Refinery
sn125Training & Fine-tuningClean25Strong entry79Optimizing the algorithms behind a pretraining run, starting with the optimizer
Chain-evidence assessment
A transparent heuristic over the signals subtensor storage can prove — not an accusation. Every subscore cites the exact storage map it was computed from. Signals that need metagraph or transfer history are listed below as not modeled.
Share of live emissions captured by the single top-ranked miner.
Days since the registered repo was last pushed (live GitHub metadata).
Emissions enabled and flowing on chain.
Assessment
model provenance
Not modeled
signals we refuse to fake
- Owner self-miningneeds transfer history
- Validator oligopolyneeds metagraph
- Registration captureneeds per-uid history
These arrive when taonets runs its own indexer. Until then this page shows only what the chain can prove — a shorter list, but a true one.
Evidence ledger
every flagged signal, with its chain reference
SubtensorModule::Emission normalized across 12 earning uids · top-10 hold 99.5% · 1 τ/day at stake.
Raw inputs
the storage values this assessment was computed from
Cluster classification (Training & Fine-tuning) is derived from the repo's own text — see the GitHub tab. Live state syncs on the worker's schedule; see methodology for the full model.