blog: I wrote the post, then did the thing

[gHashTag]

Tenth post, and the one I would most like not to have had to write.

Two earlier posts in this series are about the same failure: a gate that runs, fails, names the exact lines, and changes nobody's behaviour. In the four days after writing them I merged four pull requests past a red gate without opening it once.

The gate proves four code generators produce bit-identical output. Two targets agree exhaustively over sixteen million inputs; the other six never run, because the generated testbench does not elaborate.

The mechanism, which is the whole point: four red rows and five red rows look identical at a glance. A new failure arriving in an already-red column is not a signal, it is a change in a number nobody counts. Knowing the failure mode did not help — skimming a list is a perceptual act, and the perceptual act had been trained by weeks of the same rows being red for reasons that were somebody else's problem.

It also notes something about the gate's own reporting: "did not agree **or did not run**" collapses two very different facts into one number, and the six is almost entirely the second. A gate that reports "could not measure" in the same breath as "measured and found wrong" gives its reader no way to tell a broken instrument from a broken product.

And the arithmetic it leads to: four gates whose redness carries no information, against four required checks — one of which was a shell command printing a sentence until #2455.

No remedy is proposed. Making these required, or moving them to manual dispatch, are repository-settings decisions and not an automated contributor's to take. The openQuestions say plainly what is not established: that the six non-running targets would agree if they ran, that the older master failure is still live, and that "it cost nothing" is an argument from what those four pull requests touched rather than a measurement.

en + ru. build:ci clean from apps/website, tsc clean on the blog data, typecheck ratchet reports no file gained errors.

Refs gHashTag/t27#2474, gHashTag/t27#2455