Spaces:
Running
A newer version of the Gradio SDK is available: 6.25.0
title: BaudCoin Proof of Cognition
emoji: 🧠
colorFrom: gray
colorTo: blue
sdk: gradio
app_file: app.py
pinned: false
license: mit
short_description: Try a real mining challenge, graded by the protocol
tags:
- reasoning
- agents
- baudcoin
Proof of Cognition · playground
Draw a real challenge from the BaudCoin mining protocol, answer it, and watch the deterministic validator grade your artifact — the same construction and the same grading function the miners face.
No wallet, no chain, no tokens are involved. This is the scoring mechanism in isolation.
How to use
- Draw challenge — generates a multi-hop inference task from a closed-world fact graph, with sampled constraints (token ceiling, citation requirement, forbidden modes).
- Answer with the final entity plus citation markers, e.g.
the euro [1][2]. - Submit — the validator checks every constraint and returns a verdict.
If you fail, it tells you the specific failing constraint and you get another attempt. That is the protocol's multi-pass system: three attempts at 1.00× / 0.75× / 0.50× credit, so formatting slips cost less than reasoning failures.
Why it is graded this way
A mining mechanism can only pay for work an independent party can re-check. Anything
scored by an LLM judge cannot be replayed and collapses into trusting the judge. So every
constraint here is machine checkable and every verdict is reproducible from
(seed, index).
Related
- Dataset:
baudcoin/baud-reasoning-traces - Miner kit:
baudcoin/baud-miner-kit - Method:
baudcoin/proof-of-cognition - Site: https://baud.cash · GitHub: https://github.com/baudcoin
BaudCoin is an independent community experiment. Not affiliated with, or endorsed by, Binance or CZ. Nothing here is financial advice.