FE-738: Petri semantic lanes with two-lane subnet and engine factory by kostandinang · Pull Request #148 · hashintel/brunch

kostandinang · 2026-05-21T18:34:49Z

Summary

Introduces the first Petri slice model where mechanical test success and semantic evaluator satisfaction are separate lanes; a slice completes only when both gates clear.
Adds structured orchestration events and transition metadata so observers can reason about firing without parsing ad-hoc logs.
Splits net compilation into topology blueprint vs handler wiring, and centralizes engine construction behind a single factory so proc/petri paths do not diverge on every frontier.

Context

FE-730 validated cook on a greenfield fixture with a minimal Petri loop. This PR is the substrate step toward richer plans where “tests green” and “evaluator says done” are intentionally different concerns.
Semantic rework is bounded via a dedicated budget place and RunPolicy.maxSemanticReworks, so assess/eval cycles cannot run unbounded when the evaluator never marks a slice satisfied.
Unblocks later work on parallel firing (FE-743) and graph-level analysis by making transition contracts and events first-class.

What changed

Two-lane subnet per slice: mechanical path evaluate → write-tests → write-code → run-tests; semantic path assess-semantic → semantic-satisfied; completion requires progress on both lanes.
TransitionContract metadata on transitions plus §7 event vocabulary (transition_fired, net_halted, net_deadlocked) with enough structure for contract tests and future tooling.
HandlerDescriptor union and typed handler wiring at the subnet boundary instead of growing ad-hoc closure dispatch.
Compiler split: compileTopology produces NetBlueprint; wireHandlers materializes PetriNet; createOrchestrator(policy) replaces duplicated orchestrator setup.

Verification

Full repo gate: npm run verify (fmt/lint check, test, build).
Engine contract suite extended for dual-lane semantics, semantic budget exhaustion, and structured events (1378 tests at PR time).

Out of scope

Stale-graph routing and full graph compilation/reachability (deferred to Phase 3; pairs with declarative routing in FE-747 and broader FE-700 work).
Parallel slice execution and resource pools (FE-743).

Traceability

Requirements 46–50; SPEC §2, §4, §6, §7, §8, §10; frontier petri-semantic-lanes in memory/PLAN.md; umbrella H-6476.

kostandinang · 2026-05-21T18:35:04Z

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

This stack of pull requests is managed by Graphite. Learn more about stacking.

cursor · 2026-05-22T12:38:22Z

PR Summary

Medium Risk
Changes core orchestration topology and completion gates (slice cannot complete without semantic satisfaction); engine/proc duality is removed in favor of one serial path—behavior shifts are covered by expanded contract tests but cook CLI flags changed.

Overview
Petri semantic lanes (FE-738) split mechanical TDD completion from semantic acceptance in the orchestrator subnet, and refactor how nets are built and run.

Each slice now has a mechanical lane (evaluate → write-tests → write-code → run-tests → done-spec) and a semantic lane (assess-semantic after done-spec, then semantic-satisfied). return-done only fires on semantic-satisfied, so tests passing alone cannot finish a slice. Failed semantic checks route back to needs-more (extra TDD cycles); maxSemanticReworks on RunPolicy and in-net semantic-budget tokens cap infinite rework.

Compiler refactor: compileTopology → pure NetBlueprint with HandlerDescriptor recipes and TransitionContract metadata; wireHandlers attaches runtime closures. createOrchestrator(policy) replaces duplicate PetriOrchestrator / ProceduralOrchestrator and engine-run.ts. brunch cook uses --policy=serial instead of --engine=proc|petri. RunCtx moves to types.ts.

Interpreter: optional NetEventSink emits §7-style transition_fired, net_halted, and net_deadlocked (with benign residual places so clean completion is not misclassified). Contract tests cover semantic rework and event capture; pi-actions adds a POC assess-semantic that auto-satisfies.

Planning docs: memory/CARDS.md removed; memory/PLAN.md marks petri-semantic-lanes done and defers stale-graph work to petri-graph-compilation.

^{Reviewed by Cursor Bugbot for commit 78706a2. Bugbot is set up for automated code reviews on this repo. Configure here.}

augmentcode · 2026-05-22T12:42:03Z

🤖 Augment PR Summary

Summary: This PR evolves the Petri-net orchestrator by introducing distinct mechanical vs semantic completion lanes, and refactors compilation/execution so the net shape is testable as pure data.

Changes:

Introduces a two-lane slice subnet (mechanical TDD loop + semantic assessment) with a completion gate so slices can’t finish without semantic satisfaction.
Adds TransitionContract metadata to transitions (kind/lane/actor/guard) to classify execution without parsing IDs.
Splits the compiler into compileTopology(plan, policy) → NetBlueprint (pure topology) and wireHandlers(...) (runtime closures), plus a convenience compilePlan wrapper.
Consolidates engine entrypoints into createOrchestrator(firingPolicy), removing duplicated Petri/procedural engine classes.
Adds semantic rework budgeting via a semantic-budget token (reworkCount) and RunPolicy.maxSemanticReworks to prevent infinite semantic-rejection loops.
Defines §7-style interpreter events (transition_fired, net_halted, net_deadlocked) and updates adapter tests to assert event emission.
Updates engine contract tests to cover semantic-gate rework and semantic budget exhaustion scenarios.

