They are comparison baselines, not general ARC solvers. The current CogniARC pipeline is rule-first and may use a micro-NN only as a hint or comparison. Low-confidence output must abstain or escalate to a separately verified path.

Bundled models

Model	Role	Authority
`domain_classifier.json`	Compare movement/rotation/transform/hybrid hypotheses	Advisory
`action_predictor.json`	Estimate whether a candidate action may succeed	Advisory
`captcha_classifier.json`	Experimental CAPTCHA-family classification fixture	Research only; never a bypass mechanism

The historical accuracy values in older cards were measured on small or synthetic splits. They are not claims of ARC-AGI-3 generalization. Consult the source repository's dev/holdout reports and reproduce the relevant benchmark before drawing conclusions.

Where it is used

Minimal JSON inference

python

import json
import numpy as np

with open("domain_classifier.json", encoding="utf-8") as stream:
    model = json.load(stream)

x = np.asarray([1.0, 1.0, 0.3, 0.35, 0.45, 0.02], dtype=float)
for index, (weights, bias, activation) in enumerate(zip(
    model["weights"], model["biases"], model["activations"]
)):
    matrix = np.asarray(weights).reshape(model["layers"][index + 1], model["layers"][index])
    x = x @ matrix.T + np.asarray(bias)
    if activation == "relu":
        x = np.maximum(0, x)
    elif activation == "sigmoid":
        x = 1 / (1 + np.exp(-x))
    elif activation == "softmax":
        shifted = x - np.max(x)
        x = np.exp(shifted) / np.exp(shifted).sum()

print(x)

Use the source feature extractors rather than inventing anonymous feature vectors. The example demonstrates the stored feed-forward format only.

Licence

Released under the MIT License.

zedgamer/cogniarc-nano-nn

CogniARC Nano-NN — experimental ARC classifiers

Bundled models

Where it is used

Minimal JSON inference

Licence

zedgamer/
cogniarc-nano-nn