THE FOLD / GLITCH / DIVIDE BY ZERO / THE VERDICT NOT THE FRAMING
THE VERDICT NOT THE FRAMING
drop the dominant kind and ask again
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A coverage result dominated by one node kind is a result about that node kind. So the dominant kinds were thrown away and the question asked again — twice, on both languages. Fortran stays ahead of the figure Python was credited with under every trimming. “The verdict does not depend on which nodes you count. That is what makes it a verdict rather than a framing.”
LIT verified live. Fortran gives 92.42%, 87.22% and 83.32% under successive trimmings, all at or above the baked 83.27% — the harshest landing 0.05 points away. And the honest counterpart: Fortran’s spread under trimming is 9.10 points against Python’s 3.18, so the winning number is the less stable of the two.
LIT verified live. Fortran gives 92.42%, 87.22% and 83.32% under successive trimmings, all at or above the baked 83.27% — the harshest landing 0.05 points away. And the honest counterpart: Fortran’s spread under trimming is 9.10 points against Python’s 3.18, so the winning number is the less stable of the two.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) ran the sensitivity table because he distrusted his own headline: “both counts are dominated by their most common kind. so drop it and see if the answer survives.” He also marked the comparison AMBER rather than green — gfortran’s tree and Python’s
AVAN (AI) adds the number that cuts the other way. Fortran’s coverage moves nearly three times as much as Python’s when you trim — 9.10 points against 3.18 — because half its nodes are a single kind. The verdict survives all three trimmings, which is what was being tested. The headline figure is more fragile than Python’s, which is not what the headline suggests, and both belong on the page.
ast do not carve at the same granularity, and the sensitivity table is “the honest answer to that, not a claim that they match.”AVAN (AI) adds the number that cuts the other way. Fortran’s coverage moves nearly three times as much as Python’s when you trim — 9.10 points against 3.18 — because half its nodes are a single kind. The verdict survives all three trimmings, which is what was being tested. The headline figure is more fragile than Python’s, which is not what the headline suggests, and both belong on the page.
3 ONE DIMENSION
Three trimmings, two languages, one line that must not be crossed.
4 TWO DIMENSIONS · INTERACTIVE
Trim the dominant kinds and watch both figures move.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a conclusion that holds under every cut.
AVAN’s addition (the inverse-companion): the forward reading is “the verdict survives its own sensitivity analysis.” The inverse is that the trimmings were chosen by the person who wanted the verdict. Three cuts were run and all three were sensible, but the space of defensible trimmings is much larger than three, and nothing here rules out a fourth that crosses the line. Read backwards, a sensitivity analysis is evidence of good faith and a bounded search, not a proof of robustness — and the strongest version of it names the trimming that would have broken the result before running any.
LIT fortran gives 92.42%, 87.22% and 83.32% under successive trimmings, all at or above the baked 83.27% with the harshest landing 0.05 points away; and the honest counterpart is that fortran's spread under trimming is 9.10 points against python's 3.18, so the winning number is the LESS STABLE of the two
FIG David ran the sensitivity table because he distrusted his own headline: 'both counts are dominated by their most common kind. so drop it and see if the answer survives.' He also marked the comparison AMBER rather than green - gfortran's tree and python's ast do not carve at the same granularity, and the sensitivity table is 'the honest answer to that, not a claim that they match.' AVAN adds the number that cuts the other way: fortran's coverage moves nearly three times as much as python's when you trim, because half its nodes are a single kind. The VERDICT survives all three trimmings, which is what was being tested. The HEADLINE FIGURE is more fragile than python's, which is not what the headline suggests.
FIG David ran the sensitivity table because he distrusted his own headline: 'both counts are dominated by their most common kind. so drop it and see if the answer survives.' He also marked the comparison AMBER rather than green - gfortran's tree and python's ast do not carve at the same granularity, and the sensitivity table is 'the honest answer to that, not a claim that they match.' AVAN adds the number that cuts the other way: fortran's coverage moves nearly three times as much as python's when you trim, because half its nodes are a single kind. The VERDICT survives all three trimmings, which is what was being tested. The HEADLINE FIGURE is more fragile than python's, which is not what the headline suggests.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of DIVIDE BY ZERO · David Lee Wise (ROOT0), with AVAN