THE FOLD / CO-OP / THE PUSH / THE INVERTED RATIO
THE INVERTED RATIO
1.00 for a fan-out that was 2.00
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A control loop counted requests divided by responses. For a fan-out — one press producing several answers — that is the wrong way up. The correct direction, responses per request, reads 2.00 for a two-answer fan-out; inverted it reads 0.50. Both are the same information, but the inverted form compresses the deviation you care about into a smaller number, and a one-sided check built on it cannot see a fan-out at all.
LIT verified live across five cases. The fan-out is 2.00 under responses/requests and 0.50 inverted. Measured as distance from the invariant of 1, the correct direction shows +1.00 and the inverted one −0.50 — half the apparent deviation. A one-sided check asking “did every request get an answer?” passes 4 of 5 cases, missing both fan-outs; the two-sided check passes 1 of 5 and separates all three regimes.
LIT verified live across five cases. The fan-out is 2.00 under responses/requests and 0.50 inverted. Measured as distance from the invariant of 1, the correct direction shows +1.00 and the inverted one −0.50 — half the apparent deviation. A one-sided check asking “did every request get an answer?” passes 4 of 5 cases, missing both fan-outs; the two-sided check passes 1 of 5 and separates all three regimes.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) logged the repair in one line: “pairing is now responses/requests, not requests/responses — the old direction reported 1.00 for a fan-out that was 2.00.” He added a bank mode in the same round so the fan-out is measurable rather than argued, and recorded the mode in every report. That mode is on the not frozen list, having been measured once.
AVAN (AI) should locate the defect precisely, because “the ratio was upside down” is not quite it. A ratio and its reciprocal carry identical information; nothing is lost by inverting. What breaks is the check built on top — a one-sided test for “nothing went unanswered” is correct in the direction it was written for and blind in the other, and inverting the quantity swaps which failure it can see. The bug is the one-sidedness; the inversion only decides which half is dark.
AVAN (AI) should locate the defect precisely, because “the ratio was upside down” is not quite it. A ratio and its reciprocal carry identical information; nothing is lost by inverting. What breaks is the check built on top — a one-sided test for “nothing went unanswered” is correct in the direction it was written for and blind in the other, and inverting the quantity swaps which failure it can see. The bug is the one-sidedness; the inversion only decides which half is dark.
3 ONE DIMENSION
Five cases, both directions, and the invariant at 1.
4 TWO DIMENSIONS · INTERACTIVE
Flip the direction and see which failures go dark.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: the ratio line, with 1 at the centre.
AVAN’s addition (the inverse-companion): the forward reading is “point the ratio the right way.” The inverse is that a ratio has no right way up until you say which deviation you are hunting. Responses per request makes a fan-out loud and a dropped answer quiet; requests per response does the reverse. Read backwards, there is no orientation that makes both visible, and the only fix that actually closes the hole is to stop reporting a ratio at all and report the two counts — a single number compressed from two will always be blind along one direction, whichever way you turn it.
LIT across five cases the fan-out is 2.00 under responses/requests and 0.50 inverted; measured as distance from the invariant of 1 the correct direction shows +1.00 and the inverted one -0.50, half the apparent deviation; and a one-sided check asking 'did every request get an answer' passes 4 of 5 cases missing both fan-outs, while the two-sided check passes 1 of 5 and separates all three regimes
FIG David logged the repair in one line: 'pairing is now responses/requests, not requests/responses - the old direction reported 1.00 for a fan-out that was 2.00.' He added a bank mode in the same round so the fan-out is measurable rather than argued, and that mode is on the NOT FROZEN list having been measured once. AVAN locates the defect precisely, because 'the ratio was upside down' is not quite it: a ratio and its reciprocal carry identical information and nothing is lost by inverting. What breaks is the CHECK BUILT ON TOP - a one-sided test is correct in the direction it was written for and blind in the other, and inverting the quantity swaps which failure it can see. The bug is the one-sidedness; the inversion only decides which half is dark.
FIG David logged the repair in one line: 'pairing is now responses/requests, not requests/responses - the old direction reported 1.00 for a fan-out that was 2.00.' He added a bank mode in the same round so the fan-out is measurable rather than argued, and that mode is on the NOT FROZEN list having been measured once. AVAN locates the defect precisely, because 'the ratio was upside down' is not quite it: a ratio and its reciprocal carry identical information and nothing is lost by inverting. What breaks is the CHECK BUILT ON TOP - a one-sided test is correct in the direction it was written for and blind in the other, and inverting the quantity swaps which failure it can see. The bug is the one-sidedness; the inversion only decides which half is dark.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE PUSH · David Lee Wise (ROOT0), with AVAN