# Mini data snapshot — Palmer Penguins

Dataset: **Palmer Archipelago (Antarctica) penguins**, `penguins.csv`.
URL (verified live at fetch time, HTTP 200): `https://raw.githubusercontent.com/mwaskom/seaborn-data/master/penguins.csv`
344 rows, 8 columns. Computed locally with Python 3 (stdlib only, no pandas).

## Stat 1 — Percentile

```python
body = [float(r["body_mass_g"]) for r in rows if r["body_mass_g"]]
sorted(body)[int(round(0.75*(len(body)-1)))]   # linear-interpolated p75
```
**p75 body_mass_g = 4750 g.** → Three quarters of the sampled penguins weigh under 4.75 kg.

## Stat 2 — Correlation (with a twist)

```python
pearson([(float(r["bill_length_mm"]), float(r["bill_depth_mm"])) for r in rows if r["bill_length_mm"] and r["bill_depth_mm"]])
```
**All 342 penguins: r(bill_length, bill_depth) = −0.235** (longer bills, shallower bills).
Split by species, the sign flips: **Adelie +0.391, Chinstrap +0.654, Gentoo +0.643.**
→ The pooled negative correlation is a textbook Simpson's paradox: a grouping effect, not a trait-level relationship.

## Stat 3 — Group-by count

```python
Counter(r["species"] for r in rows)
```
**Adelie 152, Chinstrap 68, Gentoo 124** (total 344).
→ Adelie is the largest cohort and roughly 2.2× the Chinstrap sample.

## Claim check

**Claim:** "Across the whole dataset, penguins with longer bills have shallower bills."
**Verdict: not-supported** — by our own output the pooled r is indeed negative (−0.235), but that is driven entirely by species mix; *within every species* the correlation is positive (+0.39 to +0.65), so the claim describes the grouping artifact, not the birds.

_(Words: 197. Dataset URL live-checked at fetch time. Deliverable in English except this note; column names and headings kept in English.)_