From e1db4b036a7632a17c13a22f8a133ba34f3d5630 Mon Sep 17 00:00:00 2001
From: "Jonathan D.A. Jewell" <6759885+hyperpolymath@users.noreply.github.com>
Date: Mon, 21 Sep 2026 15:28:54 +0100
Subject: [PATCH] fix(strip-history): strip inputs/ too, and print what
 actually remains

The previous strip set left 71.58 MiB behind while every check reported
success. Measured in a scratch clone: 269.43 MiB -> 81.13 MiB, with
"VESPA/Multiplex blobs remaining (expect 0): 0" printed truthfully, and
71.58 MiB of that pool's content still present under inputs/fastq/.

The cause is structural, not a typo in the path list. `git rev-list
--objects` emits each object EXACTLY ONCE, paired with one of the paths it
is reachable under. Those two blobs lived at both data/Multiplex_pool/ and
inputs/fastq/; the census that chose the strip set saw only the first name,
so inputs/ never appeared in it at all. Removing the path did not remove the
content, and the post-strip grep for the stripped path returned 0 because it
could not have returned anything else. A check that asks "is the path I just
deleted absent" is a tautology, not evidence.

So this does two things:

  - adds `inputs` and `logs_*.zip` (a committed CI log artefact no path class
    covered) to the strip set. Re-measured: 269.43 MiB -> 9.17 MiB, and the
    heaviest remaining path is the 7.71 MiB of LIVE .fastq.gz fixtures, which
    is the correct floor.

  - prints the heaviest REMAINING path aggregates after the rewrite. A path
    strip can never prove a blob is gone, because the blob may have a second
    name; the only honest verification is to look at what survived. The
    predicted figure was ~21 MiB and the run produced 81 MiB -- that gap was
    the entire signal, and nothing in the named checks would have raised it.

The HEAD-tree-identity assertion is kept and still passes. It answers a
different question -- did the strip set catch a LIVE file -- and it was never
capable of detecting dead content that was missed.

Still never runs on its own: refuses without --i-have-read-the-warnings, and
the force-push commands remain printed instructions inside a heredoc.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X3hgXxWm6umMgZkjYyHnnm
---
 scripts/strip-history.sh | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/scripts/strip-history.sh b/scripts/strip-history.sh
index 8148403..278a68a 100755
--- a/scripts/strip-history.sh
+++ b/scripts/strip-history.sh
@@ -113,6 +113,8 @@ git filter-repo --force --invert-paths \
   --path      'data/VESPA_pool' \
   --path      'data/Multiplex_pool' \
   --path      'web' \
+  --path      'inputs' \
+  --path-glob 'logs_*.zip' \
   --path-regex '^data/MiSeq_SOP/.*\.fastq$' \
   --path-regex '(^|/)node_modules/'
 
@@ -146,6 +148,23 @@ else
 fi
 echo "    run the test suite in $WORK before believing any of this."
 
+# A path-based strip CANNOT prove a blob is gone: the same blob can be reachable
+# under a SECOND path, and `git rev-list --objects` names each object exactly
+# once, so the census that chose the strip set credits it to one path only.
+# Measured 2026-09-21: stripping data/Multiplex_pool left its 71.58 MiB intact
+# under inputs/fastq/, and the pre-rewrite listing never mentioned inputs/ at
+# all. The named checks above all passed while 71 MiB survived. So print what
+# actually REMAINS and read it -- an unexpected heavy path here is the tell.
+echo
+echo "    heaviest paths REMAINING (read this; do not trust the checks above alone):"
+git rev-list --objects --all \
+ | git cat-file --batch-check='%(objecttype) %(objectsize:disk) %(rest)' \
+ | awk '$1=="blob" && $3!="" {
+     n=split($3,a,"/"); k=(n>=2 ? a[1]"/"a[2] : a[1]); s[k]+=$2; c[k]++
+   }
+   END { for (k in s) printf "      %10.2f MiB %6d  %s\n", s[k]/1048576, c[k], k }' \
+ | sort -rn | head -8
+
 cat <<'NEXT'
 
 ═══════════════════════════════════════════════════════════════════════════