339 timestamps, unique per-run event IDs,
+  explicit zero-attempt settlements, portable failure identities, 32-bit host
+  timer bounds, and fail-closed recovery for ambiguous unsafe side effects.
+- Closed adversarial state-machine cases covering duplicate or stale starts,
+  pending retry reservations, global attempt races, premature downstream
+  settlements, prototype-named JSON keys, mutable executor registries and graph
+  views, event factory/clock failures, completion commit order, and non-cooperative
+  Python handlers that swallow cancellation after a durability failure.
+- Expanded conformance from Python-produced interrupted history resumed by
+  TypeScript to bidirectional terminal-history consumption for missing executors
+  and output-binding failures. Each cross-language terminal replay reconstructs
+  the recorded result with zero new events and zero executor calls.
+- Final local release gates are green: Core 37 tests, Runtime 81 tests, Python
+  486 tests, all workspace tests, build/typecheck/lint, strict mypy, both Ruff
+  configurations, eight compiler fixtures, runtime/primitives/persistence and
+  durable cross-language conformance, seven npm tarballs and clean installs,
+  Python wheel/sdist audits, source and isolated quickstarts, documentation links,
+  fixture validation, production dependency audit, and diff hygiene.
+- Synchronized the canonical Graph Schema into CLI and MCP packaged assets and
+  corrected CI/documented mypy commands so they explicitly load the Python
+  strict configuration. The 07:40 automatic progress scan reported every
+  completed lane healthy and the remaining conformance lane recently active;
+  the post-gate manual scan then reported 30 of 30 tasks healthy with zero
+  waiting, stale, blocked, warning, or integration-risk items.
+- Distributed leases/fencing, checkpoint acceleration, replay/fork, approval
+  callbacks, and universal exactly-once effects remain deliberately unclaimed
+  follow-up work. Registry publication remains deferred.
+
+## 14:55 UTC durable-recovery delivery checkpoint
+
+- Committed the integrated feature as
+  `854b2e3a35e36c2f41a82704485934a96cce644c` with author and committer
+  `reacher-z <mtrxcop@gmail.com>` and no co-author trailer, then pushed
+  `feat/durable-recovery` to the existing GitHub remote.
+- Opened draft pull request
+  [#14](https://github.com/reacher-z/GraphEngineering/pull/14) against protected
+  `main`. The GitHub App lacked collaborator authority for PR creation, so the
+  authenticated `gh` fallback completed the same draft-only operation.
+- Every reported remote CI, package, cross-language, progress-scanner,
+  dependency-review, and Python/JavaScript CodeQL check passed. The pull request
+  remains intentionally unmerged and review-required; protected-branch policy
+  was not bypassed.
@@ -1,6 +1,6 @@
 {
   "schema_version": 1,
-  "updated_at": "2026-07-26T12:20:00Z",
+  "updated_at": "2026-07-26T13:40:44Z",
   "repository": ".",
   "scan_policy": {
     "warning_after_minutes": 60,
@@ -428,6 +428,102 @@
       "risk": "medium",
       "blocker": null,
       "next_action": "Use the 9,416+ moving parity benchmark, finish GitHub launch settings, and execute the adoption/content backlog without promising stars"
+    },
+    {
+      "id": "D6-DURABLE-SPEC-010",
+      "title": "Cross-language durable scheduler recovery contract and fixtures",
+      "owner": "main + /root/durable_contract",
+      "status": "completed",
+      "work_package": "WP3",
+      "depends_on": ["D1-SPEC-001", "D2-TS-PERSIST-001", "D2-PY-PERSIST-001"],
+      "expected_artifacts": [
+        "spec/durable-recovery-semantics.md",
+        "spec/durable-json.schema.json",
+        "spec/conformance/durable-json.case.json",
+        "spec/conformance/durable-resume.case.json"
+      ],
+      "expected_tests": [
+        "tagged Durable JSON hash parity",
+        "committed-success reuse",
+        "interrupted-attempt recovery",
+        "terminal resume idempotence"
+      ],
+      "assigned_at": "2026-07-26T12:23:00Z",
+      "started_at": "2026-07-26T12:23:00Z",
+      "last_heartbeat": "2026-07-26T13:40:44Z",
+      "risk": "high",
+      "blocker": null,
+      "next_action": "Hold the contract stable while both native implementations and the shared recovery harness are verified"
+    },
+    {
+      "id": "D6-TS-DURABLE-010",
+      "title": "TypeScript event-sourced scheduler start and resume",
+      "owner": "/root/durable_contract (handoff from /root/durable_ts)",
+      "status": "completed",
+      "work_package": "WP3",
+      "depends_on": ["D6-DURABLE-SPEC-010", "D3-TS-RUNTIME-002", "D2-TS-PERSIST-001"],
+      "expected_artifacts": ["packages/runtime"],
+      "expected_tests": [
+        "commit-before-release",
+        "successful node never reruns",
+        "attempt budgets survive crash",
+        "identity/history/CAS rejection",
+        "terminal idempotence"
+      ],
+      "assigned_at": "2026-07-26T12:42:00Z",
+      "started_at": "2026-07-26T12:42:00Z",
+      "last_heartbeat": "2026-07-26T14:39:43Z",
+      "risk": "high",
+      "blocker": null,
+      "next_action": "Maintain the 81-test durable runtime contract while shared release and package gates run"
+    },
+    {
+      "id": "D6-PY-DURABLE-010",
+      "title": "Python event-sourced scheduler start and resume",
+      "owner": "/root/durable_py",
+      "status": "completed",
+      "work_package": "WP3",
+      "depends_on": ["D6-DURABLE-SPEC-010", "D3-PY-RUNTIME-002", "D2-PY-PERSIST-001"],
+      "expected_artifacts": ["python/src/graph_engineering"],
+      "expected_tests": [
+        "commit-before-release",
+        "successful node never reruns",
+        "attempt budgets survive crash",
+        "identity/history/CAS rejection",
+        "terminal idempotence"
+      ],
+      "assigned_at": "2026-07-26T12:42:00Z",
+      "started_at": "2026-07-26T12:42:00Z",
+      "last_heartbeat": "2026-07-26T14:39:43Z",
+      "risk": "high",
+      "blocker": null,
+      "next_action": "Maintain the 486-test durable runtime contract as the branch enters CI review"
+    },
+    {
+      "id": "D6-DURABLE-CONFORMANCE-011",
+      "title": "Cross-language crash/resume and adversarial history conformance",
+      "owner": "main",
+      "status": "completed",
+      "work_package": "WP3-WP11",
+      "depends_on": ["D6-TS-DURABLE-010", "D6-PY-DURABLE-010"],
+      "expected_artifacts": [
+        "tools/conformance/python_durable_report.py",
+        "tools/conformance/python_durable_interop.py",
+        "tools/conformance/run.mjs"
+      ],
+      "expected_tests": [
+        "Python-created crash history resumes in TypeScript",
+        "shared result and event sequence parity",
+        "terminal zero-event zero-executor resume",
+        "resigned history rejection",
+        "full repository and package gates"
+      ],
+      "assigned_at": "2026-07-26T13:28:00Z",
+      "started_at": "2026-07-26T13:28:00Z",
+      "last_heartbeat": "2026-07-26T14:49:21Z",
+      "risk": "high",
+      "blocker": null,
+      "next_action": "Monitor protected-branch CI and review feedback without weakening the recovery contract"
     }
   ]
 }