Column repair path-coverage count correction

This narrow successor preserves bc66591 and the original column/CI repair receipts unchanged. Their claimed 23 examples and 52 event/path controls were incorrect. The bound source has 22 tracked examples. The original test walked those examples plus two verifier paths and one final-PNG evidence path for each of pull_request and push: 25 paths per event, 50 controls. The frozen summary counter overstated coverage by two. Its counter is not qualifying evidence.

The current test now builds an explicit ledger from this checkout’s tracked Git inventory and executes its assertions over those exact rows. An instrumented runner records that test’s actual ledger normally and under optimized Python. The new receipt and verifier derive all counts from the executed rows and the frozen source tree; neither accepts a hard-coded summary count. The historical 52 claim is explicitly superseded, not edited into apparent prior success.

The all-example and column verifier/demo path filters, 30-minute budget, every original Rust mode, smoke/lifecycle assertion and job layout remain unchanged from bc66591. No Rust, Cargo, native evidence, old figure, physical gate or forcing-feature source is changed. Prior successful artifact-corruption and local CI runs are preserved; this correction tests the changed path-ledger method and source/evidence binding. It does not repeat an unchanged hosted job. A new hosted CI completion remains required. Older viscosity vector-transition and later wet-pressure qualifications remain separate, unresolved limits.

The packet also binds the exact ordered five-commit parent chain from 35b00247 through bc66591, including source 6e62e2b9, its parent 300366af and evidence d08ad5cd. This lane remains separate from uniform forcing.