record(ENG-RECORD-CONFLICT-SURFACES): retire the dangling check-agent-record.py line 1973 citation from row 898 and ISSUE-GH-2317

[phantomic12]

record(ENG-RECORD-CONFLICT-SURFACES): retire the dangling check-agent-record.py line 1973 citation from row 898 and ISSUE-GH-2317

The #2317 archive row cites scripts/check-agent-record.py line 1973 as the
home of the INDEX_PREAMBLE constant, and the record quoting the row carries
the same cite twice, in its Title and in its Frozen archive evidence block.
test_every_tracked_checker_line_citation_still_resolves resolves every
citation the tree carries against the tracked scripts, and all three
references dangle: the checker is 1253 lines at this stack's base, so the
cited line is 745 past end-of-file, and the gate listed the archive row plus
both record instances among its subfailures. A citation the gate cannot
resolve is a claim no reader can verify.

The row was true when written. Measured 2026-08-29 at origin/main df024dc
while unblocking #2303, and that tree's checker is 2254 lines with
INDEX_PREAMBLE opening at line 1972 -- the multiline constant froze the
archive preamble so it could not drift without a deliberate edit on both
sides, and 1973 is its second line. W6 (7dc2ef1) then retired the live
index and the constant with it: the checker today carries no INDEX_PREAMBLE
and no merge prose, the preamble sentence lives in the archive itself, and
nothing gates it. No existing line satisfies the original claim, so a
re-point would have lied; the repair re-anchors instead of re-pointing.

All three instances move in lockstep to one new sentence that keeps the
factual claim and retires the dead anchor: the checker of the day pinned the
expected preamble as a constant, INDEX_PREAMBLE, so correcting it was itself
a gated change; W6 retired the live index and that pin with it, and
correcting the preamble is today an edit to the archive and nothing else.
The record's quote stays byte-equal to the row, which is the frozen-evidence
contract this stack enforces, and the two edits are deliberately ONE commit
because a matching-pair move has no valid intermediate state:
agent-issue.py update validates the record against the archive at every
step, so record-first and archive-first each fail validation -- the pair
moves together or not at all.

Red-first against the citation gate: on the stack tip the test failed 24
subtests, three of them the dangling 1973 references; after the repair it
fails 21 -- exactly the pre-existing baseline, none of them ours -- and the
tree-wide record checker (rc=0) and the frozen-evidence census (831 records
with a block, 831 resolve, 831 quote their own issue, 0 violations) are
unchanged. The gate also resolves N-M range spellings, so the new sentence
carries no citation-shaped text at all, and historical line anchors in this
message are written out in prose for the same reason.

The row is filed under ENG-RECORD-CONFLICT-SURFACES (#364) as
ISSUE-LOCAL-01M3Q0N3Y2C29368C30S90JXGR (kind record), beside ISSUE-GH-2317
in the same directory. Stacked on #3350 through #3353 (base 9e17af3); the
restored archive is the only reason row 898 is tracked again and its
citation is visible to the gate at all.

ISSUE-LOCAL-01M3Q0N3Y2C29368C30S90JXGR

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:codebuff/buffy [freebuff]