Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Design or review identification strategy for the sewage-house-prices project. Produces strategy memos with estimand, assumptions, pseudo-code, robustness plan, falsification tests, and referee objection anticipation. This skill should be used when asked to "design the strategy", "identify the effect", "write a strategy memo", or "think through identification".
.claude/skills/brycewang-stanford-identify/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 73% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 61% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 39% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 216% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 103% | 0% |
Design or review an identification strategy for the sewage-house-prices project.
Input: $ARGUMENTS — a research question, approach name (e.g. "hedonic", "dry spills"), or "review existing" to audit all current strategies.
log(price) ~ spill_metrics + controls | lsoa + year_quarter. Assumption: spill exposure is conditionally exogenous given LSOA FE.Δlog(price) ~ Δspill_metrics | house_id. Eliminates time-invariant unobservables.docs/overleaf/ for how strategies are currently describedscripts/R/09_analysis/scripts/R/utils/spill_aggregation_utils.R for treatment constructiondocs/overleaf/refs.bib for methodological referencesFor a new or revised strategy, produce:
If reviewing an existing strategy:
markdown# Identification Strategy: [Approach] **Date:** YYYY-MM-DD **Design:** [Hedonic / Repeat Sales / Long Diff / DiD / IV / etc.] **Estimand:** [ATT / ATE / LATE] ## Strategy Summary [2-3 sentence description] ## Estimating Equation $$\log(p_{it}) = \alpha + \beta \cdot \text{SpillMetric}_{it} + \gamma X_{it} + \mu_i + \delta_t + \varepsilon_{it}$$ ## Key Assumptions 1. [Assumption 1] — [defense] 2. [Assumption 2] — [defense] ## Assessment: [SOUND / CONCERNS / CRITICAL ISSUES] ## Robustness Plan (ordered) 1. [Most important check] 2. [Second check] ... ## Falsification Tests 1. [Test 1] — [expected null and why] ## Anticipated Referee Objections 1. [Objection] — [Response] ## Next Steps - [ ] Implement main specification - [ ] Run falsification tests - [ ] Generate pre-trend evidence
Save to output/log/strategy_memo_[approach].md.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→fail | 3,667 | 3,947 | +8% | 1 | 1 | 0% | 199 | 1,366 | +586% | 0 | 0 | — |
case-01 | fail→fail | 35,481 | 5,641 | -84% | 1 | 1 | 0% | 6,211 | 1,530 | -75% | 0 | 0 | — |
case-02 | fail→fail | 23,589 | 4,686 | -80% | 1 | 1 | 0% | 4,208 | 1,442 | -66% | 0 | 0 | — |
case-03 | fail→fail | 34,096 | 6,128 | -82% | 1 | 1 | 0% | 5,479 | 1,598 | -71% | 0 | 0 | — |
case-04 | pass→fail | 10,045 | 23,350 | +132% | 1 | 1 | 0% | 1,907 | 5,725 | +200% | 0 | 0 | — |
case-06 | pass→pass | 16,652 | 27,770 | +67% | 1 | 1 | 0% | 3,432 | 6,217 | +81% | 0 | 0 | — |
case-07 | fail→pass | 20,306 | 26,692 | +31% | 1 | 1 | 0% | 3,311 | 5,737 | +73% | 0 | 0 | — |
case-08 | fail→pass | 19,416 | 26,001 | +34% | 1 | 1 | 0% | 3,233 | 5,204 | +61% | 0 | 0 | — |
case-09 | fail→fail | 22,717 | 5,612 | -75% | 1 | 1 | 0% | 4,166 | 1,472 | -65% | 0 | 0 | — |
case-10 | fail→fail | 23,214 | 27,524 | +19% | 1 | 1 | 0% | 3,972 | 5,756 | +45% | 0 | 0 | — |
case-11 | fail→fail | 17,580 | 23,502 | +34% | 1 | 1 | 0% | 2,913 | 5,216 | +79% | 0 | 0 | — |
case-12 | pass→pass | 17,879 | 22,727 | +27% | 1 | 1 | 0% | 2,981 | 4,880 | +64% | 0 | 0 | — |
case-13 | fail→pass | 7,974 | 5,491 | -31% | 1 | 1 | 0% | 1,531 | 2,126 | +39% | 0 | 0 | — |
case-14 | fail→fail | 12,132 | 28,903 | +138% | 1 | 1 | 0% | 1,982 | 5,966 | +201% | 0 | 0 | — |
case-15 | fail→pass | 40,057 | 14,857 | -63% | 1 | 1 | 0% | 1,253 | 3,965 | +216% | 0 | 0 | — |
case-16 | fail→pass | 11,196 | 15,952 | +42% | 1 | 1 | 0% | 1,862 | 3,772 | +103% | 0 | 0 | — |
case-17 | fail→pass | 35,697 | 19,104 | -46% | 1 | 1 | 0% | 1,075 | 4,575 | +326% | 0 | 0 | — |
case-18 | fail→pass | 10,409 | 2,254 | -78% | 1 | 1 | 0% | 1,755 | 1,588 | -10% | 0 | 0 | — |
case-19 | pass→pass | 9,244 | 10,316 | +12% | 1 | 1 | 0% | 1,345 | 2,778 | +107% | 0 | 0 | — |
case-20 | fail→pass | 13,978 | 20,934 | +50% | 1 | 1 | 0% | 2,436 | 4,280 | +76% | 0 | 0 | — |
case-21 | pass→fail | 21,753 | 4,539 | -79% | 1 | 1 | 0% | 3,832 | 1,442 | -62% | 0 | 0 | — |
case-22 | fail→pass | 13,227 | 14,888 | +13% | 1 | 1 | 0% | 2,060 | 3,593 | +74% | 0 | 0 | — |
case-23 | fail→pass | 39,386 | 19,698 | -50% | 1 | 1 | 0% | 1,252 | 4,529 | +262% | 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. 23 cases were attempted, and 14 counted toward the lift figure. The other 9 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 +35 percentage points is the difference between those two pass rates over the 14 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.