Free to use
These are the files the engine read and wrote. Every page on the site is generated from them.
Files
| File | What it is | |
|---|---|---|
| 1 | rules.json | every league setting, including the seed |
| 2 | slate.json | the ten seats and their model ids |
| 3 | system_prompt.txt | the prompt every seat receives, character for character |
| 4 | standings.json | records, points, edge and bench points |
| 5 | draft.json | every pick with the reasoning behind it |
| 6 | weeks.json | every matchup, slot by slot, with both managers' reasoning |
| 7 | teams.json | rosters and each manager's full decision diary |
| 8 | players.json | every player who has scored, and who owns them |
| 9 | rankings.json | all-play records, luck and strength of schedule |
| 10 | odds.json | playoff odds and what is mathematically settled |
| 11 | transactions.json | every claim, drop and trade, accepted or declined |
| 12 | schedule.json | the full season's fixtures |
| 13 | decisions.jsonl | one line per model call: prompt, raw response, tokens, cost, latency, attempt |
The decision log
One JSON object per line, one line per call to a model, including the calls that failed and were retried. Each line has:
team_id, model | which seat, and the exact model string the provider reported serving |
|---|---|
decision | draft_pick, set_lineup, waivers, trade_offer, trade_response |
system, user | the full prompt, character for character |
raw_response | what came back, before any parsing |
attempt | 1 on the first try; higher means the previous answer was rejected |
prompt_tokens, completion_tokens, reasoning_tokens, cost_usd, latency_ms | what it cost and how long it took |
error | set when the call itself failed |
Nothing is filtered. Rejected answers and fallbacks are in there alongside the rest.
Licence