t1-p2-05authored4–7h
One targeted fix for the largest failure bucket
requires: t1-p2-04
Spec
Pick your single largest failure bucket and fix only that. If retrieval dominates, try the other chunking strategy or add re-ranking; if hallucination dominates, tighten the prompt to require verbatim quoting before synthesis. Record the choice and reasoning in evals/fixes/<bucket>.md before changing code.
Why this matters
Only one, deliberately. Changing three things and re-running tells you the score moved but not which change moved it. Constraining yourself to one intervention is what makes the next task's comparison mean anything.
Verify
No result yet. Run `learn verify t1-p2-05` locally, or push to trigger CI.
repo
unrun
judge
unrun
Rubric
# t1-p2-05 — One targeted fix for the largest failure bucket
**Task:** Pick your single largest failure bucket and fix only that. If retrieval dominates, try the other chunking strategy or add re-ranking; if hallucination dominates, tighten the prompt to require verbatim quoting before synthesis. Record the choice and reasoning in evals/fixes/<bucket>.md before changing code.
**Strong:** Directly answers every one of the review questions below in the affirmative, with specific evidence from the actual work, not a restatement of the task.
- Is the chosen bucket actually the largest?
- Is the fix causally linked to it?
**Passing but weak:** The deliverable exists and technically satisfies the spec, but is generic, restates the task instructions instead of engaging with them, or only partially addresses the review questions below.
**Fails:** The deliverable is missing, or engages with none of the review questions below - e.g. it describes what was built without explaining the reasoning the task actually asks for.