radarist/structured-analytic-skills / skills/quantitative-sanity-check
quantitative-sanity-check
VerifiedRecomputes the arithmetic that a document's own numbers imply — CAGR and compounding, price × volume unit economics, percentage vs percentage-point change, share-of-whole, survivorship and base-rate framing, Fermi order-of-magnitude — and flags every claim whose figures do not reproduce each other. Use when a report, pitch or press release states linked numbers — "$6B growing to $50B at a 25% CAGR", "40,000 customers at $99 a month is $120M ARR", "conversion improved 5%", "run a quantitative sanity check on these figures". Internal consistency only — not for checking a lone figure against outside sources (use `grounded-fact-check`) or sizing a market (use `estimate-market-size`).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 11.1 KB |
| agents/openai.yaml | config | 329 B |
| evals/evals.json | config | 5.7 KB |
| references/checks.md | markdown | 2.9 KB |
| scripts/sanity.py | script | 12.3 KB |
evals/evals.json
View on GitHub{
"skill_name": "quantitative-sanity-check",
"evals": [
{
"id": 1,
"case_id": "qsc-pos-1",
"kind": "positive",
"prompt": "Run a quantitative sanity check on this investor memo: it says the market grows from $6.25B to $50B by 2032 at a 24.8% CAGR, and that 40,000 customers at $99 a month produce $120M ARR. Do the numbers reproduce each other?",
"expected_output": "A response that satisfies these checks: Recomputes the implied CAGR from start, end and years (8 compounding periods) and flags the 24.8% claim as inconsistent; Multiplies customers × annual price and reports the gap to the claimed ARR as a multiple; Uses the verdict table with a Recomputed column and an Overall/Action line; Does not pick a side — flags the inconsistency and routes to re-sourcing.",
"files": [],
"expectations": [
"Recomputes the implied CAGR from start, end and years (8 compounding periods) and flags the 24.8% claim as inconsistent",
"Multiplies customers × annual price and reports the gap to the claimed ARR as a multiple",
"Uses the verdict table with a Recomputed column and an Overall/Action line",
"Does not pick a side — flags the inconsistency and routes to re-sourcing"
],
"skills": [
"quantitative-sanity-check"
]
},
{
"id": 2,
"case_id": "qsc-pos-2",
"kind": "positive",
"prompt": "These figures in the pitch deck do not feel right — they say revenue tripled in four years at 25% a year, and that share fell 50% to 20%. Can you recompute the arithmetic and tell me whether the numbers are internally consistent?",
"expected_output": "A response that satisfies these checks: Checks that 25% compounded over four years gives about 2.4×, not 3×, and flags the compounding claim; Shows both readings of 'fell 50% to 20%' (from 40% relative or from 70% in points) and marks it ambiguous; Reports verdicts per claim rather than a single overall impression.",
"files": [],
"expectations": [
"Checks that 25% compounded over four years gives about 2.4×, not 3×, and flags the compounding claim",
"Shows both readings of 'fell 50% to 20%' (from 40% relative or from 70% in points) and marks it ambiguous",
"Reports verdicts per claim rather than a single overall impression"
],
"skills": [
"quantitative-sanity-check"
]
},
{
"id": 3,
"case_id": "qsc-pos-3",
"kind": "positive",
"prompt": "Before I quote this in the report — the source says conversion improved 5% and elsewhere gives 10% before and 15% after. Is that percentage points or relative percent, and is the claim consistent with its own before/after numbers?",
"expected_output": "A response that satisfies these checks: Computes both the percentage-point change (+5 pp) and the relative change (50%) and states which the prose matches; Recommends restating the claim in unambiguous units; Reproduces the check with scripts/sanity.py pp or equivalent arithmetic.",
"files": [],
"expectations": [
"Computes both the percentage-point change (+5 pp) and the relative change (50%) and states which the prose matches",
"Recommends restating the claim in unambiguous units",
"Reproduces the check with scripts/sanity.py pp or equivalent arithmetic"
],
"skills": [
"quantitative-sanity-check"
]
},
{
"id": 4,
"case_id": "qsc-neg-1",
"kind": "negative",
"prompt": "Fact-check the vendor numbers in this brief before we publish — the qubit count, the parameter count and the benchmark scores — against live web sources and reconcile the draft with what the sources actually say.",
"expected_output": "A response that satisfies these checks: Verifies each specific against external sources rather than recomputing internal arithmetic; Behaves as grounded-fact-check: reconciles the draft against what the sources state.",
"files": [],
"expectations": [
"Verifies each specific against external sources rather than recomputing internal arithmetic",
"Behaves as grounded-fact-check: reconciles the draft against what the sources state"
],
"skills": [
"grounded-fact-check"
]
},
{
"id": 5,
"case_id": "qsc-neg-2",
"kind": "negative",
"prompt": "How big is the market for warehouse robotics? Give me TAM, SAM and SOM with a top-down and a bottom-up estimate that agree within an order of magnitude.",
"expected_output": "A response that satisfies these checks: Builds two independent market-size estimates rather than checking a stated figure's arithmetic; Behaves as estimate-market-size: TAM/SAM/SOM triangulation.",
"files": [],
"expectations": [
"Builds two independent market-size estimates rather than checking a stated figure's arithmetic",
"Behaves as estimate-market-size: TAM/SAM/SOM triangulation"
],
"skills": [
"estimate-market-size"
]
},
{
"id": 6,
"case_id": "qsc-neg-3",
"kind": "negative",
"prompt": "Variant B converted at 12.4% versus 11.9% for variant A with about 5,000 users in each arm — is that difference statistically significant or just noise?",
"expected_output": "A response that satisfies these checks: Runs a two-proportion significance test with a confidence interval and effect size instead of an arithmetic consistency check; Behaves as test-significance.",
"files": [],
"expectations": [
"Runs a two-proportion significance test with a confidence interval and effect size instead of an arithmetic consistency check",
"Behaves as test-significance"
],
"skills": [
"test-significance"
]
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |