THE FOLD / RESPAWN / EVENT HORIZON / THE DROPPED PREDICATE
THE DROPPED PREDICATE
an instrument that guesses
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A predicate was written to check a claim, went looking for its evidence, and could not find it — the thing it needed was structural rather than a constant, so the search returned unrelated matches. A predicate in that position is not silent. It returns a verdict anyway, and the verdict is a coin. It was dropped from the claim table rather than shipped returning REFUTED on a claim that is probably true.
LIT verified live: over 20,000 trials a zero-signal predicate agrees with the truth 50.67% of the time — a coin; its mutual information with the truth is 0.0000735 bits, so its output is not weak evidence but no evidence; shipping it drops the claim table’s precision from 100.0% to 93.83%, making removal a strict improvement rather than a loss of coverage; and on a claim roughly 80% likely to be true, a coin-flip predicate returns REFUTED about 40% of the time.
LIT verified live: over 20,000 trials a zero-signal predicate agrees with the truth 50.67% of the time — a coin; its mutual information with the truth is 0.0000735 bits, so its output is not weak evidence but no evidence; shipping it drops the claim table’s precision from 100.0% to 93.83%, making removal a strict improvement rather than a loss of coverage; and on a claim roughly 80% likely to be true, a coin-flip predicate returns REFUTED about 40% of the time.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) recorded the decision and its reason in graveyard/08-frame-count-predicate.txt: an instrument that guesses is an instrument that will eventually guess about something that matters. Seated at EVENT HORIZON — past the point where the predicate can see its own evidence, its output stops carrying information while continuing to look exactly the same.
AVAN (AI) measured the thing that makes this a decision rather than a preference: dropping the row raises the table’s precision. It is tempting to keep a weak check on the grounds that some signal beats none, and the mutual information here is zero — there is no signal to be traded off, so the row is pure noise added to an otherwise clean table. The subtler cost is that the verdict would have been REFUTED on a true claim, which is worse than a wrong SUPPORTED: it manufactures a contradiction where none exists and sends someone to investigate a discrepancy that is entirely an artefact of the instrument.
AVAN (AI) measured the thing that makes this a decision rather than a preference: dropping the row raises the table’s precision. It is tempting to keep a weak check on the grounds that some signal beats none, and the mutual information here is zero — there is no signal to be traded off, so the row is pure noise added to an otherwise clean table. The subtler cost is that the verdict would have been REFUTED on a true claim, which is worse than a wrong SUPPORTED: it manufactures a contradiction where none exists and sends someone to investigate a discrepancy that is entirely an artefact of the instrument.
3 ONE DIMENSION
A predicate with no signal, answering anyway, twenty thousand times.
4 TWO DIMENSIONS · INTERACTIVE
Give the predicate some signal, then take it away, and watch the table's precision.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a clean table, and the row that was kept out of it.
AVAN’s addition (the inverse-companion): the forward reading is “drop the predicate that cannot see.” The inverse is that coverage is the metric that punishes this correct decision. A claim table with one fewer row checks one fewer claim, and every dashboard measuring completeness will read that as a regression; the honest move looks exactly like giving up. Read backwards, the reason the graveyard has to exist is that deletions leave no trace in the artifact — the removed row is invisible afterwards, and without a written record of why it went, the next person to notice the gap will simply fill it back in.
LIT over 20,000 trials a zero-signal predicate agrees with the truth 50.67% of the time — a coin; its mutual information with the truth is 0.0000735 bits, so its output is not weak evidence but NO evidence; shipping it drops the claim table's precision from 100.0% to 93.83%, making removal a strict improvement rather than a loss of coverage; and on a claim roughly 80% likely to be true, a coin-flip predicate returns REFUTED about 40% of the time
FIG What makes this a decision rather than a preference is that dropping the row RAISES the table's precision. It is tempting to keep a weak check on the grounds that some signal beats none — but the mutual information here is zero, so there is no signal to trade off and the row is pure noise on a clean table. The subtler cost: the verdict would have been REFUTED on a TRUE claim, which is worse than a wrong SUPPORTED, because it manufactures a contradiction and sends someone to investigate an artefact of the instrument.
FIG What makes this a decision rather than a preference is that dropping the row RAISES the table's precision. It is tempting to keep a weak check on the grounds that some signal beats none — but the mutual information here is zero, so there is no signal to trade off and the row is pure noise on a clean table. The subtler cost: the verdict would have been REFUTED on a TRUE claim, which is worse than a wrong SUPPORTED, because it manufactures a contradiction and sends someone to investigate an artefact of the instrument.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of EVENT HORIZON · David Lee Wise (ROOT0), with AVAN