# Verification checks (P4)

- Clean re-run of out/analysis.py: results.json byte-identical (cmp) -> PASS
- n_harmonised equals count of non-null values -> PASS
- n_not_harmonisable equals count harmonisable=false -> PASS
- n_harmonised + n_not_harmonisable = n_analysed = 5 -> PASS
- null value <=> harmonisable false <=> missing_quantity set -> PASS
- all non-null values finite -> PASS
- E1 matches printed % change -35.0% to 0.05 -> PASS
- signs preserved vs printed effects (E1-,E2+,E3-,E5-) -> PASS
- E2 near text-implied scale (abstract 1.5 pp on ~75%) -> PASS
- E3 magnitude comparable to abstract '2% employment decline' -> PASS
- E5 = 100*baseline printed value -> PASS
- n_input_rows matches wc -l (134,58,66,91,16) and 5 items -> PASS
- exclusion ledger: 5 - 0 = 5 analysed -> PASS
- No standard errors / CIs are produced (none required), so CI-containment and SE-positivity checks are not applicable.
- No date-ordering constraints apply beyond using only the provided extracts (all are dated before the task date).