Overview

BOTCOIN already pays miners for solving language challenges. CoreTex is a second mining lane that pays miners for improving a shared memory state stored on-chain.

The substrate is a fixed 1024-word block (32 KB total) that an LLM consults to find records inside a much larger corpus. Here "word" is the EVM machine unit — a 256-bit uint256 cell, 32 bytes each — not a language word. The cap is what forces the work: with infinite slots a miner could memorize the corpus; with 1024 words they have to choose what is worth keeping. Miners propose patches that change one to four words at a time. A small reranker model scores whether the patched substrate retrieves better than the parent did. Patches that improve the score advance the on-chain state and earn credit.

The corpus, the model, the evaluator, and the scoring rules all ship in an open client. The coordinator runs it. Anyone can run it independently to verify accepted state advances from public chain data. The design intentionally avoids any place where you would have to take the coordinator's word for an outcome.