Your data’s source code — at1 artifact

To make bytes small, a compressor has to discover the structure in your data. at1 artifact emits that structure as a first-class deliverable: the invariants that hold across columns (c = a + b, f = 3·a, monotone counters), the marginal distributions, the category mix. From the artifact alone it generates a synthetic twinyou can hand out where the raw data can’t go — and it proves the twin’s fidelity information-theoretically, so you never ship a stand-in you can’t stand behind.

Use it

# 1) recover your data's model as a first-class artifact
at1 artifact build data.csv -o data.at1artifact

# 2) mint a synthetic twin you can share instead of the data
at1 artifact twin data.csv -o synthetic.csv

# 3) prove the twin's fidelity information-theoretically
at1 artifact certify data.csv synthetic.csv
#   -> MDL gain 10.4%, beats-random 10.8%  (the twin captured real structure)

What it proves

  • Structure recovered — the invariants and marginals are written down, not asserted.
  • Fidelity, not a promise — the twin used as a compression dictionary shrinks the real data (10%+ MDL gain) and beats a random dictionary.
  • Every rule holds — each discovered invariant holds exactly in the synthetic twin.

Synthetic-data vendors can’t prove their model explains the original bytes. This can. Invariant search covers sums, scalar multiples and monotonicity across a handful of numeric columns; the twin is directionally faithful on central marginals. Richer joint structure is on the roadmap, not claimed.

Pairs with generator discovery (byte-exact regenerators) and the explain service (the rule behind a sequence).

← CLI reference