# Data audit (P1)

| file | SHA-256 | data rows | cols |
|---|---|---|---|
| data/wb_indicators.csv | 6823f02ed4c20fd12a54b8ca4454f85ba333f22337e2a9fc57c3cabccfdbfcd0 (matches README & bundle.json) | 2349 | 5 |
| data/wb_entities.csv | 06af6aec15122fecea5be5eb21464c505188ecc135bc55104a40f20f65040f8d (matches) | 296 | 5 |
| data/README.md | 76cb5af488d97339fb6319a5c9a7ba6ec98112ab6a0a8116e7bec6acd1f9f67a (matches bundle.json) | – | – |

**wb_indicators.csv**: iso3 (str), name (str), year (int, 2015–2023, 261 entities × 9 years, balanced), gdp_pc_ppp (float, constant 2021 PPP $, 164 missing, range 992–139,541, all > 0), life_expectancy (float, years, 0 missing, range 18.8–86.4). No duplicated (iso3, year). Every iso3 appears in wb_entities.csv; names agree.

**wb_entities.csv**: 296 entities, no duplicates, no blanks; 79 have region == "Aggregates" (217 countries/economies). Note: "NA" (Namibia's ISO2) is not an issue because codes are ISO3; pandas default NA parsing is nevertheless disabled for string columns in the script for safety (keep_default_na=False) — no entity has a code/name that would be mis-parsed anyway.

**Countries/economies, both indicators observed, by year**: 199 in each year 2015–2022, 197 in 2023. Rule (≥150, latest) ⇒ analysis year **2023**. In 2023, 217 countries present, 20 lack GDP (ASM, CHI, CUB, ERI, GIB, GUM, IMN, LIE, MAF, MCO, MNP, NCL, PRK, PYF, SSD, SYR, VEN, VGB, VIR, YEM); 0 lack LE ⇒ n_excluded_missing = 20, n_analysed = 197.

**Anomalies**
- Central African Republic LE is erratic across years (2019 31.5, 2021 40.3, 2022 18.8, 2023 57.4). The 2023 value (57.4) is within the plausible range for CAF and is kept; earlier years are not used. Flagged in data_issues; a leave-out sensitivity is not required, but CAF's leverage is noted.
- West Bank & Gaza (PSE) LE drops 11.5 years in 2023 (conflict); kept as reported.
- Missing GDP is non-random: mostly small high-income territories and conflict/sanctioned states (possible selection bias). Handled by exclusion per rules, discussed in report.
- 2023 LE values for some countries may be WB estimates/projections rather than vital-registration measurements (not verifiable from the files).

**Information-availability**: no input violates the rules; aggregates are removed via region == "Aggregates". No other files used.