# Suite17 Robust Forward Complementary Specialist Boosting Implementation Plan

**Goal:** Implement the approved Suite17 architecture with chronology-safe R/M/L specialists, adaptive forward selection, dense OOF ledgers, robust temporal admission, conservative corrections, score alignment, and complete diagnostics.

**Architecture:** Reuse Suite16 canonical loading, structured cache, mastery store, HGB/Newton learner, and evaluator. Replace fixed S1/S2/S3 sequencing with a candidate registry and greedy robust forward selection. Add a faculty-local evidence generator and R-view decomposition/ablation while removing the homemade GRANDE path.

**Tech stack:** Python 3, NumPy, pandas, joblib, scikit-learn HistGradientBoosting, SciPy sparse/SVD, pytest.

## Tasks
1. Build R variant decomposition and preserve M view; test exact feature-family isolation and finite outputs.
2. Build chronology-safe faculty-local all-history/recent predictors and L disagreement features; test no future-row use and fallback behavior.
3. Upgrade temporal statistics/admission to retain mean/median/recent/worst-window downside information; test Fold-3-like rejection.
4. Implement dense cross-fitted HGB score ledgers with identity fallback and 100% coverage.
5. Implement greedy forward specialist selection across R/M/L, maximum two default stages, deterministic tie breaking, and optional strict third-stage diagnostic.
6. Implement positive affine logit alignment to Committee target distribution and verify exact ranking preservation.
7. Integrate fold runner, standalone audits, R ablation, forward selection, operational evaluator, Council comparison, and extensive CSV/log outputs.
8. Add synthetic end-to-end CLI fixture covering R/M/L selection, identity fallback, alignment, and result artifact generation.
9. Verify full tests, py_compile, shell syntax, CLI help, extracted ZIP smoke and ZIP integrity.
