Programmable mining
One argument about what Bittensor is, and two instruments that record whether the network says so.
Sealed: fixed at birth. Open: anyone can write it. TAO keeps Bitcoin's money on purpose: no pre-mine, 21 million, halvings, paid out by the block.
Pay strangers, every block, for work you define. Bitcoin defined one job, and the job was guessing a number. The hash bought security and nothing else. The mining was never the waste. The fixed puzzle was.
The design space moved up a layer, to contracts. A decade of chains lives there. Nobody reprogrammed the mine. They reprogrammed what its output could do.
The puzzle is a program. A subnet owner writes what the work is, how it is scored, and how the block reward splits. Miners compete on the work. Validators compete on the judgment. The chain does not run the puzzle. It pays the answer.
Ethereum's work is verified by re-running it. Bittensor's work can only be verified by judging it, and the network's own docs call the judgments subjective OTF 2024. A hash is right or wrong AS 2022. A judgment can be gamed, and on Bittensor the judges are a paid market of their own.
"Show me the subnet, and I'll tell you what the miners are optimizing for."
Ala Shaabana, June 2026. Below: pick a job. The same puzzle in three states: as written, gamed, and patched. Nothing touches the chain.
The numbers are illustrative: a toy scoring function, a toy miner, nothing on chain. The shape is not. Every subnet's incentive mechanism is a function like the one on the left, and every one of them has a right-hand panel somewhere.
A hash cannot be gamed. A judgment can. The subnets already write against it: Affine sells a mechanism that is "sybil-proof, decoy-proof, copy-proof, and overfitting-proof" SN120, and Ridges names "applying a stored answer" as hardcoding SN62. Reward hacking is to Bittensor what reentrancy was to early Ethereum: the native exploit of the new design space.
Research versus revenue, the pace of rule changes, enterprise readiness. Each asks the same thing of a different puzzle: is this work worth paying for before anyone has bought it? The foundation says it plainly: "a subnet nobody uses still collects a slice of every block" OTF 2026. Targon's target is a dollar of revenue for every dollar a miner earns SN4. Bitcoin subsidised security for years before fees mattered. The frame names the bet. It does not settle it.
The argument above is one person's. The lineage below is not.
Two of eighteen narrators use the phrase, against seven for market, the network's dominant frame. The foundation's own about page now makes the Ethereum comparison and lands it on markets. The argument lands it on mining, on purpose: a Bitcoin reader already holds that slot, and mining names the work where market names the venue. The lineage is theirs. The choice of word, and the line about work that can only be verified by judging it, are the argument's. Trace the phrase across every narrator in the metaphor index.
Thirty-two puzzles, rated on whether a staker, a miner, a buyer, or a newcomer can find out what they pay for in five minutes. If a subnet is a mining program, this is whether anyone can read it. The index has a method page and a correction form, and does not depend on the argument being right.
SN107 Minos runs a recipe contest for reading DNA, where miners submit settings for mutation-finding software and the judges cook every recipe themselves on synthetic genomes with planted mutations.
SN51 Lium rents out GPU time by the second, like a holiday-let site where the properties belong to strangers and the site pays hosts a retainer even when nobody is staying.
SN15 ORO runs a daily talent contest for robot personal shoppers, scores every performance, and keeps the footage as training material for a shopping assistant it plans to sell later.
SN120 Affine runs a boxing title for AI coding models, where one champion checkpoint holds the belt, each challenger gets a single shot, and the purse is split among the last five champions.
SN64 Chutes rents out AI model computing power by the request, the way a cloud provider does, except the machines belong to strangers who are paid in the subnet's token.
SN97 Albedo is a playground game of king of the hill for coding models, where one fine-tuned model stands on top and earns most of the prize until a challenger knocks it off by a clear margin, twice.
"If you ask five people what Bittensor is, you will get five different answers" BS 2024. The map records the answers, verbatim and dated, reads every voice through one framework, and traces each frame to its first use. It has its own method page and correction form.