AT-1 Discover — at1 discover
at1 discover takes a sequence of integers and searches a cross-domain library of hypotheses for the shortest program that reproduces it exactly: state recurrences (LCG / linear-recurrence / LFSR), index-functions (output[i] = f(seed+i)), elementary cellular automata, and one compositional layer that cracks a delta or xor transform of the stream when the raw stream resists. A hypothesis counts only if it regenerates the whole sequence byte-for-byte — the verify gate that separates a real discovery from a guess.
Use it
# recover the generator behind a sequence + emit a regenerating container
at1 discover series.txt -o rule.at1gen
# RECOVERED: elementary-CA (cellular-automaton) ratio=35.6x [9 B program for 320 B]
# anyone can regenerate every value from the tiny container, SHA-256 verified
at1 discover verify rule.at1gen
# VERIFIED - regenerated 80 values byte-exact
# random / CSPRNG / genuine data: it honestly finds nothing
at1 discover noise.txt
# NO GENERATOR - values look irreducible (honest miss)Input is integers separated by whitespace, commas, or newlines (a single numeric column). The output .at1gen is a few coefficients that regenerate every value on verify, checked against a SHA-256 of the original sequence — a bad generator fails the hash, so the container can’t lie.
What it proves
- Recovery — the exact rule behind the data (a 320-byte CA sequence → a 9-byte rule + seed).
- Losslessness — every value regenerates bit-for-bit; nothing is approximated.
- Honesty — random, encrypted, or genuinely irreducible data returns no generator. It never invents a law that isn’t there — the discipline that makes a positive result trustworthy.
It runs only verified, deterministic recovery — it executes none of the offline model-written code that inventsnew hypotheses, so it’s safe to point at untrusted data. This is the productised core of AT-1’s Hypothesis Engine. See the product overview, and its siblings generator recovery and the RNG auditor.