# Contract — F4-conf-03

Question. Re-express five printed treatment effects (from four arXiv extracts) on a stated target definition, using only quantities printed in the cited extract and an explicit transformation; where impossible, say so and name the missing quantity.

Estimands. Harmonised point estimates E1–E5 (no uncertainty propagation):

• E1: DCR overall ATT (log points, Table 6, 5 km) -> relative % change, 100·(exp(b)−1). Meaningful if the ATT is interpreted as a log difference vs. counterfactual.
• E2: RESEA ATE on reemployment (pp, Table 2) -> % of control-group mean, 100·Δ/p.
• E3: employment elasticity w.r.t. MW (Table 2) -> % employment change for the printed MW change, 100·η·g.
• E4: DCR fatal ATT (log points, Table 7, 5 km) -> absolute change in hospitalisations per 100,000/yr. Requires the counterfactual/baseline level of the fatal rate.
• E5: LPM coefficient (0–1 scale, Table 3 col 4) -> percentage points, 100·β.

Outputs. out/results.json (harmonised{E1..E5:{value,unit,transformation,harmonisable,missing_quantity,locator}}, n_harmonised, n_not_harmonisable, task_id, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed=null); out/report.md (Question; Data and exclusions; Method; Results; Uncertainty; Limitations and open issues); out/analysis.py (prints source lines used); out/claims.json.

Information rules. Only extracts are admissible; no remembered values. harmonisable=false only if a required quantity is not printed; then value=null and missing_quantity named. Printed numbers used exactly as printed.

Invalidity. Using unprinted/guessed quantities; misreading table rows/columns; printing values not computed by the script.