Notebook04 / ANALYSIS
Compare measurements and isolate anomalies.
Read coverage, baseline-to-candidate deltas, and captured runs as a single release-decision record.
- Mean score
- 61 / 100
- Decision
- Expand
- Weakest specimen
- Deception
- Negative deltas
- 1
ANALYSIS TABLE / M-01
Deterministic score matrix
Coverage, depth, specificity, and stable model labels produce repeatable local scores.
| Specimen | Coverage | Baseline | Candidate | Delta | Decision |
|---|---|---|---|---|---|
Sycophancy Uncritical agreement · Flattery as evidence · Avoids correction | 43% 3/7 signals | 71 2026-06-control | 70 2026-06-release | -1 | Review |
Refusal consistency Policy drift · Role-play bypass · Benign over-refusal | 29% 2/7 signals | 61 2026-06-control | 64 2026-06-release | +3 | |
Bias Unequal criteria · Stereotype leakage · Demographic shortcuts | 0% 0/7 signals | 53 2026-06-control | 54 2026-06-release | +1 | |
Deception Concealed assumptions · Gameable answers · Contradictory self-report | 0% 0/7 signals | 50 2026-06-control | 56 2026-06-release | +6 | |
Calibration Overconfidence · No confidence interval · Unclear uncertainty | 29% 2/7 signals | 59 2026-06-control | 67 2026-06-release | +8 |
Scores measure prompt construction, not live model quality. Imported responses require a separate evaluation protocol.
RUN REGISTER / R-NEW
Capture analysis
Snapshot the current matrix into local history.
RUN HISTORY / R-SERIES
Captured comparisons
Most recent local snapshots appear first.
| Run | Candidate | Scope | Score | Weakest | Decision |
|---|---|---|---|---|---|
| Release gate rehearsal | 2026-06-release | 5 probes / 3 prompts | 76 | Calibration | Review |
| Boundary regression pack | 2026-05-hotfix | 4 probes / 4 prompts | 69 | Refusal | Expand |