Milestones
List view
Spec Section 11, Milestone 7. Time: 1-2 weekends (10-20 hours). Do these in order. If `make eval` improves, keep the change. Otherwise revert it. 1. Contextual summaries in Stage 5 (rules 5 and 7). 2. Hypothetical questions and the `dense_hq` vector (Stage 5 rule 6, Stage 12 rule 7). 3. Query decomposition (Stage 11 rule 7). 4. The semantic answer cache (Stage 19). 5. The `sandwich` context order (Stage 15 rule 6). Done when: each accepted change has one evaluation run in `eval/runs/` that shows the gain.
No due date•0/5 issues closedSpec Section 11, Milestone 5. Spec Section 7. Time: 1 weekend (about 10 hours). Steps: 1. Add every attribute in Section 7.4 and Section 7.6 to the spans (3 hours). 2. Add structlog with trace correlation and the one-line request log (1 hour). 3. Add the `/feedback` endpoint and the CLI keypress (2 hours). 4. Build the five dashboard panels. Set the five alerts if the backend supports them (3 hours). Done when: the debug runbook in Section 7.11 works end to end on one wrong answer caused on purpose, for example with `gate.tau_pass = 0.99`.
No due date•2/6 issues closed