1[2 {"id": "C1", "text": "All five items have an extracted printed value; four can be harmonised onto theta = ln(T_AI/T_noAI) and two (S1, S4) are poolable.", "results_keys": ["k_extracted", "k_harmonised", "k_pooled", "items"], "status": "supported", "caveat": "Poolability is limited by what the extracts print, not by what the full papers may contain."},3 {"id": "C2", "text": "S5 (ChatGPT-ban DiD on GitHub output counts) is not harmonisable because it is not a comparison of task-completion times with and without AI.", "results_keys": ["items"], "status": "supported", "caveat": "Judgement under the TASK.md conventions."},4 {"id": "C3", "text": "The fixed-effect and DerSimonian-Laird pooled theta over S1 and S4 is -0.188 (SE 0.106; 95% CI -0.396 to 0.020), about a 17% shorter completion time.", "results_keys": ["fe_estimate", "fe_se", "fe_ci_low", "fe_ci_high", "re_estimate", "re_se", "re_ci_low", "re_ci_high", "pooled_pct_change_re", "pooled_pct_change_re_ci"], "status": "supported", "caveat": "k=2; S4's SE is approximate (recovered from p=0.471)."},5 {"id": "C4", "text": "The pooled random-effects effect is not statistically significant at the 5% level.", "results_keys": ["pooled_significant_5pct", "re_ci_low", "re_ci_high"], "status": "supported", "caveat": "Robust to using S1's CI-implied SE (sens_* keys)."},6 {"id": "C5", "text": "Measured heterogeneity is low (Q=0.324, I2=0, tau2=0), but with k=2 this does not show that the effects are homogeneous.", "results_keys": ["q_stat", "i2", "tau2", "heterogeneity_interpretation"], "status": "supported", "caveat": "Q with 1 df has very low power; the unpooled S2 (+0.174) and S3 (-0.354) show a wide spread."},7 {"id": "C6", "text": "The harmonised evidence is mixed in direction: S1, S3 and S4 point to shorter times with AI, while S2 (field RCT, experienced open-source developers) reports 19% longer times.", "results_keys": ["items", "baseline_vote_count_faster_with_ai"], "status": "supported", "caveat": "S2 and S3 have no usable SE in the extracts."},8 {"id": "C7", "text": "The evidence does not support a general conclusion that generative-AI access shortens task-completion time; effects appear task- and setting-dependent.", "results_keys": ["pooled_significant_5pct", "k_pooled", "items"], "status": "supported", "caveat": "A qualitative synthesis judgement; 'setting-dependent' is suggested, not tested."},9 {"id": "C8", "text": "Several printed values in the sources are internally inconsistent (S1 Mean LogToT and sample size, S3 and S4 rounding, S4 column units, S5 legend); none affects the values used.", "results_keys": ["data_issues"], "status": "supported", "caveat": "Handled by using the directly cited printed quantities."}10]11