Minos

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.

The commodity, explained from zero

When a lab sequences a person's DNA, the machine produces a long, noisy readout. Software then has to decide which differences from a reference genome are real mutations and which are noise. That step is called variant calling: "the process of identifying DNA mutations from sequencing data." Getting it right matters because a missed mutation can be a missed diagnosis.

Minos does not sell that software. Its output today is a set of tuned recipes: the parameter settings that make four existing variant callers, named in the roadmap, perform best on a standard test. The about page frames the value as a fix for a trust problem: "Today, labs self-report accuracy with no independent verification. Pipelines go unchanged for years."

Nobody buys the recipes yet. The whitepaper page describes a future where "hospitals, biobanks, and pharmaceutical companies can submit sequencing samples and receive high-accuracy variant calls." The roadmap lists that production service as phase five, marked VISION, while phase one, tuning parameters, is the only active phase. Here the comparison breaks: in a real recipe contest someone eventually eats the dish, and at Minos no customer has yet been served.

Why it is on Bittensor at all

The about page's claim is independence. A lab grading its own pipeline has every reason to be generous; a network of strangers who are paid only when their recipe beats everyone else's on a fresh, unpredictable test has none. The genomes are synthetic, generated every 72 minutes with hidden mutations, so a miner cannot memorize the answer key. What decentralization changes here is who does the grading and how often, not cost or coverage; the materials give no cost figure.

How the work gets done

Miners submit a configuration file for a variant caller; the README is explicit that "miners will never be asked to upload outputs." Validators download each configuration, run it inside an isolated container on the round's synthetic genome, and score the result against the known planted mutations using hap.py, a standard genomics comparison tool. The score is mostly accuracy: under the current v2 formula, 70 percent for correctly finding variants and 30 percent for avoiding false positives, with a gate that zeroes any implausible result. The about page says the top miner "wins 100% of the emissions," emissions being the subnet's share of newly minted TAO, while the live network configuration sets the winner's share at 0.8 with the rest split among the next twenty in decaying slices; the sources disagree.

How you would know it works

The scoring configuration validators use is public at https://api.theminos.ai/scoring/network-config, which showed v2 scoring and a five-of-twenty-rounds eligibility rule. The dashboard at https://theminos.ai/dashboard showed a live round and 63 of 510 miners active, but its accuracy figures load only with JavaScript.

What is missing

There is no price list and no customer. The on-chain contact address, contact@theminos.ai, appears nowhere on the site, the docs, or the README. The docs quickstart still lists freebayes, which the README deprecated on 2026-05-09. The chain description, "The Foundational Layer of Genomics," is a tagline, not a sentence.

Go deeper

• Front door with plain explanation: https://theminos.ai/about
• Scoring formulas and overcall guardrail: https://raw.githubusercontent.com/minos-protocol/minos_subnet/main/docs/scoring.md
• Miner tuning guide with weights: https://raw.githubusercontent.com/minos-protocol/minos_subnet/main/docs/tuning_guide.md

Metaphor: a recipe contest where the judges cook every entry themselves. Every claim is drawn from the evidence set or the subnet's own materials; "(inferred)" marks a conclusion rather than a quote. Corrections.