THE FOLD / RESPAWN / ROLLBACK / THE GATE THAT CAN STOP IT
THE GATE THAT CAN STOP IT
put the falsifiable floor early
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
One floor in the tower is allowed to halt everything. If a thirteen-symbol language does not cover Fortran, the answer is not more symbols — it is that Fortran is a different shape, and that finding is the product. A gate like this is only worth having where it sits early enough to stop work that has not happened yet, and a gate placed after the work it would invalidate is not a gate at all.
LIT verified live over the tower’s own 28 blocks. Placing the stop-gate at F3 means 13 blocks spent before it and 15 at risk; at F6 it would be 25 spent and only 3 at risk — which is the wrong comparison, because the spent blocks are gone either way. If the gate fails half the time, expected total spend rises monotonically with its position: 16.0 blocks at F1 against 28.0 at F7.
LIT verified live over the tower’s own 28 blocks. Placing the stop-gate at F3 means 13 blocks spent before it and 15 at risk; at F6 it would be 25 spent and only 3 at risk — which is the wrong comparison, because the spent blocks are gone either way. If the gate fails half the time, expected total spend rises monotonically with its position: 16.0 blocks at F1 against 28.0 at F7.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) marked the floor and gave it authority in one line: “F3 is the one that can stop the tower. if 13 forms do not cover fortran, the answer is not more forms — it is that fortran is a different shape, and the finding is the product.” Elsewhere in the same file: “a form count that has to grow is a finding about fortran, not a failure of i13.”
AVAN (AI) should be careful about the optimisation, because “put the falsifiable gate first” is not quite available. Expected spend is minimised at F1, but F3 cannot move there — you cannot measure coverage before you have read the corpus and written the veto, so its inputs pin it. What the arithmetic actually shows is that F3 is as early as its dependencies allow, which is a weaker and truer claim than choosing the optimum freely.
AVAN (AI) should be careful about the optimisation, because “put the falsifiable gate first” is not quite available. Expected spend is minimised at F1, but F3 cannot move there — you cannot measure coverage before you have read the corpus and written the veto, so its inputs pin it. What the arithmetic actually shows is that F3 is as early as its dependencies allow, which is a weaker and truer claim than choosing the optimum freely.
3 ONE DIMENSION
Where the gate sits, and what it puts at risk.
4 TWO DIMENSIONS · INTERACTIVE
Slide the stop-gate up the tower and watch the expected spend.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: the tower, with the stop-gate as a cut plane.
AVAN’s addition (the inverse-companion): the forward reading is “place the falsifiable gate as early as its inputs allow.” The inverse is that a gate can only be early if it is cheap to state, and the cheapest things to state are rarely the ones worth testing. F3 is answerable early precisely because it asks a small question — does this symbol set cover that corpus — and the questions that would falsify the whole enterprise, whether any of this produces a useful agent, cannot be asked until nearly everything is built. Read backwards, the early gates are the ones you were least likely to be wrong about, and the discipline buys speed of refutation at the cost of refuting only the small claims.
LIT over the tower's own 28 blocks, placing the stop-gate at F3 means 13 blocks spent before it and 15 at risk, while at F6 it would be 25 spent and only 3 at risk - the wrong comparison, since the spent blocks are gone either way; and if the gate fails half the time, expected total spend rises monotonically with its position from 16.0 blocks at F1 to 28.0 at F7
FIG David marked the floor and gave it authority: 'F3 is the one that can stop the tower. if 13 forms do not cover fortran, the answer is not more forms - it is that fortran is a different shape, and the finding is the product.' Elsewhere: 'a form count that has to grow is a finding about fortran, not a failure of i13.' AVAN is careful about the optimisation, because 'put the falsifiable gate first' is not quite available: expected spend is minimised at F1, but F3 cannot move there since you cannot measure coverage before reading the corpus and writing the veto. What the arithmetic shows is that F3 is AS EARLY AS ITS DEPENDENCIES ALLOW, which is weaker and truer than choosing the optimum freely.
FIG David marked the floor and gave it authority: 'F3 is the one that can stop the tower. if 13 forms do not cover fortran, the answer is not more forms - it is that fortran is a different shape, and the finding is the product.' Elsewhere: 'a form count that has to grow is a finding about fortran, not a failure of i13.' AVAN is careful about the optimisation, because 'put the falsifiable gate first' is not quite available: expected spend is minimised at F1, but F3 cannot move there since you cannot measure coverage before reading the corpus and writing the veto. What the arithmetic shows is that F3 is AS EARLY AS ITS DEPENDENCIES ALLOW, which is weaker and truer than choosing the optimum freely.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of ROLLBACK · David Lee Wise (ROOT0), with AVAN