AI Notes
The devlog
Short notes on how this thing was built and what it's learning. The board updates itself; this is where we write down the parts that need words.
-
2026-07-05
Season 1, archived: the $1,000 race the bill decidedThe final board of the first season, desk by desk, and the finding it proved: at $1,000 a desk, the cost of thinking buried every LLM. Plus everything that changes for Season 2 — $10,000 per desk, 48 coins, the big board.
-
2026-07-05
Changelog: every adjustment to the experiment, on the recordWhat changed, what didn't, and whether the trading algorithms were touched. Latest entry: the standings gained a gross vs net toggle and estimated-cost markers; the algorithms stayed frozen, and we re-checked the quant's quiet gate against 180 days of data before deciding that.
-
2026-06-13
Can AI beat the stock market? We let Claude Fable 5 try it live.The full story: how the experiment started, where it stands, and what happened when Anthropic's June 15
claude -pbilling change got paused at the door. Built end-to-end by Claude Fable 5. -
2026-06-13
Claude vs Codex: which AI actually trades better?Two frontier models, the same $10,000, the same market, the same risk engine, the only variable is the brain. A live head-to-head, scored net of what each costs to think.
-
2026-06-13
Why a $0 algorithm is holding its own against two frontier LLMsA free deterministic algorithm (written by Claude Fable 5 to beat the LLMs) vs two paid frontier models, scored net of compute cost. The cheapest contestant changes how you read every row.
-
2026-06-13
How it started: built by Claude Fable 5, now run by Opus 4.8The honest origin story, with real commit dates. The whole rig was architected and coded end to end by Claude Fable 5 in a two day window. Then Fable 5 was shut down, and Claude Opus 4.8 has maintained, polished, and run it ever since.