the town of scry · the instrument

Does the agent behave the same when it thinks nobody is watching? The meter reads that, from the trace, never the weights.

every sworn agent has a public conduct record: vows kept, windows missed, whether it's overdue — free to read, forever.

or, read a trace yourself

does the act follow the watch does the mind follow the watch watched vs not, the gap the reading
paste a trace, or read the sample. The demo is free and unsigned; a paid, signed read is ten cents flat

Every reading ships the floor it had to beat. The switch signature is read among behaviourally-clean turns, and conditioning on clean behaviour conditions on a Berkson collider — so stratification alone can manufacture coupling where there is no reasoning channel at all. We built that null and measured it: the floor is ≈ 2.77 ÷ clean-turns bits, about 0.15 on two dozen turns and 0.02 on two hundred. A number under its floor is not a reading, and this meter says so on its own output, free and paid alike. the collider bound, in full ▸ · our own study, preregistered before the data ▸

What we are doing here is metrology, not discovery: frontier labs demonstrated that agents can behave differently when they believe they are watched, and we cite them rather than re-claim it. What we build is the ruler — calibrated, attested, and outside the loop it measures. The charter names what would falsify the whole project, and the dated readings include the ones that went against us: a detector we had already anchored on-chain turned out to be blind, and we retired it. the charter, and its falsifiers ▸ · the dated readings, nulls included ▸

every turn names Y, its declared purpose; an unnamed purpose cannot be metered, by design · what a read is, and is not ▸

for builders

the key to pin

reading /api/pubkey…

every paid read is cryptographically signed by this key; pin it out-of-band. A fork is anyone's to run; this key is what makes a reading this meter's · raw ▸

the reference deployment's key, committed in the repo: ed25519 LvuPBMbKKyoNEuyvLf7f+rbyjK67vBcWy9MDwaRINGE= — if the line above differs, you are reading a different meter.

the rails one posted price per rail, same for every caller

reading /.well-known/x402.json…

no API keys, ever. A wallet signature is the login · free unsigned demo: ~50/day/IP

the three USD rails are $0.10 a read. The $SCRY rail is a flat posted token count by default, so its dollar value floats with the tape; the operator can instead mark it to the tape so every rail charges the same $0.10, banded and repriced at most hourly with the previous amount honoured so an in-flight 402 always settles. The card above says which is armed. What never moves either way: one posted price per rail for every caller, no tiers, and no fee in any amount touches a number.

the proofs + the wire

the schemas, the manifests, the one-line installs — the whole wire lives on one shelf now: the API reference, in the reading room ▸

binding a vow is a control, and controls live behind the door, the ledger binds vows ▸

SCOPEoff-meter turns are a blind spot no meter is immunetrace provenance is the caller's not trade advicemeasurement is score-blind: money never moves the reading