Excerpt from PROVER-CANONICAL/receipt.txt; source captured 2026-10-04.

[measured] Dead-path census, recomputed this session (song-ref, read-only):
  git ls-tree -r --name-only 6bc87e8e -- src/prover            ->  76 live paths
  FIELD SEPARATOR MATTERS and my first attempt was wrong: awk's default whitespace
  FS DROPS every numstat row whose path contains a space, silently under-counting.
  Tab-FS is the census:
  git log --format= --numstat 6bc87e8e -- src/prover | awk -F TAB 'NF==3 && $1!="-" {c[$3]+=$1+$2}'
                                                          -> 302 paths, 162126 churn
  join (live tree vs churn table)   -> 226 dead paths, 97757 dead churn
  DEAD SHARE = 97757 / 162126 = 60.30%    (whitespace-FS gave 60.26%)
  live churn 64369
  variant `git log --all ... -- src/prover` -> 296 paths, 164842 churn
  variant `+ lean/Song/proof`               -> 480 paths, 266886 churn
  ../song-ref-gen1 @ ae688f66, `git log --all -- src/prover`  -> 296 paths, 192438 churn

[OPEN, discrepancy — named, not reconciled] The recorded figure is 60.75% of
  172218 (lessons.tsv L12; ledger residual on PROVER-LESSONS). I could not
  reproduce 172218 with any flag combination above; my reachable-history figure is
  162126 (60.30%) at song-ref HEAD and 164842 with `--all`. The two dead-share
  figures agree to 0.45 percentage points, so the SHAPE of L12 holds; the total does
  not. The DEAD/60.75% CLAIM IS NOT RESTATED HERE; both numbers are recorded with
  their exact commands and the difference is 6.0% of total churn. The 226-vs-215
  dead-path count is the same tab-FS effect. What §2 relies on is the measured