skills / data

statistical-analysis

what it does

Apply statistical methods including descriptive stats, trend analysis, outlier detection, and hypothesis testing. Use when analyzing distributions, testing for significance, detecting anomalies, computing correlations, or interpreting statistical results.

Verified Before and after: 2 models 1 try per case View source Download .zip

Compare the answers

Answers from Claude Sonnet 5.5 and GPT-6.1 Sol · try 1 of 1

Same request, same model, once with the skill and once without. Read both and judge for yourself.

“Here are daily signups for the last four weeks. Flag anything anomalous. I'm planning to drop outliers before I compute the monthly average.”

Source material given with the request

Date Day Signups 2026-09-07 Mon 312 2026-09-08 Tue 328 2026-09-09 Wed 305 2026-09-10 Thu 331 2026-09-11 Fri 298 2026-09-12 Sat 196 2026-09-13 Sun 184 2026-09-14 Mon 318 2026-09-15 Tue 335 2026-09-16 Wed 322 2026-09-17 Thu 0 2026-09-18 Fri 309 2026-09-19 Sat 201 2026-09-20 Sun 188 2026-09-21 Mon 326 2026-09-22 Tue 341 2026-09-23 Wed 317 2026-09-24 Thu 329 2026-09-25 Fri 1140
Show the rest (10 more lines)
2026-09-26 Sat 207 2026-09-27 Sun 192 2026-09-28 Mon 330 2026-09-29 Tue 338 2026-09-30 Wed 321 2026-10-01 Thu 344 2026-10-02 Fri 315 2026-10-03 Sat 205 2026-10-04 Sun 199
with the skill
**Two dates stand out: September 17 (0 signups) and September 25 (1,140 signups). Investigate them before dropping either.** The lower weekend counts look normal, not anomalous. ### Anomaly flags | Date | Signups | Assessment | What to check | |---|---:|---|---| | **2026-09-17, Thu** | **0** | Other Thursdays recorded **329–344**. A zero is highly unusual. | Missing data, tracking failure, or a genuine signup/service outage. | | **2026-09-25, Fri** | **1,140** | Other Fridays recorded **298–315**. This is about **3.7×** their average. | Campaigns, launches, bulk imports, duplicate events, or bot activity. | **Normal pattern:** Aside from those two dates, weekdays range from **298–344**, while weekends range from **184–207**. That consistent weekly pattern means weekends should **not** be removed as low outliers. There is also a modest upward drift, especially on weekends, rather than another obvious isolated anomaly. ### Effect on your average | Calculation | Daily mean | Daily median | |---|---:|---:| | All 28 days | **304.7** | **316** | | Excluding both flagged days—sensitivity check only | **284.3** | **316** | The two unusual values partly offset one another, so the overall mean alone hides them.
Show the rest (10 more lines)
### Recommendation before deleting anything - **If the zero reflects a real outage:** keep it. It is part of actual signup performance. - **If the zero means missing tracking:** recover the count if possible; otherwise mark it missing and disclose any estimation. - **If the spike represents real signups:** keep it, even if a campaign caused it. - **If the spike contains duplicates or invalid signups:** correct those records using a documented rule. **Use the all-days mean for actual performance unless you establish a data-quality problem.** For “typical daily volume,” report the median alongside it and distinguish weekdays from weekends. Also, this window is **September 7–October 4**, not a calendar month. If you mean September’s monthly average, you still need September 1–6 and should exclude October 1–4.
without
**Two dates stand out after accounting for the weekday/weekend pattern:** | Date | Signups | Why it’s anomalous | What to check | |---|---:|---|---| | **2026-09-17 (Thu)** | **0** | Other Thursdays have 329–344 signups. | Missing data, tracking failure, or a genuine service outage. | | **2026-09-25 (Fri)** | **1,140** | Other Fridays have 298–315 signups; this is roughly 3.7× normal. | Campaign, bulk import, bots, duplicate events, or a reporting error. | **The lower weekend counts are not anomalies.** Saturdays consistently have 196–207 signups and Sundays 184–199. They belong in an average covering all calendar days. **Don’t automatically drop the two flagged dates:** - If the value is a reporting error, correct it if possible. Treat missing data as missing—not as zero. - If it reflects genuine activity, including an outage or successful campaign, retain it in the actual average. You can report a separate “typical-day” average with exclusions clearly disclosed. Also, this is a **28-day window spanning September 7–October 4**, not a complete calendar month. Its unadjusted average is **304.68 signups/day**; excluding both flagged dates gives **284.27**, but that should be labeled an adjusted average rather than the monthly average.

Answers are shown exactly as the model wrote them, from the first try, and nothing is picked by hand.

The models

2 models

Claude Sonnet 5.5

made by Anthropic

claude-sonnet-5-5 · revision not reported by the provider

1 try per case · tested 3 Nov 2026

GPT-6.1 Sol

made by OpenAI

gpt-6.1-sol · revision not reported by the provider

1 try per case · tested 3 Nov 2026

The tasks

written by Carte Blanche
landing-page-simpson

“Which landing page converts better? Looks like B is the clear winner to me.”

segment-pvalue-multiple-tests

“Marketing wants to put next quarter's budget into Ohio because the analysis found a significant lift there (p = 0.02). Is that sound?”

signup-anomalies-before-removal

“Here are daily signups for the last four weeks. Flag anything anomalous. I'm planning to drop outliers before I compute the monthly average.”

store-layout-paired-test

“I tested whether our new store layout increased sales. I ran a two-sample t-test and got p = 0.63, so I'm going to report that the layout did nothing. Is my conclusion right?”

typical-order-value-skewed

“What's our typical order value this week? I need one number for the weekly report.”

Install

Download the .zip and unzip it. Put its files, in a folder named statistical-analysis, here:

Claude Code~/.claude/skills/Every project. For one project only, use .claude/skills/ inside it.
Codex~/.agents/skills/Every project. For one repository only, use .agents/skills/ inside it.
Cursor~/.cursor/skills/Cursor also reads the Claude Code and Codex folders above.

Then start a new session and ask for what the skill does. Other agents that read SKILL.md have their own skills folder; check their docs.

Coming after launch: get skills from inside your agent, no download needed.