step,rule,n_before,n_after,n_removed,rationale
0,All study items in data/studies.json assessed and extracted,5,5,0,No study dropped from assessment; all 5 have a printed value extracted
1,Harmonisable onto theta = ln(T_AI/T_noAI),5,4,1,S5 removed: DiD effect on GitHub output counts around a ban - not a completion-time comparison on the same tasks and not randomised
2,Poolable (SE obtainable from printed quantities),4,2,2,S2 removed (CI only in non-reproduced Figure 1); S3 removed (significance stars only); S1 (printed SE) and S4 (approx SE from p-value) pooled