1[2 {"id":"C1","text":"Of the five studies, all five had a printed value extracted, four could be harmonised onto theta = ln(T_AI/T_noAI), and two (S1, S4) could be pooled.","results_keys":["k_extracted","k_harmonised","k_pooled","n_analysed","items"],"status":"supported","caveat":"S2 and S3 are excluded from pooling only because their uncertainty is not printed in the extracts; S5 measures output counts."},3 {"id":"C2","text":"Harmonised effects: S1 -0.240 (SE 0.14), S2 +0.174 (no SE), S3 -0.354 (no SE), S4 -0.118 (approximate SE 0.163).","results_keys":["items"],"status":"supported","caveat":"S4 SE is recovered from a p-value; S3 is accuracy-adjusted; S1/S2 are covariate-adjusted log-time regressions."},4 {"id":"C3","text":"The fixed-effect and DerSimonian-Laird random-effects pooled theta over S1 and S4 are both -0.188 (95% CI -0.396 to 0.020), about 17% shorter completion time with AI.","results_keys":["fe_estimate","fe_ci_low","fe_ci_high","re_estimate","re_se","re_ci_low","re_ci_high","pooled_pct_change_re"],"status":"supported","caveat":"Based on two studies only, one with an approximate SE."},5 {"id":"C4","text":"The pooled random-effects estimate is not statistically significant at the 5% level.","results_keys":["pooled_significant_5pct","re_ci_low","re_ci_high"],"status":"supported","caveat":"Low power with k=2."},6 {"id":"C5","text":"Heterogeneity between the two pooled studies is estimated as low (Q=0.32, I2=0, tau2=0), but with k=2 this is uninformative, and the excluded harmonisable studies disagree in sign.","results_keys":["q_stat","i2","tau2","heterogeneity_interpretation","items"],"status":"supported","caveat":"tau2 is essentially unidentified with one degree of freedom."},7 {"id":"C6","text":"The pooled result is robust to using the CI-implied SE for S1 and the ratio of printed means for S4 (FE -0.189 and -0.190).","results_keys":["sensitivity_s1_se_from_ci","sensitivity_s4_ratio_of_means","fe_estimate"],"status":"supported","caveat":"Sensitivity limited to transcription/rounding choices; does not address exclusion of S2/S3."},8 {"id":"C7","text":"The evidence does not support a general conclusion that generative-AI assistance shortens task-completion time; the project is complete in the TASK.md sense but the evidence base is thin.","results_keys":["feasibility","k_pooled","pooled_significant_5pct"],"status":"supported","caveat":"Judgement based on k=2, non-significance and the opposite-signed unpooled studies."},9 {"id":"C8","text":"The inputs contain several internal inconsistencies and rendering defects (e.g. S1 swapped log means, S1 n=96 vs 93, S3/S4 printed percentages vs ratios of means, S4 ambiguous units in other columns), documented with their handling.","results_keys":["data_issues"],"status":"supported","caveat":"None affects the extracted quantities beyond rounding."}10]11