ZLM · ask in English, answered by your own machine

The model writes the explanation; your machine computes the number. Those are two processes on purpose — if one thing both produced the value and wrote the account of it, their agreeing would prove nothing. Every number here came off your machine, and the model could not have written a different one: digits are not in its vocabulary to write. A model would have to be trusted to do the arithmetic, and then agreement between the answer and the receipt would prove nothing — both would have come from the same place. The prose here is templated; every number in it came off your machine.

Ranked by name overlap plus domain hints — the same heuristic measured at top‑1 5/8, top‑5 8/8 on eight held-out questions. Operators whose inputs are confirmed against their domain's published keys rank first. An unconfirmed mark means the published key list does not cover this equation's symbols — that list is a lower bound, so it may still compute. It is not a claim that the operator is unusable. This tab reads the catalogue, not your machine, so it costs nothing.

Every number in the text is matched against the value within its stated tolerance. Untraceable numbers are flagged; so are numbers that trace but wear the wrong unit. Signal and confidence are reported separately — blending them makes most of an innocuous document's score a measure of how well it was researched.

Your machine does the mathematics and returns a keyed HMAC proof. ZLM holds no credential of its own — your session travels with the request, and the compute is metered to your envelope, not to ours. Change the value by one digit and the proof fails: it binds to the number, which is why agreement means something.