01 / Agents Nine agents. One desk. Each agent is a way of reaching what the engine has already measured. They share one rule: every connection is measured, mechanical, or labeled conjecture, and none of them will tell you what to do. Four of the nine cost no model call per question.
All09 Read05 Watch01 Verify02 Connect01 ↑ ↓ to walk the index
Index Agent Type Model calls
A01 Market State The whole market's structure, read across all six factor families at once. read 0 A02 Token Briefing One token, read inside the market it is trading in. read 0–2 Incident Mechanics A post-mortem when something breaks: trigger, amplifier and mechanism held apart. read 1–2 A04 Mechanism How a transmission channel works, from the curated graph. read 1–2 A05 What Changed The difference between the last time you looked and now. read 0 A06 Watch A standing condition, in plain language, that alerts you when it becomes true. watch 1 A07 Claim Check Paste a post or a thread. Get each claim checked against the desk's own data. verify 1 A08 The Record The desk's own track record, including the claims it had to drop. verify 0 A09 Tool Endpoint The desk's read-only tools, for an agent you already run. connect 0 Incident Mechanics Grounds on the divergence flags that fired, the candidate transmission channels in the mechanism graph, the immutable news corpus from that week and typed claims with their claimant and incentive. It never blesses a single cause; where a causal story comes from a source, it is attributed to that source.
Reads How it answers Plan Gather Compute Generate Verify Guardrail Every number must resolve to a store row; a mechanical claim must cite a real graph edge. Failures are dropped, never softened.
Model calls 1–2 one plan (often skipped) + one generation
You get A mechanism-level account, with every causal link either documented in the graph or labeled conjecture.
It won't No hindsight: it reasons only over what had been captured at that moment. 
LensAI