# Verification checks (P4)

• Clean re-run: deleted out/results.json, re-ran analysis.py, byte-identical to previous output (cmp): PASS
• No CIs/SEs produced (no uncertainty propagation required), so CI-contains-estimate and SE>0 checks are not applicable.
• Information rule: every input is a printed cell of the cited extract (line numbers in locators); E4 baseline not substituted.
• 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 non-null: PASS
• E1 computed rounds to printed -35.0%: PASS
• E1 sign negative like log-point ATT: PASS
• E2 recompute 100*1.47/75.0: PASS
• E3 recompute 100*-0.388*0.047: PASS
• E3 magnitude close to abstract 'about 2% employment decline': PASS
• E5 recompute -0.0962*100: PASS
• all non-null values finite: PASS
• random_seed null, task_id correct: PASS
• n_input_rows matches wc -l (66,91,58,134,16) and 5 items: PASS
• exclusions: 5 raw - 0 removed = 5 analysed: PASS