product delivery evalsnextlab
Evaluate components and trajectories
Consolidate the vertical slice with component-level correctness checks and end-to-end trajectory evaluation before claiming readiness.
- Stable ID
sprint.consolidation- Item version
- 1.0.0
- Time
- 120 min
- Status
- ready
Reading lens
Focus
- component evals
- trajectory evals
- regression evidence
- unsupported-claim checks
Ignore for now
- single aggregate quality scores
- demo-only happy paths
- feature expansion during consolidation
Canonical sources
Open the material
Before reading
Which failures can pass an end-to-end demo while remaining visible in component-level evidence?
After reading
Build an eval matrix covering validators, alignment, marking, ECF, routing, hints, selection, and full trajectories.
Knowledge check
Why are both component and trajectory evals necessary?
Write an answer to unlock the reference
Component evals localise correctness and policy failures; trajectory evals verify that routing and state transitions compose into the intended learner experience. Either layer alone can hide important defects.
INTERNAL_REFERENCE · Evaluation strategy derived from repository contracts.Working notes