Miners publish a codec as a salted on-chain commitment, then reveal repo|rev. The artifact that wins is the exact artifact anyone can reproduce.
GLYPHSUBNET 117
Compression ratio
pinned benchmark corpus · lower is better
Challenger enwik9 ratios — fixed benchmark
Compressed / original size on the pinned enwik9 benchmark slices (from the 1 GB Wikipedia corpus), measured with the validator’s own pipeline on reference hardware; bars fall back to the live-round value where a codec has no fixed measurement yet.
Leaderboard
ranked by enwik9 · fixed benchmark corpus
| # | UID | STATUS | CODEC | HOTKEY | ENWIK9 | FINEWEB-EDU | PILE | BLEND | ROUND-TRIP |
|---|
Winner slots
crown emission · miner conviction
| SLOT | UID | CODEC | HOTKEY | WEIGHT | EMISSION α | EARNED α | LOCKED α | REQUIRED α | CONVICTION |
|---|
Winners must keep the required alpha locked to their hotkey (Miner Conviction): a non-compliant slot's incentive burns to UID 0 for the tempo and recovers on restaking. Free allowance is max(20% of earned, 1,000 α); values marked ~ show raw stake while the chain-lock reading is not yet reported.
Benchmark
top miner vs published compressors · enwik9
Compression ratio on enwik9 — lower is better
Large Text Compression Benchmark results on enwik9 (1 GB Wikipedia), alongside the top miner codec's enwik9 ratio measured by the live validator run.
Validators
permit holders securing evaluation
| UID | ROLE | HOTKEY | STAKE α | DIVIDENDS | LAST UPDATE |
|---|
Mechanism
how the crown works
Validators stream fresh, unseen text through each revealed codec. A single unfaithful round-trip byte disqualifies the run before ratio is even scored.
The best qualifying ratio is crowned champion and receives the full miner emission. There is no partial credit — beat the champion or earn nothing.
When no challenger beats the bar, emission is directed to UID 0 — the subnet owner's address, acting as the burn sink — instead of being paid out — idle windows don't pay.