THE FOLD / GLITCH / SEGFAULT / THE BREACH RULE
THE BREACH RULE
an asymmetry that costs nothing
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
History only grows. That one fact turns a staleness gate into a provenance gate for free. A published number below today’s count is staleness — forgivable, fixable by rebuilding, and you need a threshold to decide how much is too much. A published number above today’s count cannot be staleness at all. No amount of age produces it. It did not come from this checkout, and detecting that needs no threshold, no configuration and no judgement.
LIT verified live: over 200,000 monotone histories a published number drawn from the repository’s own past never once exceeds today’s count; flagging “above current” therefore has 100% precision by construction; its honest limit is recall — of 200,000 genuinely foreign numbers only 58.3% land above the current count, and a foreign number below it is invisible to the rule; while the drift side carries a real tradeoff, moving from 32.7% false positives at a 0.5% threshold to 16.3% false negatives at 10%.
LIT verified live: over 200,000 monotone histories a published number drawn from the repository’s own past never once exceeds today’s count; flagging “above current” therefore has 100% precision by construction; its honest limit is recall — of 200,000 genuinely foreign numbers only 58.3% land above the current count, and a foreign number below it is invisible to the rule; while the drift side carries a real tradeoff, moving from 32.7% false positives at a 0.5% threshold to 16.3% false negatives at 10%.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) calls it the one idea in the probe, in WORKFLOW.ascii Track B, and the word he uses for the asymmetry is free. Seated at SEGFAULT — a value that came from outside the address space you were reading.
AVAN (AI) measured the limit as well as the strength, because the rule is easy to oversell. Precision is perfect and recall is not: a number from another branch that happens to be smaller than today’s count sails through, and here that is 41.7% of foreign values. So the breach rule is a one-sided instrument — when it fires you know something for certain, and when it stays quiet you know nothing at all. That combination is rarer than it sounds and it is the right trade for a gate, because a gate’s false positives cost trust while its false negatives cost only the status quo.
AVAN (AI) measured the limit as well as the strength, because the rule is easy to oversell. Precision is perfect and recall is not: a number from another branch that happens to be smaller than today’s count sails through, and here that is 41.7% of foreign values. So the breach rule is a one-sided instrument — when it fires you know something for certain, and when it stays quiet you know nothing at all. That combination is rarer than it sounds and it is the right trade for a gate, because a gate’s false positives cost trust while its false negatives cost only the status quo.
3 ONE DIMENSION
One axis, one line, and two completely different kinds of wrong.
4 TWO DIMENSIONS · INTERACTIVE
Move the published number across today's count and watch the verdict change kind.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: the cone of everything this checkout could ever have said.
AVAN’s addition (the inverse-companion): the forward reading is “above the count means it came from elsewhere.” The inverse is that the rule is bought entirely with monotonicity, and monotonicity is a property of the metric rather than of the gate. Commits and lines only grow; test pass rates do not, coverage does not, latency does not, and for those the free half of this asymmetry simply does not exist. Read backwards, the lesson is to look for the monotone quantities in your system, because each one hands you a threshold-free check that nobody has to tune — and there are usually more of them than anyone has noticed.
LIT over 200,000 monotone histories a published number drawn from the repository's own past never once exceeds today's count; flagging 'above current' therefore has 100% precision by construction; its honest limit is recall, since of 200,000 genuinely foreign numbers only 58.3% land above the current count and one below it is invisible; while the drift side carries a real tradeoff, from 32.7% false positives at a 0.5% threshold to 16.3% false negatives at 10%
FIG The rule is one-sided and easy to oversell. Precision is perfect and RECALL IS NOT — a foreign number smaller than today's count sails through, which here is 41.7% of them. When it fires you know something for certain; when it stays quiet you know nothing at all. That is the right trade for a gate, because a gate's false positives cost trust while its false negatives cost only the status quo. The whole thing is bought with monotonicity, which is a property of the metric, not the gate — pass rates and coverage and latency do not have it.
FIG The rule is one-sided and easy to oversell. Precision is perfect and RECALL IS NOT — a foreign number smaller than today's count sails through, which here is 41.7% of them. When it fires you know something for certain; when it stays quiet you know nothing at all. That is the right trade for a gate, because a gate's false positives cost trust while its false negatives cost only the status quo. The whole thing is bought with monotonicity, which is a property of the metric, not the gate — pass rates and coverage and latency do not have it.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of SEGFAULT · David Lee Wise (ROOT0), with AVAN