analysis(p595): channel-leverage scan over archived threshold-rank histograms

[LightChainr]

Summary

Zero-new-sampling scan of the readable threshold-rank readout family for maximal |A4|/se(A4), per #595.

• Family scanned (all reconstructable from results/server-20260829/P205-norm5-conjugate-coalescence/raw/): M(p_ref), M on a 15-point p-grid, tail derivatives M'(p)/M''(p), histogram-native functionals (mean rank, gap, variance), and the GE-optimal linear combination solved against the delete-one jackknife covariance.
• Pre-registration (GOVERNANCE 2D/2E) enforced: only se(A4)-equivalent (SE of the orientation difference) and amplification-vs-M are reported. No delta/rho/z/contamination bound for any selected channel. Selected combination validated on held-out batches (split-half control, 40 random splits). Deliverable is a design recommendation with a stated selection-optimism penalty, not a measurement.
• Exact S/D/Sp/Dp obs() channels are not reconstructable from the archived threshold-rank histograms and are declared a prospective buy-back (not fabricated).

Key numbers

• Best apparent readout: varKminus (variance of threshold rank), apparent SNR up to 5.35, apparent amplification vs M up to 65x (325_C_B).
• Transport (snr_heldout / snr_selected), 40 splits: single best readout 0.83-1.00 (real, transportable leverage); GE-optimal combination 0.37-0.74 (dominated by selection optimism).
• Conclusion: spread is not pure noise (one readable channel transports), but the data-snooped combination does not; recommend scoring varKminus prospectively.

Deliverables

• notes/p595-channel-leverage-20260913.md
• scripts/channel_leverage_scan.py
• results/channel-leverage-20260913/ (REPORT.md, metadata.json, commands.txt, raw/, derived/)

Full Matching-One repository CI has not been run for this commit.