docs(pm): express the entry guard's self-test size as a floor, not a reading (#14002)

The docblock above the `--self-test` entry guard carried two figures frozen at
PR #9897 and both had rotted:

  * "THIS file's 77 assertions" read as a present-tense claim about this
    file's current suite. It printed 77 at PR #9897 and prints 255 today — a
    factor of 3.3.
  * "the same reason line 810 carries it" pointed at the sibling
    main-invocation guard, which really was line 810 at PR #9897. Line 810 is
    now an unrelated `try {` inside the regeneration predicate; the guard has
    moved to line 2336.

Writing today's numbers would re-freeze the same defect one value later, so
the 77 is now anchored to PR #9897 as the historical fact it is, today's size
is a FLOOR ("more than three times that") stated as a floor on purpose and
pointing at `--self-test`'s `checked` as the live reading, and the line number
is replaced by a reference to the guard itself, which cannot drift.

Measured before choosing the shape, on this file's own history rather than
inherited from the sibling: across all 13 commits that have ever touched it
(introduced 2026-08-18, never renamed), the static `assert(` count went up 10
times, held flat twice, and has never once gone down (22 -> 248 static). The
counting method reproduces the two independent readings on record — 70 static
at a065e46 and 223 at 9c4c431.

The runtime figure is not even constant across environments, which is a second
reason a floor beats a reading: `--self-test` prints 255 in an installed tree
and 253 without node_modules, where the generator-provenance row fails closed
and says so on its `live:` line. Both readings are green and truthful.

No behaviour change: the guard and the self-test are untouched, and the count
is identical before and after this commit (255 at the merge base, 255 here).


Claude-Session: https://claude.ai/code/session_01Msg17tAHJ3jVTYFgHydCm2

Co-authored-by: Claude <noreply@anthropic.com>

Author: os-sam

scripts/pm/check-governed-merges.mjs

@@ -3498,13 +3498,23 @@ function runTestModeExitFor(paths) {
   return testVerdict(paths).governed ? EXIT_TEST_GOVERNED : EXIT_TEST_NOT_GOVERNED;
 }
 
-// `invokedDirectly` for the same reason line 810 carries it: this module is
-// imported for its exported predicates (`proxyRearmPlan` — see
-// scripts/pm/ci-failure.mjs), and an unguarded trigger ran THIS file's 77
-// assertions inside the importer's own `--self-test`, printing a second
-// summary and putting an unrelated file's failures on the importer's exit
-// code. A self-test is a mode of the file that is being RUN, never a side
-// effect of importing it.
+// `invokedDirectly` for the same reason the main-invocation guard above
+// carries it: this module is imported for its exported predicates
+// (`proxyRearmPlan` — see scripts/pm/ci-failure.mjs), and an unguarded
+// trigger ran this file's whole self-test inside the importer's own
+// `--self-test`, printing a second summary and putting an unrelated file's
+// failures on the importer's exit code. The suite was 77 assertions at PR
+// #9897, where this guard landed, and is now more than three times that: the
+// 77 is anchored to that PR and frozen as a historical fact, while the
+// multiple is a FLOOR and is written as one on purpose. The live figure is
+// whatever `--self-test` prints from `checked`; it moves on most edits to
+// this file and has never once gone down across this file's history, so a
+// floor stays true where a reading rots. Both figures that stood here had
+// rotted before anyone looked — 77 was written as THIS file's current size
+// and was by then off by a factor of three, and the neighbouring "line 810"
+// had drifted onto unrelated code — so do not "helpfully" refresh either
+// back into a reading. A self-test is a mode of the file that is being RUN,
+// never a side effect of importing it.
 if (invokedDirectly && process.argv.includes('--self-test')) {
   await selfTest();
 }