dydact orthogonaLabs applied sciences
dydact orthogonaLabs applied sciences
A universal reasoning compiler.
Zero parameters. Zero training.
One compiler beats Kona on Sudoku, ChemBERTa on chemistry property prediction, and Cadence on circuit equivalence — without any learned weights. API access for falsification, reproduction, and your own experiments.
0
trained parameters
0
training samples
100%
extreme Sudoku
30.6ms
avg solve
interactive demos
Sudoku
Paste 81-char puzzle. 100% extreme @ 30.6ms.
vs Kona 96.2% @ 313ms / 36.5M params
Try it
Math Evaluator
Any expression → exact symbolic evaluator → exact result.
Sub-ms warm. LaTeX + sympy + Python forms.
Try it
Chemistry / SMILES
SMILES → 3D conformer + property predictions.
MW r=0.998 / TPSA r=0.979 zero-shot
Try it
Circuit EC
.bench netlist → equivalence verification.
ISCAS'85 100% incl. c6288 32-bit multiplier
Try it
Boolean SAT
CNF clauses → assignment + energy trajectory.
100% at N=10 / 90% gradient at N=25
Try it
Want an API key?
Academic researchers verifying a falsification claim, labs that want programmatic access, and corporate teams running benchmark suites — reach out and tell us your intended use. We'll spin up access and rate limits that fit.
Contact us We publish our failures.
Gradient SAT drops to 90% at N=25. Zero-shot logP r=0.581 (ChemBERTa-trained wins). TSPLIB 11.8% above optimal. We list every known mode here so you can decide where it fits your work.
See the failure dashboard using dydact Lab intelligences in a publication?
Compute by Contraction describes the foundational mathematics of the dydact architecture. Academic tier requires citation. Every result returned by the API includes a BibTeX snippet + a permanent reproducibility hash you can paste into your paper.
@unpublished{ukpeh2026compute,
author = {Ukpeh, Francis A. and {orthogonaLabs}},
title = {Compute by Contraction: Determinism in Energy-Based Models},
year = {2026},
month = may,
note = {Preprint. \url{https://lab.dydact.io/papers/compute-by-contraction.pdf}}
}