# Suite11 Opportunity-Aware Latent Domains Implementation Plan

**Goal:** Build a chronology-safe cross-faculty latent-domain MoE that discovers correction-opportunity domains from historical OOF behavior, predicts them prospectively, trains dedicated residual specialists, and reports directly comparable metrics against prior systems.

**Architecture:** Reuse the Suite5 cache and Suite7 backbone/checkpoints. Discover NULL + K domains from OOF committee/base-expert utility signatures, distill the resulting labels into an observable-feature router, train Suite7-initialized residual experts on soft domain responsibilities, select specialist authority chronologically, and retain exact Suite7 fallback.

**Tech Stack:** Python, NumPy, pandas, scikit-learn, PyTorch.

## Tasks
- [x] Unit-test opportunity signature construction, latent discovery and soft responsibilities.
- [x] Unit-test latent specialist loss, router output and authority shrinkage.
- [x] Unit-test router diagnostics, residual correlation and top-budget rescue transitions.
- [x] Implement fold trainer with train/reliability/selection chronology.
- [x] Implement exact prior-score loaders for Committee, Suite7/8/9/10, V2 and Suite4 Council.
- [x] Implement fair common evaluator: fold/mean-fold/pooled AP-AUC-budget metrics, fixed/drift Precision-Recall-F1-F2, full confusion matrices and aggregate TP/FN/FP/TN.
- [x] Save domain/router/specialist diagnostics and print concise research summaries.
- [x] Run syntax/unit/synthetic smoke verification and package suite11 zip.