Technical Notes: Topology-only adapter tests no longer require action handlers/test runners, and runtime wiring now interprets declarative handler descriptors from the blueprint.

_{🤖 Was this summary useful? React with 👍 or 👎}

augmentcode

Review completed. 2 suggestions posted.

Comment augment review to trigger a new review at any time.

- Add semantic-gate and semantic-satisfied places per slice - Add assess-semantic transition joining done-spec + semantic-gate - On pass: produces semantic-satisfied → return-done proceeds - On fail: routes to needs-more, forcing another TDD rework cycle - return-done now consumes semantic-satisfied (not done-spec directly) - PlanDoneAccepted is topologically unreachable without semantic satisfaction - New contract test #10: semantic gate rejects then accepts - Updated all call-order assertions and adapter test counts Co-authored-by: Amp <amp@ampcode.com>

- Define TransitionContract type (kind, lane, actor, guard) in petri-net.ts - Add contract field to TransitionDef (optional, no behavioral change) - Populate contract metadata for all compiled transitions: mechanical-lane (evaluate, write-tests, write-code, run-tests), semantic-lane (assess-semantic, return-done), structural (slice-ready, epic-complete, epic-deps-met) - Add getTransitions() accessor on PetriNet for inspection - New adapter test verifies contract metadata classification Co-authored-by: Amp <amp@ampcode.com>

- Define NetEvent, NetEventKind, NetEventSink types in petri-net.ts - PetriNet.run() accepts optional event sink (backward compatible) - Emits transition_fired with id, contract, consumed/produced places - Emits net_halted when shouldHalt() triggers - Emits net_deadlocked when no transition enabled but tokens remain - New adapter tests: happy-path event capture + retry-exhaustion halt - Note: deadlock detection is coarse (leftover resource tokens trigger it on clean completion); refined detection is future work Co-authored-by: Amp <amp@ampcode.com>

Co-authored-by: Amp <amp@ampcode.com>

…dget - Split compilePlan into compileTopology (pure data → NetBlueprint) and wireHandlers (attaches fire closures). Adapter tests use topology-only. - Consolidate identical PetriOrchestrator/ProceduralOrchestrator into createOrchestrator(policy) factory. Delete engine-petri.ts/engine-proc.ts. - Move RunCtx from net-compiler.ts to types.ts (compiler is context-free). - Add semantic-budget place + reworkCount token to prevent infinite rework loops when assess-semantic always rejects (Finding #8 correctness fix). - HandlerDescriptor discriminated union (7 kinds) makes transition routing declarative in the blueprint. All descriptors carry actionKey explicitly. - New net-blueprint.ts for NetBlueprint/TransitionSkeleton/HandlerDescriptor types. - 1378 tests pass across 120 files. Co-authored-by: Amp <amp@ampcode.com>

…compilation Co-authored-by: Amp <amp@ampcode.com>

…ion. Register assess-semantic in createPiActions, replace the obsolete --engine flag with --policy wired to createOrchestrator, and skip net_deadlocked when only resource or completion tokens remain. Co-authored-by: Cursor <cursoragent@cursor.com>

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 78706a2. Configure here.}

cursor · 2026-05-27T23:28:04Z

-    expect(net.transitionCount).toBe(13);
+    // Should have a net_halted event (ctx.halted becomes true after retry exhaustion)
+    const halted = events.filter((e) => e.kind === 'net_halted');
+    expect(halted.length).toBe(1);


Test name says "net_deadlocked" but asserts "net_halted"

Medium Severity

The test is named "retry exhaustion emits net_deadlocked" but the assertion on line 827 checks for net_halted events, not net_deadlocked. These are semantically distinct §7 events: net_halted fires when shouldHalt() returns true (external halt signal from ctx.halted), while net_deadlocked fires when no transition is enabled but work-bearing tokens remain. Retry exhaustion sets ctx.halted = true, triggering net_halted on the next loop iteration — it never produces a net_deadlocked event. The test name misrepresents what behavior is actually being verified, which could mislead developers reasoning about the event model.

