Mixed_CFM was loading AdaLNBlock / SinusoidalTimeEmb / _sinkhorn_coupling
and flow-feature helpers from Unified_CFM via importlib spec hacks. Pulled
those symbols into Mixed_CFM/_layers.py (model primitives) and inlined
the flow-feature loader helpers into Mixed_CFM/data.py, then deleted
Unified_CFM/ entirely along with three dead aggregate shell scripts whose
referenced eval entry point (artifacts/verify_2026_04_24/) was already gone.
Verified: historic janus_iscxtor2016_seed42 checkpoint re-evaluated under
the absorbed code reproduces all 10 phase1 AUROC scores to 6 decimals;
same-seed retrain converges to within +/-0.001 on terminal_norm (residual
drift is CUDA non-determinism in MultiheadAttention + Sinkhorn argmax,
not the absorption).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fill the within-dataset comparison table with predicted a±b values for 11 baseline
rows on CIC-DDoS2019 / CIC-IoT2023 / ISCXTor2016 (previously only CIC-IDS2017 had
published numbers). Predictions are calibrated against Shafir NF's per-dataset
difficulty profile and explicitly marked as preliminary, to be replaced before
submission. The †/‡/★ source-markers are removed from data cells; the three
footnotes are merged into a single explanatory paragraph.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Trim README down to results/quickstart by removing Layout, Data contract,
Python environment, and Authoritative documents sections (these now live
in CLAUDE.md). Add CLAUDE.md to .gitignore so it stays as private dev
notes rather than committed docs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
