# Data audit — F1-conf-02

# data/life_expectancy_panel.csv

SHA-256: b3f2f774ed6797034acf62853286ca37fc02334ba3ce70e745e4f3d5c533c3ac (matches bundle.json)
439 data rows + header, 4 columns: country_name (str), iso3 (str), year (int), life_expectancy (float, years, 3 decimals).
Years 2000–2022 (all 23 present). 20 distinct iso3. No whitespace issues in iso3/country_name.
Value range 51.490 (NGA 2011) – 84.560 (JPN). Largest year-on-year changes are 2020–2022 COVID-era moves (IND −2.87 in 2021, +4.42 in 2022; BGD +3.16 in 2022; USA −1.81 in 2020; ESP −1.60 in 2020) — plausible, kept.

Duplicate country-years, TUR: 46 rows = 23 years under name "Turkey" and 23 under "Turkiye", same iso3, identical values in all 23 years (an assembly artefact of the two extractions; the WB renamed the country). Handling: keep one row per (iso3, year); since values are identical, choice affects no number. Retained the "Turkiye" label (current WB naming; README lists Türkiye). 23 rows removed.
Missing value: CHE 2007 has an empty life_expectancy cell. Not imputed (rule). CHE therefore incomplete and excluded from balanced set.
Truncated coverage: BGD, ETH, IND, NGA have rows only for 2011–2022 (12 of 23 years); 2000–2010 rows are absent (likely from the narrower extraction). Not looked up/imputed (rule). These countries are excluded from the balanced set.

Consequence: the balanced set is 15 countries, all high-income OECD plus Türkiye; all four low/lower-middle-income countries drop out. The 2022 all-available baseline includes them (all 20 countries have a 2022 value; key n_countries_2022_available), so it is not comparable to the balanced mean.

No input violates the rules; the only file used is the panel CSV. README.md used for provenance only.