Null resultAI usage (tokens)
Rule checks out of sample
Across every band measured on unseen skills, how many rule checks paid?
measured-local2026-09-1980e4a13
- docs/experiments/DETECTION-2026-09-19.md · 80e4a13 · 2026-09-20
Headline figures
18
Unseen skills compared
measured-local · 2026-09-19
0of 18
Unseen skills accepted
95% interval 0.0% to 17.6%
measured-local · n=18 · 2026-09-19
4of 8
Seen, ranked lane K skills accepted
95% interval 21.5% to 78.5%
measured-local · n=8 · 2026-09-19
Notes
- Seen and unseen are never pooled. The four accepted skills are all from the ranked, seen set.
- The conditional band has 20 candidates in the whole measurable corpus and every one is now measured, so a larger conditional draw is not possible.
Tables
n: Skills compared in the band. 5 rows.
| band | provenance | source_table | compared | accepted | interval_low | interval_high | mean_token_reduction | n | evidence_class | date | commit |
|---|---|---|---|---|---|---|---|---|---|---|---|
| conditional, ranked | seen | rule-checks | 8 | 4 | 0.2152 | 0.7848 | -0.0073 | 8 | measured-local | 2026-09-19 | 80e4a13fb96ebb96d8efff8690a8c77215657590 |
| conditional, rest of band | unseen | conditional-rule-checks-rest | 2 | 0 | 0 | 0.6576 | -0.2897 | 2 | measured-local | 2026-09-19 | 80e4a13fb96ebb96d8efff8690a8c77215657590 |
| absolute prohibition | unseen | absolute-rule-checks | 12 | 0 | 0 | 0.2425 | 0.0569 | 12 | measured-local | 2026-09-19 | 80e4a13fb96ebb96d8efff8690a8c77215657590 |
| path-scoped write | unseen | path-rule-checks | 4 | 0 | 0 | 0.4899 | -0.6494 | 4 | measured-local | 2026-09-19 | 80e4a13fb96ebb96d8efff8690a8c77215657590 |
| all unseen combined | unseen | conditional-rule-checks-rest;absolute-rule-checks;path-rule-checks | 18 | 0 | 0 | 0.1759 | -0.1385 | 18 | measured-local | 2026-09-19 | 80e4a13fb96ebb96d8efff8690a8c77215657590 |