Every miner gets the same random start, the same 0.82 billion tokens, the same six hours, the same GPU. Only two files differ: the model and the training recipe. This is our measurement log.
Score is bpb — how many bits the model needs per byte of held-out text. The uniform-guess floor is 15.617; everything below is learning. Emission is winner-take-all, so second place earns exactly what last place earns.
Every dot is a completed six-hour run. Hover for detail.
Each bar is a single change measured against an otherwise identical run. Negative is an improvement. The headline: the learning-rate schedule dominates, and model size does nothing at all.
Orthogonalised-momentum group, full six-hour runs at 218.5M.
Monotonic and flattening — 0.04→0.02 buys 0.0054, 0.02→0.01 buys 0.0018. We are at the optimum.
Four full runs, identical recipe, 218.5M → 295.6M.
Total spread 0.012 across a 35% range of size, and not monotonic. Chain #1 is a 130.7M model — smaller than anything here.
Eight full six-hour runs and four short ablations. The position bar places each score on the same axis as the race chart above.
| Run | Params | Tok/step | Ortho LR | Tokens | Passes | Tok/s | bpb | Position |
|---|
"Passes" counts how many times the run swept the 0.82B-token shard. Over-repeating hurts: the best chain result sits near 2.4 passes, and a 3.1-pass run on the same architecture scored 0.11 worse.
Five boxes testing whether a smaller model wins, since size proved flat above 218M.
| Run | Params | Ortho LR | Tok/s | Step | State |
|---|
Training-only entries — our recipe on other miners' architectures. These cost nothing; the subnet operator pays for the GPU.
| Hotkey | Architecture | State | Queued |
|---|
Fewer than one submission in five ever produces a score — the rest die to infrastructure faults or the anti-cheat gate. The queue runs roughly a day deep.
| Hotkey | Architecture | State | Queued |
|---|
Any miner can submit a training recipe against a published architecture. The architecture's owner is then credited with the best score anyone reaches on it — so a good design earns even while someone else trains it.
| Architecture | Owner | Best bpb | Position |
|---|
PRISM's entire share of subnet emission collapses onto a single hotkey each epoch. Rank 2 earns zero. A run that improves our score but doesn't take first place is worth nothing in emission — only in information.
Beating everyone on a borrowed architecture credits its owner with our exact score — a guaranteed tie. Ties break by lexicographically smallest hotkey, which is why we mined hotkeys beginning 0x000000 and sent them to the architectures whose owners have low-valued keys.
Only rank matters under winner-take-all; the lattice value is a re-mapping so higher means better.
A cheat or suspicious verdict is terminal: score 0, no retry.