Data Skill Suite
Audited tool use, independent scoring, and skill version comparisons.
Condition and version comparisons
Differences use complete paired tasks. Positive values favor the right condition.
| Comparison | Pass-rate change | Tasks: improved / tied / regressed | Evidence |
|---|---|---|---|
| baseline → v1 | +0.0 pp | 0 / 10 / 0 | Descriptive only · insufficient or incomplete task sample 0 excluded pairs · 10 complete tasks |
| baseline → v2 | +0.0 pp | 0 / 10 / 0 | Descriptive only · insufficient or incomplete task sample 0 excluded pairs · 10 complete tasks |
| v1 → v2 | +0.0 pp | 0 / 10 / 0 | Descriptive only · insufficient or incomplete task sample 0 excluded pairs · 10 complete tasks |
Intervals are suppressed for fewer than five tasks, incomplete pairs, or no observed task variation. A tie does not establish equivalence.
baseline → v1 · per-task outcomes
| Task | Outcome | Change |
|---|---|---|
| contact-dedupe | tie | +0.0 pp |
| csv-cleanup | tie | +0.0 pp |
| customer-join | tie | +0.0 pp |
| decimal-invoices | tie | +0.0 pp |
| grouped-sales | tie | +0.0 pp |
| missing-versus-zero | tie | +0.0 pp |
| payment-reconciliation | tie | +0.0 pp |
| schema-validation | tie | +0.0 pp |
| stable-ranking | tie | +0.0 pp |
| timezone-events | tie | +0.0 pp |
baseline → v2 · per-task outcomes
| Task | Outcome | Change |
|---|---|---|
| contact-dedupe | tie | +0.0 pp |
| csv-cleanup | tie | +0.0 pp |
| customer-join | tie | +0.0 pp |
| decimal-invoices | tie | +0.0 pp |
| grouped-sales | tie | +0.0 pp |
| missing-versus-zero | tie | +0.0 pp |
| payment-reconciliation | tie | +0.0 pp |
| schema-validation | tie | +0.0 pp |
| stable-ranking | tie | +0.0 pp |
| timezone-events | tie | +0.0 pp |
v1 → v2 · per-task outcomes
| Task | Outcome | Change |
|---|---|---|
| contact-dedupe | tie | +0.0 pp |
| csv-cleanup | tie | +0.0 pp |
| customer-join | tie | +0.0 pp |
| decimal-invoices | tie | +0.0 pp |
| grouped-sales | tie | +0.0 pp |
| missing-versus-zero | tie | +0.0 pp |
| payment-reconciliation | tie | +0.0 pp |
| schema-validation | tie | +0.0 pp |
| stable-ranking | tie | +0.0 pp |
| timezone-events | tie | +0.0 pp |
Every agent run
| Task / repetition | Condition | Outcome | Tool calls | Skill read | Seconds | Evidence |
|---|---|---|---|---|---|---|
| decimal-invoices / 1 | v1 | passed | 7 | Yes | 47.607 | Tools · Score · Output |
| decimal-invoices / 1 | baseline | passed | 6 | No | 55.978 | Tools · Score · Output |
| decimal-invoices / 1 | v2 | passed | 7 | Yes | 34.79 | Tools · Score · Output |
| missing-versus-zero / 1 | baseline | passed | 6 | No | 40.616 | Tools · Score · Output |
| missing-versus-zero / 1 | v2 | passed | 7 | Yes | 52.357 | Tools · Score · Output |
| missing-versus-zero / 1 | v1 | passed | 7 | Yes | 55.329 | Tools · Score · Output |
| csv-cleanup / 1 | v2 | passed | 8 | Yes | 53.024 | Tools · Score · Output |
| csv-cleanup / 1 | v1 | passed | 7 | Yes | 43.236 | Tools · Score · Output |
| csv-cleanup / 1 | baseline | passed | 6 | No | 65.916 | Tools · Score · Output |
| contact-dedupe / 1 | v1 | passed | 7 | Yes | 42.72 | Tools · Score · Output |
| contact-dedupe / 1 | baseline | passed | 6 | No | 43.509 | Tools · Score · Output |
| contact-dedupe / 1 | v2 | passed | 8 | Yes | 61.826 | Tools · Score · Output |
| grouped-sales / 1 | baseline | passed | 6 | No | 29.501 | Tools · Score · Output |
| grouped-sales / 1 | v2 | passed | 8 | Yes | 33.35 | Tools · Score · Output |
| grouped-sales / 1 | v1 | passed | 7 | Yes | 29.075 | Tools · Score · Output |
| timezone-events / 1 | v2 | passed | 8 | Yes | 51.31 | Tools · Score · Output |
| timezone-events / 1 | v1 | passed | 7 | Yes | 27.325 | Tools · Score · Output |
| timezone-events / 1 | baseline | passed | 6 | No | 46.555 | Tools · Score · Output |
| customer-join / 1 | v1 | passed | 8 | Yes | 29.298 | Tools · Score · Output |
| customer-join / 1 | baseline | passed | 7 | No | 31.473 | Tools · Score · Output |
| customer-join / 1 | v2 | passed | 8 | Yes | 28.694 | Tools · Score · Output |
| stable-ranking / 1 | baseline | passed | 6 | No | 29.751 | Tools · Score · Output |
| stable-ranking / 1 | v2 | passed | 7 | Yes | 26.56 | Tools · Score · Output |
| stable-ranking / 1 | v1 | passed | 7 | Yes | 29.482 | Tools · Score · Output |
| payment-reconciliation / 1 | v2 | passed | 9 | Yes | 32.532 | Tools · Score · Output |
| payment-reconciliation / 1 | v1 | passed | 8 | Yes | 62.082 | Tools · Score · Output |
| payment-reconciliation / 1 | baseline | passed | 7 | No | 32.329 | Tools · Score · Output |
| schema-validation / 1 | v1 | passed | 7 | Yes | 35.581 | Tools · Score · Output |
| schema-validation / 1 | baseline | passed | 6 | No | 36.488 | Tools · Score · Output |
| schema-validation / 1 | v2 | passed | 8 | Yes | 59.459 | Tools · Score · Output |
Scope and reproducibility
Original synthetic data tasks; this is not a representative real-world benchmark.
The model is an alias, not a pinned provider snapshot. Repeat outcomes may vary.
Explicit skill discovery through MCP; not native implicit skill routing or arbitrary shell/code execution.
Hashes verify bundle integrity, not model authenticity or independent replication.
Check_output validates JSON syntax only. Hidden expected outputs are scored after the agent exits.
Usage is reported tokens and elapsed time; monetary cost is not independently measured.
Suite: data-workflows-v1 / 1.0.0 · Created: 2026-10-14T08:58:08.015407+00:00
Metrics · Frozen run plan · Integrity manifest · Environment
Offline integrity check: skillarena verify <this-directory>. Integrity checks do not certify performance or provenance.