# ModelProof: Bit-Exact, Permanently Stored AI Models You Can Verify in a Browser Tab

DevFeed: [ModelProof: Bit-Exact, Permanently Stored AI Models You Can Verify in a Browser Tab](<https://devfeed.tech/articles/modelproof-bit-exact-permanently-stored-ai-models-you-can-verify-in-a-browser-tab-102018.md>)

Original publisher: [Read original article](<https://hackernoon.com/modelproof-bit-exact-permanently-stored-ai-models-you-can-verify-in-a-browser-tab?source=rss>)

Author: ModelProof

Published: 2026-10-26T23:20:25Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [ML model versioning](<https://devfeed.tech/topics/ml-model-versioning.md>), [AI Models](<https://devfeed.tech/topics/ai-models.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [Python](<https://devfeed.tech/topics/python.md>), [Web app](<https://devfeed.tech/topics/webapp.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-inference](<https://devfeed.tech/tags/ai-inference.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [ans](<https://devfeed.tech/tags/ans.md>), [arithmetic](<https://devfeed.tech/tags/arithmetic.md>), [arweave](<https://devfeed.tech/tags/arweave.md>), [behaviour](<https://devfeed.tech/tags/behaviour.md>), [browser](<https://devfeed.tech/tags/browser.md>), [cli](<https://devfeed.tech/tags/cli.md>), [code](<https://devfeed.tech/tags/code.md>), [content](<https://devfeed.tech/tags/content.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [csv](<https://devfeed.tech/tags/csv.md>), [decentralize-ai](<https://devfeed.tech/tags/decentralize-ai.md>), [decentralize-ai-hackathon](<https://devfeed.tech/tags/decentralize-ai-hackathon.md>), [decentralized-storage](<https://devfeed.tech/tags/decentralized-storage.md>), [format](<https://devfeed.tech/tags/format.md>), [manifest](<https://devfeed.tech/tags/manifest.md>), [model](<https://devfeed.tech/tags/model.md>), [modelproof](<https://devfeed.tech/tags/modelproof.md>), [models](<https://devfeed.tech/tags/models.md>), [permanent-storage](<https://devfeed.tech/tags/permanent-storage.md>), [publisher](<https://devfeed.tech/tags/publisher.md>), [stored](<https://devfeed.tech/tags/stored.md>), [verifiable-ai](<https://devfeed.tech/tags/verifiable-ai.md>), [verify](<https://devfeed.tech/tags/verify.md>)

## AI overview

ModelProof aims to make a model's published behavior reproducible: its MPW1 format uses int8 weights, int32 biases, integer accumulation, ReLU activations and shift-based requantization so inference outputs are exact across machines. A manifest records the weights' SHA-256 and size, preprocessing rules, metrics and exact test vectors; the weights and manifest are stored on Arweave. Verification fetches the files, checks the weight hash, reruns the vectors and compares output digests, with Python and JavaScript runtimes required to agree. The live browser verifier does this locally without the publisher's server. The demo is a small 4-8-3 Iris network with 100% accuracy on 30 holdout samples before and after int8 quantization; Python verification uses only the standard library, and training is reported to reproduce the weights byte for byte. The approach verifies behavior against the manifest, not the honesty of training data or claimed metrics. Larger models need chunked weights and a Merkle root, which is on the roadmap.

## Source excerpt

An integer-only model format with a content-hashed manifest on Arweave lets anyone prove a model still behaves as its publisher claimed.