Status: ReleasedData ownership: Institute-ownedLicence: CC BY 4.0
Skill-library utility curve (measured, N=1..10000)
The measured Utility-Problem curve behind the Skill Runtime note: naive-scan vs indexed matching cost as the library grows.
Purpose
Show where naive scanning stops being viable and indexing is required.
Source
A skill-runtime matching probe on Apple Silicon using a model-free synthetic harness, median of 5 reps, measured 2026-06-13.
Composition
CSV, 8 rows. Fields: N, naive_scan_match_us, indexed_match_us, re_derive_wall_us, candidates_scanned, candidates_idx_visited. N in {1,10,100,1000,2000,5000,8000,10000}.
Known biases
- A synthetic harness rather than a real model, so absolute speedups differ from real-model runs.
Limitations
- Single machine; µs wall-clock point measurements.
Access conditions
The full measured curve is available for download below, under CC-BY-4.0.
| N | naive_scan_match_us | indexed_match_us | re_derive_wall_us | candidates_scanned |
|---|---|---|---|---|
| 1 | 8.11 | 12.0 | 3357.1 | 1 |
| 10 | 10.01 | 12.0 | 3357.1 | 10 |
| 100 | 39.94 | 12.0 | 3357.1 | 100 |
| 1000 | 384.93 | 12.0 | 3357.1 | 1000 |
| 2000 | 899.91 | 12.0 | 3357.1 | 2000 |
| 5000 | 2029.90 | 12.0 | 3357.1 | 5000 |
| 8000 | 3607.99 | 12.0 | 3357.1 | 8000 |
| 10000 | 4361.03 | 12.0 | 3357.1 | 10000 |