Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use this skill when the user discusses experiment design, ablations, training runs, evaluation, baselines, metrics, failures, or result interpretation that should be logged into Obsidian experiment and result notes.
.claude/skills/brycewang-stanford-obsidian-experiment-log/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 481% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 378% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -22% | 0% |
Use this skill whenever project work changes the experimental state.
This is a supporting skill under obsidian-project-memory.
It should help maintain canonical experiment and result notes, not create note sprawl.
Experiments/Results/, if a durable finding existsDaily/ noteDaily/ until they are interpreted.planned, running, done, failed)Link experiments and results directly to each other, and link both back to 00-Hub.md, 01-Plan.md, or Daily/ only when those references improve the main working surface.
Treat experiment notes as the bridge between Papers/ and Results/:
Results/,Writing/ rather than leaving the chain unfinished.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→pass | 12,066 | 6,923 | -43% | 1 | 1 | 0% | 2,235 | 1,681 | -25% | 0 | 0 | — |
case-01 | fail→pass | 5,713 | 9,607 | +68% | 1 | 1 | 0% | 429 | 2,491 | +481% | 0 | 0 | — |
case-02 | fail→pass | 6,061 | 7,982 | +32% | 1 | 1 | 0% | 408 | 1,951 | +378% | 0 | 0 | — |
case-03 | fail→pass | 11,975 | 10,789 | -10% | 1 | 1 | 0% | 2,285 | 2,517 | +10% | 0 | 0 | — |
case-04 | pass→pass | 6,952 | 4,528 | -35% | 1 | 1 | 0% | 1,280 | 1,341 | +5% | 0 | 0 | — |
case-05 | pass→pass | 7,836 | 3,367 | -57% | 1 | 1 | 0% | 1,517 | 1,100 | -27% | 0 | 0 | — |
case-06 | fail→pass | 9,262 | 4,215 | -54% | 1 | 1 | 0% | 1,628 | 1,273 | -22% | 0 | 0 | — |
case-07 | fail→pass | 8,082 | 1,671 | -79% | 1 | 1 | 0% | 1,388 | 671 | -52% | 0 | 0 | — |
case-08 | fail→pass | 7,363 | 1,361 | -82% | 1 | 1 | 0% | 1,252 | 634 | -49% | 0 | 0 | — |
case-09 | fail→pass | 10,415 | 4,086 | -61% | 1 | 1 | 0% | 1,886 | 1,179 | -37% | 0 | 0 | — |
case-11 | fail→pass | 7,737 | 2,343 | -70% | 1 | 1 | 0% | 1,524 | 920 | -40% | 0 | 0 | — |
case-12 | pass→pass | 8,235 | 4,057 | -51% | 1 | 1 | 0% | 1,380 | 1,093 | -21% | 0 | 0 | — |
case-13 | pass→pass | 8,810 | 2,386 | -73% | 1 | 1 | 0% | 1,545 | 838 | -46% | 0 | 0 | — |
case-14 | fail→fail | 13,753 | 9,824 | -29% | 1 | 1 | 0% | 2,305 | 2,082 | -10% | 0 | 0 | — |
case-15 | pass→pass | 12,627 | 5,889 | -53% | 1 | 1 | 0% | 2,510 | 1,568 | -38% | 0 | 0 | — |
case-16 | pass→pass | 10,277 | 5,655 | -45% | 1 | 1 | 0% | 2,101 | 1,401 | -33% | 0 | 0 | — |
case-17 | fail→fail | 7,244 | 1,337 | -82% | 1 | 1 | 0% | 1,329 | 616 | -54% | 0 | 0 | — |
case-18 | pass→pass | 4,314 | 1,459 | -66% | 1 | 1 | 0% | 736 | 635 | -14% | 0 | 0 | — |
case-19 | pass→pass | 7,539 | 4,284 | -43% | 1 | 1 | 0% | 1,406 | 1,245 | -11% | 0 | 0 | — |
case-20 | pass→pass | 8,039 | 6,153 | -23% | 1 | 1 | 0% | 1,767 | 1,876 | +6% | 0 | 0 | — |
case-21 | pass→pass | 14,855 | 10,016 | -33% | 1 | 1 | 0% | 3,052 | 2,602 | -15% | 0 | 0 | — |
case-22 | pass→pass | 7,940 | 5,103 | -36% | 1 | 1 | 0% | 1,463 | 1,287 | -12% | 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 21 counted toward the lift figure. The other 1 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 +41 percentage points is the difference between those two pass rates over the 21 comparable cases.
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.