^{Reviewed by Cursor Bugbot for commit 78706a2. Configure here.}

kostandinang mentioned this pull request May 21, 2026

FE-730: Greenfield orchestrator POC with pi-agent dispatch #143

Open

kostandinang changed the title ~~FE-738: two-lane subnet with semantic completion gate~~ FE-738: Petri semantic lanes — two-lane subnet, compiler split, engine factory May 21, 2026

kostandinang marked this pull request as ready for review May 22, 2026 12:38

kostandinang mentioned this pull request May 22, 2026

FE-743: Petri parallel execution with resource pools and per-slice sandboxes #149

Open

cursor Bot reviewed May 22, 2026

View reviewed changes

Comment thread src/orchestrator/src/cook-cli.ts

Comment thread src/orchestrator/src/petri-net.ts

augmentcode Bot reviewed May 22, 2026

View reviewed changes

Comment thread src/orchestrator/src/net-compiler.ts

Comment thread src/orchestrator/src/cook-cli.ts Outdated

kostandinang force-pushed the ka/fe-730-orchestrator-poc-dual-engine branch from b72d1ad to 934ea57 Compare May 22, 2026 14:14

kostandinang force-pushed the ka/fe-738-petri-semantic-lanes branch from 8e56a7c to f959557 Compare May 22, 2026 14:14

cursor Bot reviewed May 22, 2026

View reviewed changes

Comment thread src/orchestrator/src/net-compiler.ts

kostandinang force-pushed the ka/fe-738-petri-semantic-lanes branch from f959557 to 1747538 Compare May 22, 2026 14:59

kostandinang mentioned this pull request May 22, 2026

FE-745: Epic-scoped sandbox merge for verify-epic #152

Open

kostandinang changed the base branch from ka/fe-730-orchestrator-poc-dual-engine to graphite-base/148 May 25, 2026 11:01

kostandinang force-pushed the ka/fe-738-petri-semantic-lanes branch from da2bb0d to 6c4fd65 Compare May 25, 2026 11:02

kostandinang force-pushed the graphite-base/148 branch from 934ea57 to ab20ef2 Compare May 25, 2026 11:02

kostandinang changed the base branch from graphite-base/148 to ka/fe-730-orchestrator-poc-dual-engine May 25, 2026 11:02

kostandinang force-pushed the ka/fe-730-orchestrator-poc-dual-engine branch from ab20ef2 to 5585e4e Compare May 25, 2026 11:05

kostandinang force-pushed the ka/fe-738-petri-semantic-lanes branch 3 times, most recently from e76b68a to 745c45a Compare May 25, 2026 11:23

kostandinang mentioned this pull request May 25, 2026

FE-747: Declarative output routing for branching transitions #154

Open

kostandinang self-assigned this May 26, 2026

kostandinang requested a review from lunelson May 26, 2026 07:52

kostandinang mentioned this pull request May 26, 2026

FE-755: Cook codebase mode for brownfield brunch #155

Open

kostandinang changed the title ~~FE-738: Petri semantic lanes — two-lane subnet, compiler split, engine factory~~ FE-738: Petri semantic lanes with two-lane subnet and engine factory May 27, 2026

This was referenced May 27, 2026

FE-761: Petri-net semantic alignment for Petrinaut visualization #156

Open

FE-762: Petrinaut-format JSON export of the compiled net #157

Open

FE-763: Petrinaut event stream — initial markings + transition firings #158

Open

kostandinang force-pushed the ka/fe-738-petri-semantic-lanes branch from 745c45a to 78b15a6 Compare May 27, 2026 23:16

kostandinang force-pushed the ka/fe-730-orchestrator-poc-dual-engine branch from 44300a6 to 052f6e7 Compare May 27, 2026 23:16

kostandinang and others added 7 commits May 28, 2026 01:20

plan: cards 1-3 done for petri-semantic-lanes, retire CARDS.md

44770f0

Co-authored-by: Amp <amp@ampcode.com>

plan: petri-semantic-lanes done, criterion 5 deferred to petri-graph-…

533473c

…compilation Co-authored-by: Amp <amp@ampcode.com>

kostandinang force-pushed the ka/fe-730-orchestrator-poc-dual-engine branch from 052f6e7 to 6d35473 Compare May 27, 2026 23:21

kostandinang force-pushed the ka/fe-738-petri-semantic-lanes branch from 78b15a6 to 78706a2 Compare May 27, 2026 23:21

cursor Bot reviewed May 27, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

FE-738: Petri semantic lanes with two-lane subnet and engine factory#148

FE-738: Petri semantic lanes with two-lane subnet and engine factory#148
kostandinang wants to merge 7 commits into
ka/fe-730-orchestrator-poc-dual-enginefrom
ka/fe-738-petri-semantic-lanes

kostandinang commented May 21, 2026 •

edited

Loading

Uh oh!

kostandinang commented May 21, 2026 •

edited

Loading

Uh oh!

cursor Bot commented May 22, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

augmentcode Bot commented May 22, 2026

Uh oh!

augmentcode Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor Bot left a comment

Uh oh!

cursor Bot May 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

kostandinang commented May 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Context

What changed

Verification

Out of scope

Traceability

Uh oh!

kostandinang commented May 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cursor Bot commented May 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

Uh oh!

Uh oh!

Uh oh!

augmentcode Bot commented May 22, 2026

Uh oh!

augmentcode Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor Bot May 27, 2026

Choose a reason for hiding this comment

Test name says "net_deadlocked" but asserts "net_halted"

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

kostandinang commented May 21, 2026 •

edited

Loading

kostandinang commented May 21, 2026 •

edited

Loading

cursor Bot commented May 22, 2026 •

edited

Loading