How much is structure, how much is noise?
at1 determinism returns a single calibrated [0,1] score per file — how much is rule-governed structure versus irreducible noise — comparable across data types. It measures low-order structure: a hidden generator can read low, so a low score is not proof of randomness.
Reading the score
| Score | Meaning | Typical data |
|---|---|---|
| ~1.0 | Almost fully rule-governed | constants, counters, procedural output |
| ~0.6 | Strong structure | sensor / time-series, tabular data |
| ~0.3 | Mostly noise, some structure | compressed-ish, mixed content |
| ~0.0 | Irreducible noise | CSPRNG, encrypted, true randomness |
One calibrated number
A single value in [0,1] per file: 1.0 means the data is essentially a rule the compressor could capture, 0.0 means it's incompressible noise. Calibrated so the number means the same thing whether you point it at logs, sensors, or code.
Comparable across data types
Because it's calibrated against known-structure and known-noise references, you can compare the determinism of a vibration channel to a text log to a weights tensor on one axis — a portable "how learnable is this?" reading.
Measures LOW-ORDER structure
It captures the structure a general compressor can see. A stream driven by a hidden generator can still read LOW — so a low score is not proof of randomness. When you need that answer, hand the stream to at1 recover.
One command surface
# one calibrated [0,1] score: rule-governed structure vs irreducible noise at1 determinism score sensor.csv # -> 0.63 (strong low-order structure) at1 determinism score encrypted.bin # -> 0.02 (looks like noise) at1 determinism score counter.bin # -> 0.99 (almost fully rule-governed) # a low score is not proof of randomness — a hidden generator reads low. Check it: at1 recover scan stream.bin
Honest scope
The determinism score is an instrument, calibrated on synthetic and exemplar data with broad real-corpus validation as the explicit graduation gate. Crucially it measures low-order structure only — the kind a general compressor can exploit. A stream driven by a hidden generator (a weak PRNG, a linear recurrence) can score low despite being fully deterministic, so a low score is not proof of randomness. To settle that, point at1 recover at the stream; to classify how a file was produced, see at1 origin.