Channel Correlation Diagnostic

A real, standing diagnostic that tracks pairwise correlation between the four channels across actual accumulated history — kept deliberately separate from the estimator suite, so channel separation stays checkable rather than folded into any single score.

What it does

For every pair among S/D/I/C, compute_channel_correlation() computes the real Pearson correlation across the run log's accumulated history, not a single snapshot. If the strongest pairwise correlation exceeds 0.7, it raises a contamination warning — a signal that two channels meant to stay structurally independent are starting to move together, which is exactly the failure mode the whole four-channel architecture exists to prevent. Requires at least 10 data points before it will report anything at all, rather than draw a conclusion from too little history.

Verified by direct execution

Real channel correlation, computed across 30 actual readings:
  S-D: r=+0.3207
  D-I: r=+0.2865
  S-I: r=+0.4397
  D-C: r=+0.0352
  S-C: r=+0.3251
  I-C: r=+0.3142

Contamination warning: False

All six pairwise correlations well under the 0.7 warning threshold on this run — the channels are behaving independently, not collapsing into each other. This is a live, repeatable check, not a one-time report: run it again on a different history and it gives a real, potentially different answer, which is the point.

Where this fits

This is the real dependency an earlier, more elaborate Channel Divergence Monitor was built on top of — that specific extension isn't currently available to republish, but this diagnostic doesn't need it to be useful on its own. It stands alone as a genuine, ongoing correlation check, and is a natural candidate for wiring into any live estimation pipeline on this site that maintains a run log over time.