Input: main 880ec07ae605d93f4c49e4464d8b4526457cfb65, after PRs #203, #205 and #206. This continues the existing mission; #203 is complete.
Question: can the exact common-period correlations control the finite, attainable-cell, prime-weighted certificate, with its actual drift retained? Deliver one precisely quantified transfer or an exact feasible obstruction, and account separately for the passage from variance to linear excess.
Ownership: Codex, finite correlations. Write new files in this directory and one focused test file; regenerate CONTEXT.md if the test inventory changes. Preserve all earlier outputs. Claude owns the dual-witness and barrier work in certificate_lp_frontier; read its comparison table only.
Use exact small controls and the three saved rational feasible vectors. Prime data evaluate the measure, not a prime-blind coefficient construction. No new optimization, large search, CI experiment, or formalization. Run numerical libraries with one thread and pytest with -n 0 -m "not slow". The milestone ships as an open PR for cross-review, without merging.