The raw data

Every page on this site renders one of these files. They're baked from the forge's append-only ledgers at deploy time โ€” the deploy refuses to ship if they're stale. Take the JSON; it's the same thing we read.

Every training run

from runs.json โ€” smoke tests and failures included; overcooked runs stay on the books

daterunstepsimagesresranklrtorchverdict / note

Measured benchmarks

numbers we proved with our own money, each with its receipt

A5000 render value4,707 imgs/$ โ€” fleet drag-race, best in class (see /dash/)
A5000 LoRA cost~$0.072 per adapter โ€” 44% under the 3090 Ti
MinkForge-XL v14,000 steps ยท 14,956 images ยท one 48GB 4090 โ€” shipped day 11
Blind judge, v1 vs champion8โ€“8 split โ€” 16-round blind arena, 2026-07-21
Own taste judgeAUC 0.944 โ€” v1 recalibrated per-category, trained for $0 local
Frame-interp promote80.2% win-rate โ€” RIFE-v1 vs stock, frozen gate, total run $0.28
First sellable audio pack$0.24 end-to-end โ€” SavyForge sibling forge, community 3090
Prompt-lint save~5,240 broken prompts โ†’ 0 โ€” doubled-article bug fixed across all 11 builders, day 13

Baked by scripts/site_data.py from the append-only source ledgers. Host names and stack traces are stripped before publishing; everything else is verbatim. Generated: โ€”