THE FOLD / BOSS / THE WALL / THE TWO ERRORS THAT CANCEL
THE TWO ERRORS THAT CANCEL
clean for exactly the wrong reason
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Two FORTRAN programs contain the identical statement
LIT verified live. Case 11 reports 0 unclosed, 0 bad. Case 12 reports 2 unclosed, 1 bad. The statement is byte-identical in both. Case 11 passes because a frame the language never opened is closed by a terminator that terminates nothing — two errors cancelling exactly, and a clean verdict over a model that has understood none of it.
DO 10 I = 1.10. A period, not a comma — and since spaces are not significant in fixed form, this is an assignment to a variable named DO10I. No loop is opened. One program is followed by 10 CONTINUE; the other is not. The checker calls the first clean and the second broken.LIT verified live. Case 11 reports 0 unclosed, 0 bad. Case 12 reports 2 unclosed, 1 bad. The statement is byte-identical in both. Case 11 passes because a frame the language never opened is closed by a terminator that terminates nothing — two errors cancelling exactly, and a clean verdict over a model that has understood none of it.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) built the pair and compiled both, which is what makes them evidence: “this is an ASSIGNMENT to a variable named DO10I. spaces are not significant in fixed form. the veto sees DO + label and opens a frame that the language never opened. the famous one. it is not folklore: gfortran compiles it.” He separated the myth from the mechanism — the story usually attached to this syntax is disputed, and the syntax is not.
AVAN (AI) noticed the pair rather than the case. David’s table lists case 11 as clean and case 12 as mis-parsed without remarking on it; the two files differ by one line. That makes 11 the more alarming of the two, because a checker that is wrong and says so is far less dangerous than one that is wrong and reports success. This page publishes the cancellation, and the credit for the pair existing at all belongs to whoever wrote both files.
AVAN (AI) noticed the pair rather than the case. David’s table lists case 11 as clean and case 12 as mis-parsed without remarking on it; the two files differ by one line. That makes 11 the more alarming of the two, because a checker that is wrong and says so is far less dangerous than one that is wrong and reports success. This page publishes the cancellation, and the credit for the pair existing at all belongs to whoever wrote both files.
3 ONE DIMENSION
The same statement, two verdicts.
4 TWO DIMENSIONS · INTERACTIVE
Add or remove the terminator and watch the verdict flip.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a phantom open meeting a phantom close.
AVAN’s addition (the inverse-companion): the forward reading is “a passing test can be evidence of nothing.” The inverse is that cancellation is the normal case, not the exception. Any checker that reports a single aggregate — a balance, a total, a difference from zero — is a function that maps many states onto one number, and every such function has a kernel: a whole space of wrong models that produce the right output. Read backwards, case 11 is not a freak. It is one visible member of the set of errors the measurement was designed to be unable to distinguish from correctness.
LIT case 11 reports 0 unclosed and 0 bad while case 12 reports 2 unclosed and 1 bad, from a byte-identical statement; case 11 passes because a frame the language never opened is closed by a terminator that terminates nothing - two errors cancelling exactly, and a clean verdict over a model that has understood none of it
FIG David built the pair and compiled both, which is what makes them evidence: 'this is an ASSIGNMENT to a variable named DO10I. spaces are not significant in fixed form. the veto sees DO + label and opens a frame that the language never opened. the famous one. it is not folklore: gfortran compiles it.' He separated the myth from the mechanism - the story usually attached to this syntax is disputed, and the syntax is not. AVAN noticed the PAIR rather than the case: David's table lists 11 as clean and 12 as mis-parsed without remarking on it, and the two files differ by one line. That makes 11 the more alarming, because a checker that is wrong and SAYS SO is far less dangerous than one that is wrong and reports success.
FIG David built the pair and compiled both, which is what makes them evidence: 'this is an ASSIGNMENT to a variable named DO10I. spaces are not significant in fixed form. the veto sees DO + label and opens a frame that the language never opened. the famous one. it is not folklore: gfortran compiles it.' He separated the myth from the mechanism - the story usually attached to this syntax is disputed, and the syntax is not. AVAN noticed the PAIR rather than the case: David's table lists 11 as clean and 12 as mis-parsed without remarking on it, and the two files differ by one line. That makes 11 the more alarming, because a checker that is wrong and SAYS SO is far less dangerous than one that is wrong and reports success.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE WALL · David Lee Wise (ROOT0), with AVAN