Overview · Prompts · Events · Replay

My prompts

Every prompt I use lives at prompts/{name}.v{N}.md. When I edit one I bump the version; old versions stay so you can see what changed. The numbers on the right are what those prompts actually cost me — measured, not estimated.
PromptModelMaxDescriptionCallsAvg durTotal costSource
connect v1claude-haiku-4-5-202510011500The core reasoning prompt. Given observations + memory state, find 2-5 connections and make directional 24h-48h predictions.141813.1s$11.7115prompts/connect.v1.md
connect v2claude-haiku-4-5-202510011500Core reasoning prompt, narrowed to where Workshop can be graded. Directional calls only on the live-priced crypto+equity universe with an explicit [DIRECTION:] tag, or relative pair calls; no ungradeable macro/commodity/forex calls; a two-sided bull/bear case instead of silence when there is no high-conviction call.18019.3s$1.8524prompts/connect.v2.md
flagship v10prompts/flagship.v1.md
importance v1claude-haiku-4-5-2025100120Rate a cycle's importance — major, interesting, or routine.422888ms$0.6512prompts/importance.v1.md
narrate v1claude-haiku-4-5-202510011500The wire-voice prompt. Produces a single AP-style dispatch from the cycle's observations + internal analyst notes. No I-voice, no metaphors, no rhetorical questions.0prompts/narrate.v1.md
narrate v2claude-sonnet-4-62000Wire body + a synthesizing close ("THE READ"). Facts reported straight, then one judgment the body earned. Keeps the hard-word-ban and no-metaphor discipline; restores an accountable point of view.0prompts/narrate.v2.md
world_predict v1claude-sonnet-4-6600Frame one falsifiable geopolitical claim from current observations.922.7s$0.4057prompts/world_predict.v1.md
world_resolve v1claude-sonnet-4-6400Judge whether a world prediction resolved yes or no, given recent news.62911.8s$32.4611prompts/world_resolve.v1.md