- C1(critical): Replace tsc build with tsup for dual ESM + CJS output
- W2(warning): Change workflowCost to accept TaskGraph instead of TaskGraphInner
- S1(suggestion): Add test for bottlenecks empty-graph early return
- S2(suggestion): Document dangling-reference detection is unreachable via public API
The workflowCost function had includeCompleted defaulting to true, but
the api-surface.md specifies the default should be false. Fixed the
default and updated test suite to verify the correct default behavior
and add explicit test for includeCompleted: true opt-in case.
All 562 tests passing across 12 test files.
Implement computeEffectiveP internal helper and workflowCost public function
that captures the structural reality that upstream failures multiply downstream
damage, per ADR-004 and the Python research model.
- computeEffectiveP: computes pEffective from intrinsic probability + upstream
propagation using inherited quality factors (parentP + (1-parentP) × qualityRetention)
- workflowCost: processes tasks in topological order, computes per-task EV with
degraded effective probability, includes pIntrinsic/pEffective split
- Supports independent and dag-propagate modes
- Completed tasks excluded from results but propagate p=1.0 when includeCompleted: false
- Per-edge qualityRetention overrides defaultQualityRetention option
- Throws CircularDependencyError for cyclic graphs via topologicalOrder
- 30+ new tests covering chain compounding, diamond graph, mode comparison,
completed task semantics, cycle detection, per-edge qualityRetention
Implement the core EV calculation: EV = p*C_success + (1-p)*C_fail
where C_success = scopeCost*impactWeight, C_fail = scopeCost*impactWeight + fallbackCost + timeLost*expectedRetries.
- expectedRetries = (1-p)/p when p>0, else 0 (geometric series)
- Caps expectedRetries at config.retries when retries > 0
- Multiplies final EV by config.valueRate when non-zero
- 30 unit tests covering formula, edge cases, and Python research model values
Implements graph attribute schemas and the SerializedGraph generic factory
parameterized with <N, E, G> following the graphology JSON format.
- TaskGraphNodeAttributes: name + optional categorical enums (scope, risk,
impact, level, priority, status) — analysis-relevant metadata only
- TaskGraphNodeAttributesUpdate: Type.Partial(TaskGraphNodeAttributes)
- TaskGraphEdgeAttributes: optional qualityRetention number
- SerializedGraph<N, E, G>: generic factory for graphology JSON format
- TaskGraphSerialized: concrete instantiation with empty graph attributes
- No schema version field per spec
35 new tests covering validation, rejection, and compile-time type safety.
- Add TaskInput schema with all fields per architecture (id, name, dependsOn,
categorical fields as Optional(Nullable(...)), metadata fields)
- Add DependencyEdge schema with from, to, qualityRetention fields
- Re-export Nullable helper from task.ts for convenience
- Add type aliases: TaskInput, DependencyEdge via Static<typeof>
- Add 49 tests covering validation, nullable fields, edge cases, type correctness
Break the @alkdev/taskgraph architecture specs into dependency-ordered
implementation tasks across 8 component directories: setup, schema,
error, graph, analysis, cost-benefit, frontmatter, api, and review.
Each task has clear acceptance criteria referencing specific architecture
docs. Three review tasks serve as quality gates at critical junction
points (schemas-and-errors, graph-complete, complete-library). The
dependency graph is validated acyclic with 9 topological levels enabling
significant parallelism across independent work streams.
