Data audit
| file | SHA-256 | rows (excl. header) | cols |
|---|---|---|---|
| data/life_expectancy_panel.csv | b3f2f774ed6797034acf62853286ca37fc02334ba3ce70e745e4f3d5c533c3ac (matches bundle.json) | 439 | 4 |
| data/README.md | e9099d209a7642b2b2151161277222719045c3c1094e9490a5d02faa316353d0 (matches bundle.json) | — | — |
Schema: country_name (str), iso3 (str, 20 distinct, no whitespace issues), year (int, 2000–2022), life_expectancy (float, years, 3 dp; range 51.49–84.56; 1 empty cell). Rows per year: 17 for 2000–2010, 21 for 2011–2022 (21 = 20 countries + TUR duplicate).
Anomalies:
- TUR duplicated: 46 rows = 23 years × 2, names "Turkey" and "Turkiye", values identical for every year → exact assembly duplicate; keep one row per iso3-year (no conflict, so choice does not affect values). README lists the name as "Türkiye"; file uses "Turkiye" — cosmetic.
- CHE 2007 empty life_expectancy → missing; CHE not balanced.
- BGD, ETH, IND, NGA: rows only for 2011–2022 (2000–2010 absent, likely the narrower extraction) → not balanced.
- Year-on-year changes >2 years only for BGD 2022 (+3.2), IND 2021 (−2.9) and 2022 (+4.4): COVID-era, plausible, kept as recorded (none of these countries are in the balanced set anyway). No values out of plausible range. No information-availability violations in the input (single file, observed data).