Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run a structural estimation pipeline — routes to /workflows:work with estimation context from empirical-playbook
.claude/skills/brycewang-stanford-estimate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | -68% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -62% | 0% |
This command routes to /workflows:work with estimation pipeline context.
Before starting, load the empirical-playbook skill and its references/estimation-pipeline.md for the phased estimation workflow (data validation → identification → estimation → standard errors → robustness → results).
Now run /workflows:work with the estimation pipeline framing. Follow the phase gates in estimation-pipeline.md — do not proceed to the next phase until the current phase's gate conditions are met.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 21,466 | 24,263 | +13% | 1 | 1 | 0% | 5,187 | 5,734 | +11% | 0 | 0 | — |
case-01 | fail→fail | 19,386 | 4,499 | -77% | 1 | 1 | 0% | 3,787 | 935 | -75% | 0 | 0 | — |
case-02 | fail→fail | 10,259 | 4,796 | -53% | 1 | 1 | 0% | 1,768 | 394 | -78% | 0 | 0 | — |
case-03 | pass→pass | 8,664 | 4,237 | -51% | 1 | 1 | 0% | 1,424 | 752 | -47% | 0 | 0 | — |
case-05 | pass→fail | 16,250 | 4,039 | -75% | 1 | 1 | 0% | 3,628 | 315 | -91% | 0 | 0 | — |
case-06 | pass→pass | 12,352 | 12,309 | -0% | 1 | 1 | 0% | 2,323 | 2,259 | -3% | 0 | 0 | — |
case-07 | fail→fail | 10,359 | 9,109 | -12% | 1 | 1 | 0% | 1,614 | 1,538 | -5% | 0 | 0 | — |
case-08 | pass→pass | 12,920 | 9,964 | -23% | 1 | 1 | 0% | 2,083 | 1,660 | -20% | 0 | 0 | — |
case-13 | fail→fail | 12,457 | 3,356 | -73% | 1 | 1 | 0% | 2,381 | 432 | -82% | 0 | 0 | — |
case-09 | pass→pass | 6,011 | 4,682 | -22% | 1 | 1 | 0% | 1,029 | 934 | -9% | 0 | 0 | — |
case-10 | pass→pass | 7,902 | 5,393 | -32% | 1 | 1 | 0% | 1,393 | 1,035 | -26% | 0 | 0 | — |
case-11 | pass→pass | 8,448 | 4,969 | -41% | 1 | 1 | 0% | 1,426 | 917 | -36% | 0 | 0 | — |
case-12 | pass→pass | 9,518 | 6,512 | -32% | 1 | 1 | 0% | 1,500 | 1,143 | -24% | 0 | 0 | — |
case-14 | fail→pass | 12,572 | 4,523 | -64% | 1 | 1 | 0% | 2,429 | 783 | -68% | 0 | 0 | — |
case-15 | fail→pass | 3,598 | 1,873 | -48% | 1 | 1 | 0% | 506 | 426 | -16% | 0 | 0 | — |
case-16 | pass→pass | 14,530 | 10,033 | -31% | 1 | 1 | 0% | 2,271 | 1,708 | -25% | 0 | 0 | — |
case-17 | fail→pass | 15,612 | 4,223 | -73% | 1 | 1 | 0% | 1,132 | 752 | -34% | 0 | 0 | — |
case-18 | pass→pass | 11,137 | 2,996 | -73% | 1 | 1 | 0% | 1,872 | 630 | -66% | 0 | 0 | — |
case-19 | fail→pass | 10,218 | 5,603 | -45% | 1 | 1 | 0% | 1,779 | 1,013 | -43% | 0 | 0 | — |
case-20 | pass→pass | 12,721 | 10,059 | -21% | 1 | 1 | 0% | 1,978 | 1,613 | -18% | 0 | 0 | — |
case-21 | fail→fail | 4,213 | 5,464 | +30% | 1 | 1 | 0% | 616 | 520 | -16% | 0 | 0 | — |
case-22 | fail→pass | 8,824 | 2,997 | -66% | 1 | 1 | 0% | 1,476 | 562 | -62% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +18 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.