[tannevaled]

What

The README carried three figures for the same quantity, and they disagreed
with each other and with the repository:

where	claimed
the badge (line 8)	22,488
prose, 3 places	23,471
scripts/conformance/rubyspec/BASELINE, summed	23,481

The badge was 1,015 low and carries no date, so it read as current. The
prose at the top — "It runs 23,471 of ruby/spec's examples" — and in the
architecture section read as current too. All three now come from BASELINE,
the per-file record CI actually enforces.

The table row is deliberately left alone

It keeps 23,471, because the table declares itself "Measured 2026-09-28 on
862a9f3, darwin/arm64" against a pinned corpus. A dated measurement going
stale is not an error; an undated claim going stale is.

Two further reasons not to touch it:

• its other rows (failing, erroring, skipped, ran, share) are not derivable
  from BASELINE, which records passing counts and nothing else — so
  refreshing one cell would have made the table internally inconsistent
  (pass + fail + error would stop equalling ran);
• re-running the full sweep to refresh all of them would have made the docs
  less true right now: the machine is at load ~10, and the FILEFAIL/timeout
  rows are exactly what load moves.

A clause beside the table now says which number tracks what, so a reader is not
left with two figures and no explanation.

Machine-checked from here

A number a human retypes is a number that rots.
TestReadmeConformanceCountMatchesTheBaseline reads BASELINE, sums the OK
rows, and fails with the figure to write:

README badge claims 22,488 ruby/spec examples, BASELINE sums to 23481
  update the badge to 23,481
README.md:15 claims 23,471 examples, BASELINE sums to 23481

That output is its known-bad control — putting the old numbers back makes it
name both sites and the correct value, and restoring them makes it pass.

The rule it applies is the table's own: every bold thousands-separated number
outside a table row must equal the baseline total. There are exactly three
today and all three are this count; a future bold number about something else
should be made unambiguous rather than have the rule widened.

Also audited, and found correct

• the Go badge (go-1.26.4+) matches go.mod's go 1.26.4; CI's
  go-version: stable is above that floor, so there is no contradiction;
• scripts/conformance/rubyspec/README.md already says the gate "used to be"
  a FLOOR file — past tense, correct since conformance: switch the ruby/spec gate to the per-file baseline (#710) #739;
• the front-end campaign section already carries its own date and an explicit
  "have not been re-measured since" note;
• the table's 2,202 of 2,206 produced a result; 4 produced none matches
  BASELINE exactly today (2202 OK rows, 4 FILEFAIL).

Green

go test ./... -count=1 -timeout=90m exit 0 (10 packages), gofmt clean,
go vet clean.

🤖 Generated with Claude Code