docs(guide): make three drifted counts on ci-cd-pipeline.md self-maintaining - #8121

Fixes #7965

Three sentences on content/docs/guide/ci-cd-pipeline.md qualified a file/document population with a
copied numeral and no declared measurement point. Each numeral was re-taken from that figure's own
gate verdict line — never re-counted by hand — and each sentence was then rewritten so it stays
true as the tree moves, rather than being refreshed to today's number.

Base 59a3a233d · head 6528b3fe4 · one file changed, 10 insertions / 7 deletions.

The verdict lines these fixes rest on (verbatim, run on this branch)

node scripts/check-doc-component-types.mjs (exit 0):

Scanned 188 doc file(s) (.mdx + .md), 1105 code block(s), 889 `type` literal(s) against 658 registered key(s) derived from 143 source file(s) (240 resolved call site(s), 133 indirect, 2 open): 772 registered, 117 exempted; 4 key table(s), 24 row(s), 45 table key(s) judged (namespaced + bare), 45 registered.
✅  Every documented component type is registered.

node scripts/check-shell-escape-residue.mjs (exit 0):

✅  check-shell-escape-residue: OK (5/5 root(s) resolved -- AGENTS.md: 1 file(s), 15 fence(s); CLAUDE.md: 1 file(s), 0 fence(s); skills: 16 file(s), 210 fence(s); .claude/skills: 4 file(s), 20 fence(s); content/docs: 184 file(s), 1064 fence(s); 206 file(s) and 1309 fenced block(s) examined in total; 0 occurrence(s) outside a fence (counted, not judged); coverage -- .claude: 4/4; skills: 16/16 document(s) under a declared root).

pnpm exec vitest run scripts/__tests__/ at head 6528b3fe4 (exit 0):

 Test Files  112 passed (112)
      Tests  3360 passed (3360)
   Duration  104.70s

Difference from the card. The card measured 206 files / 889 literals / 107 test files. Two of the
three still read the same today (206 total files with the four-file .claude/skills root inside it;
889 literals). The test population has moved again since the card was written: 112 test files and
3360 tests, not 107. That movement inside one working day is itself the argument for not writing a
bare numeral here.

The three sentences

1. :782 — "measured across 143 files and 558 literals" → route (a), live reading

Before:

How a snippet is judged. type is not one vocabulary in these pages — measured across 143 files
and 558 literals, the corpus spells action schemas, block schemas, ...

After:

How a snippet is judged. type is not one vocabulary in these pages — measured across every
.md and .mdx page under content/docs, a population the gate re-derives on every run and prints
in its own verdict line beside the code-block and type-literal counts, the corpus spells action
schemas, ... — so that size is read off the run, never copied here.

Why (a) and not (b). The sentence's job is to say the corpus is large and mixed, not to state a
size, so the numeral was load-bearing for nothing — and the pairing was already wrong in a way a
refresh would have preserved: 143 is the gate's source-file count (the registry derivation),
while the literals come from the doc files it walks, which the same line reports as 188. A
sentence reading "in these pages ... measured across 143 files" therefore pinned a doc-corpus claim
to a source-file count. Naming the population (content/docs, .md + .mdx) and pointing at the
verdict line removes both the drift and the mis-pairing. The trailing clause is deliberate: it tells
the next reader why there is no number here, so the sentence is not "repaired" back into one.

2. :1083 — "the other four still return 203 of today's 207 files" → route (a), relation restated

Before:

The document floors are per root and deliberately never a whole-surface total: on a day the
four-file .claude/skills root reads zero, the other four still return 203 of today's 207 files, so
a total floor stays green through the entire outage.

After:

The document floors are per root and deliberately never a whole-surface total: on a day the
smallest root, .claude/skills, reads zero, the other four still return all but its handful of
documents, so a total floor stays green through the entire outage — the relation is the argument,
and each root's own population is printed on every run rather than copied here.

Why (a). This paragraph's point is a RELATION — a per-root floor survives one root reading zero, a
total floor does not — and 203 of 207 was only an illustration of it. Restating the relation makes
the argument independent of any count, and the gate already prints the per-root census (AGENTS.md: 1 file(s) ...; .claude/skills: 4 file(s) ...) on every run, so a reader who wants the sizes gets
today's, not a copy. A measurement point (route b) would have worked here too but reads worse: it
would have anchored two numerals in a sentence whose whole force is that the numerals do not matter.

3. :1440 — "73 files / 1996 tests" → route (b), declared measurement point

Before:

| pnpm test scripts/__tests__ | 73 files / 1996 tests — every test that reads a manifest version, QUICK_REFERENCE.md or a CHANGELOG.md | ~50 s |

After:

| pnpm test scripts/__tests__ | 112 test files / 3360 tests on main@59a3a233d — every test that reads a manifest version, QUICK_REFERENCE.md or a CHANGELOG.md | ~50 s |

Why (b). In a narrow table cell the route-(a) phrasing ("the run prints its own totals") displaces
the cell's actual job, which is to say what the command covers and roughly how big it is. A declared
measurement point keeps the sizing and makes the sentence permanently true, exactly the way :576's
Measured on main@6422aa891 does two sections up.

Subject kept accurate. This is a TEST-file population, not a document population. The cell now
says "112 test files", not "112 files", so nobody reads it as a docs count later.

What happened to the ~50 s cell. Nothing — it is untouched. It is a duration, not a count, and
only the middle cell was rewritten. Recorded for the reviewer: the two runs on this branch took
116.6 s and 104.7 s, but that is a shared container running several agents in parallel, so it is not a
measurement comparable with whatever produced the ~ figure. Changing it on this evidence would be a
guess, which is the failure mode this card exists to stop.

The two correct neighbours did not move

Both declare the tree they measured, so both are still true and neither is a defect:

• :576 — Measured on `main@6422aa891`: 18 guide files, 91 candidate spans ...
• :1658 — Across all 5281 first-parent commits on `main` ...

Proof, and it is mechanical rather than a promise — the diff has exactly three hunks and none of their
ranges contains either line:

$ git diff -U0 59a3a233d..6528b3fe4 -- content/docs/guide/ci-cd-pipeline.md | grep '^@@'
@@ -781,3 +781,5 @@
@@ -1082,3 +1084,4 @@
@@ -1440 +1443 @@

Line 576 is untouched at 576; the neighbour formerly at 1658 is byte-identical and now sits at 1661,
shifted only by the three lines the second hunk added. Both were compared against
59a3a233d:content/docs/guide/ci-cd-pipeline.md directly and match.

No gate, no pin, no new scan population

One file changed. No gate script, no workflow, no test, no other page, nothing under packages/. The
coverage-expansion work that would make this class fail loudly on content/docs is objectui#7966's
family and is deliberately not started here.

Gates (all at head 6528b3fe4, exit codes captured by redirect-then-capture)

The gate list was re-derived against the actual diff rather than taken from the dispatch: every
script under scripts/ that reads content/docs was enumerated and run, which is where the last
four doc gates in the table came from.

Live E2E (informational) is red on every branch today for an upstream reason unrelated to this
change.

Generated by Claude Code