NOVA

SN68 NOVA is a bounty board for drug candidates, where a target protein is posted, anyone may submit molecules predicted to stick to it, and one winner per round takes the whole prize.

The commodity, explained from zero

The product is a shortlist of candidate drugs. A drug works by binding to a target protein, and the first step is screening enormous libraries of molecules for the few predicted to bind tightly. NOVA runs that screen as a contest. The site calls it "a decentralized engine for therapeutic discovery" and "a back-to-back hackathon" that runs around the clock. Two kinds of entry are accepted: small molecules from a billion-scale catalogue, and nanobodies, short antibody-like protein sequences the miner designs.

The shortlist has value because the conventional route is slow: the site puts a new drug at "$1 billion USD" and "10+ years". The partnerships page offers "Screening as a Service" ("Submit a target. NOVA's network generates and ranks lists"), co-development and custom competitions, and names three partners: OnePot.AI, which synthesizes hits "in days, not months", Yalotein and Diagen AI. The value-stack page counts "11M+" molecules and "16K+" nanobodies identified. No price is given; the route in is an email.

Why it is on Bittensor at all

The site's stated difference is who may take part: "Anyone can participate", and NOVA "rewards performance, not credentials". It does not compare cost or speed with a pharmaceutical company's own screening. Here the comparison breaks: on a real bounty board the poster pays the prize, while NOVA's prize is the subnet's emission, its share of newly minted TAO, paid whether or not any drug company asked for that target.

How the work gets done

Every 361 blocks a new round starts, and the target proteins are announced on Discord a day ahead. Miners, the hunters, search for molecules or design nanobody sequences, encrypt the answer so it cannot be read until the round closes, upload it to a public GitHub repository and record a pointer on chain. Validators, the judges, decrypt every entry, throw out any that breaks the format or uniqueness rules, and score the rest with two prediction models: Boltz-2 estimates how tightly each small molecule binds, and a modified BoltzGen ranks each nanobody on fold confidence, binding strength and stability. The best molecule and the best nanobody each win their whole share of the round's emission; ties go to whoever committed first. The miner guide warns that running the supplied baseline unchanged "will NOT result in incentive distribution".

How you would know it works

The research page carries dated write-ups: a scoring-function note from 2025-12-23 reporting an AUC of about 0.81 for hit identification, and an H1 2026 report from 2026-07-08 on moving hits "From the Network to the Lab Bench". The dashboard page renders empty without a browser.

What is missing

The chain identity has no contact, no Discord and no additional text, and the site's Discord link is the shared Bittensor server. The miner task guide lives in a docs folder the README never links. The registration cost is not stated (about 0.12 TAO on chain on 2026-09-09). Validators need an API key obtained by direct message to the team. The chain snapshot lists 2 active miners. There is no support path beyond GitHub issues and an email.

Go deeper

• What NOVA is, in plain words
• Dated research results
• Miner guide with scoring rules

Metaphor: a bounty board for drug candidates. Every claim is drawn from the evidence set or the subnet's own materials; "(inferred)" marks a conclusion rather than a quote. Corrections.