Research

Verifiers, not vibes.

Snayu must prove its reasoning, not suggest it. We extract deterministic grammars from expert knowledge and check plans against them — so every replan comes with an answer that holds.

Grammar generation first

Extract a deterministic rulebook from expert text — not a classifier, a grammar.

Verification over scoring

Which rule broke, which assumption failed, which step became invalid. Structured, not a score.

Path maps as one use case

Feasibility becomes testable. The verifier generalizes far beyond planning.

Method

How verification runs.

Four steps. Deterministic by construction — same corpus, same violations.
  1. 01

    Ingest a specialized corpus: physics, planning traces, domain procedures.

  2. 02

    Extract entities, prerequisites, relations, and rules into a grammar file.

  3. 03

    Apply the grammar deterministically. Record the exact violation path.

  4. 04

    Return errors, warnings, assumptions, and missing bridges.

Diagnostics

Four kinds of finding.

Every check returns structured evidence — never a bare pass/fail.
Error
A rule is directly violated.
Warning
Plausible, but underspecified.
Assumption
Used without justification.
Omission
A necessary step is missing.

Domains

Where it applies today.

  • Physics validation

    Conservation laws apply only when the prompt supplies the activating clue.

  • Behavioral path maps

    Plans tested against habits, commitments, and real transitions.

  • Human and model text

    Any target content — the method stays general-purpose.

Research that ships in the product.