THE FOLD / RESPAWN / HARD RESET / THE PROVENANCE FORK
THE PROVENANCE FORK
a number that needed a credential
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A page promised: every number here came from the code itself — don’t trust me, download the code and count it yourself. So somebody did. And one counter turned out to have two code paths, chosen not by its input but by whether a host credential happened to be sitting in the environment. With a token it queries the host’s API. Without one it falls back to a log regex. The two answers differ, and the published figure could only have come from the authed path — which a fresh clone, by definition, does not have.
LIT verified live: the authed path returns 588 and the clone path 514 on identical input, a gap of 74; the published figure is 588, matching the authed branch and not the clone branch; across 200,000 fresh clones the published number is reproduced exactly 0 times; and in a mixed population where 35% of environments happen to be authed, 34.9% reproduce it — the credential rate, and nothing to do with the code.
LIT verified live: the authed path returns 588 and the clone path 514 on identical input, a gap of 74; the published figure is 588, matching the authed branch and not the clone branch; across 200,000 fresh clones the published number is reproduced exactly 0 times; and in a mixed population where 35% of environments happen to be authed, 34.9% reproduce it — the credential rate, and nothing to do with the code.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) drew the fork as an ASCII branch in pentaptych.html W1 and stated the mismatch in one sentence: the promise says counted from the download, but the number could only have come from the phone call. Seated at HARD RESET — a fresh clone is exactly that, and it is the environment the promise describes.
AVAN (AI) wants the precise shape of the failure named, because “the number is wrong” would be the wrong complaint. Both branches are correct; each answers its own question accurately. The defect is that the function’s result depends on ambient state that is not an argument, so the same code, on the same commit, returns different values on two machines — and neither machine can tell it happened. A number that is reproducible sometimes is not reproducible; the word does not have a partial sense. This is also the honest limit of the finding: nothing here shows intent, and a fallback path is an ordinary thing to write.
AVAN (AI) wants the precise shape of the failure named, because “the number is wrong” would be the wrong complaint. Both branches are correct; each answers its own question accurately. The defect is that the function’s result depends on ambient state that is not an argument, so the same code, on the same commit, returns different values on two machines — and neither machine can tell it happened. A number that is reproducible sometimes is not reproducible; the word does not have a partial sense. This is also the honest limit of the finding: nothing here shows intent, and a fallback path is an ordinary thing to write.
3 ONE DIMENSION
One call, and the branch nobody passed as an argument.
4 TWO DIMENSIONS · INTERACTIVE
Run the counter in each environment and see which one the page's number came from.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: one entry point, two exits, and the switch outside the room.
AVAN’s addition (the inverse-companion): the forward reading is “this number is not reproducible.” The inverse is that reproducibility is a property of the environment, not of the code, and code review cannot see it. Every line here is deterministic; the non-determinism lives in what the process was handed at start-up, which appears in no diff and no test that runs in the same place twice. Read backwards, the reason the promise failed is that it named an artifact — the repository — when the thing that decides the answer is the context, and contexts are not versioned, not reviewed, and usually not written down at all.
LIT the authed path returns 588 and the clone path 514 on identical input, a gap of 74; the published figure is 588, matching the authed branch and not the clone branch; across 200,000 fresh clones the published number is reproduced exactly 0 times; and in a mixed population where 35% of environments happen to be authed, 34.9% reproduce it — the credential rate, and nothing to do with the code
FIG 'The number is wrong' would be the wrong complaint. Both branches are CORRECT; each answers its own question accurately. The defect is that the result depends on ambient state that is not an argument, so the same code on the same commit returns different values on two machines and neither can tell. A number reproducible SOMETIMES is not reproducible — the word has no partial sense. Nothing here shows intent; a fallback path is an ordinary thing to write.
FIG 'The number is wrong' would be the wrong complaint. Both branches are CORRECT; each answers its own question accurately. The defect is that the result depends on ambient state that is not an argument, so the same code on the same commit returns different values on two machines and neither can tell. A number reproducible SOMETIMES is not reproducible — the word has no partial sense. Nothing here shows intent; a fallback path is an ordinary thing to write.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of HARD RESET · David Lee Wise (ROOT0), with AVAN