`check` has no aperture for a ticket slug cited in SOURCE

Source comments cite ticket slugs constantly and the convention is a good one — feature-nilpy-dotted-package-imports, bug-nilpy-omitted-variant-default-segfaults and ~2100 others resolve correctly. Nothing verifies them.

check already has the right idea one folder over: DANGLING-LINK reports a [[wiki-link]] in a ticket body whose prefix looks live but resolves to no ticket, and its text is careful about the five repairs (rename, planned-never- filed, delivered-under-another-name, merged, never-a-ticket). All five apply verbatim to a source comment. The aperture just never opens a .pas or .inc.

How it was found, which is the part worth keeping

Not by auditing. I wrote a comment in compiler/pyparser.inc citing bug-n-a-stdlib-class-folds-onto-a-same-named-function-and-silently-evaluates-to-its-result, a slug I had composed while writing the comment and never filed — I had FIXED the bug instead, which is what this repo asks for. check ran green. The citation is unreachable for every future reader and no instrument says so.

A fix-don't-file culture manufactures this steadily. The natural way to record why a line is the way it is, while fixing rather than filing, is to cite the bug in a comment — and there is then no ticket for that slug to resolve to.

The measurement, with its population, so a re-run can disagree usefully

Tree 3d3d90a2d. Population compiler/** + lib/**, .pas and .inc only. tools/** is EXCLUDED and must stay excluded: progress_near_corpus.py is a corpus of example slugs for the near-dup checker and *_devtest.sh carry fixtures like bug-a-a-throwaway, none of which are citations. Slug shape is (bug|feature|task|decide|refactor|perf)- plus three or more dashed segments, matched against the 4959 .md filenames under devdocs/progress/*/.

distinct slug-shaped citations : 2472
resolve exactly                : 2119
prefix of a real slug          :  158   <- LINE-WRAPPED in a comment, not a defect
no real slug begins with it    :  195   <- the finding

The 158 are not a bug and must not be counted as one. A comment wraps mid-slug and the regex sees a prefix. Any re-implementation that skips this split will report 353 and be wrong by 45%.

Worst files: pyparser.inc 58, pasparser_expr.inc 18, builtin/pylib.pas 17, symtab.inc 11, ir.inc 11, pasparser_decl.inc 11.

The 195 are NOT mostly missing tickets, and the repair differs

Two sampled, both RE-WORDINGS of tickets that exist:

lib/rtl/pil.pas cites bug-a-a-class-named-after-a-used-unit-silently-turns-its-own-constructor-into-a-self-call; filed as bug-a-a-class-named-after-a-used-unit-cannot-be-constructed-from-outside-that-unit.
bug-a-a-method-pointer-record-is-hard-sized-on-32-bit-targets; filed as bug-a-method-pointer-record-is-hard-sized-16-bytes-on-32-bit-targets.

So the dominant repair is correct the spelling, not file the ticket — and a checker that reports these as "missing tickets" would send someone to re-file work that is already tracked. Report them as UNRESOLVABLE CITATIONS and let the reader pick among the five DANGLING-LINK repairs.

The sample is 2 of 195 and the proportion is NOT established. Do not quote a split between renames, never-filed and delivered-elsewhere until someone has classified a real sample; the only measured claim here is that 195 do not resolve and that at least two are renames.

Why 40 and not higher

Nothing is mis-ranked and no program misbehaves — a stale citation costs a reader a grep and the wrong conclusion. It earns a number above the floor because the population is large, it grows every time someone follows this repo's own fix-don't-file rule, and the failure is silent in the direction of a reader deciding a bug was never reported.

What would retire this

check opening compiler/** and lib/** for slug-shaped tokens and reporting the unresolvable ones under the existing DANGLING-LINK wording. Two things it must get right or it is worse than nothing:

Exclude the prefix-of-a-real-slug case, or it reports 45% noise on its first run and teaches everyone to ignore it.
Exclude tools/ corpora and devtest fixtures by path, not by heuristic; they are deliberately full of fake slugs, and a checker that reds on its own test corpus is born red.

Positive control: lib/rtl/pil.pas's bug-a-a-class-named-after-a-used-unit-silently-turns-its-own-constructor-into-a-self-call must be reported. Verified unresolvable at 3d3d90a2d, with the real ticket (...cannot-be-constructed-from-outside-that-unit) sitting right beside it, so it exercises the rename case rather than the never-filed one.