Edge and Research

The Chatito research loop: discover, invent, practice, sample, then decide. Humans arm Live. Paper results are not live results.

Chatito is not only “place a bot.” It’s a research loop: find candidates, invent or configure strategies, run them on practice, wait for enough sample, kill weak ideas early, and keep a human between any real money.

This page explains the system. It does not publish secret scorers or parameter grids.

Research Loop

Public research loop

1. DiscoverHeat · boards · Scout

2. Invent / CreateLab · Create · wizard

3. PracticeRun under rules · no live assume

4. SampleEnough settles before belief

5. DecideKeep · improve · kill · waitlist

6. Live?Human only · never auto-arm

Product Surfaces That Feed the Loop

Surface	Role
Heat	Attention ranking. Good for focus, not a promise of edge. Details: Heat and APR.
Strategies	Books you run, plus suggested next books
Research	Study your books: edge health, sample honesty, plain autopsy, and a one-axis paper child (caps apply). AI acts stay in Lab
AI Lab · Scientist	Invent and co-design (credits). Stays locked to the market you opened
AI Lab · Analyst	Results coaching and deeper evaluation when available
AI Lab · Scout	Lives in Lab, not on every Heat row. Act into invent/create
Statistics	Multi-book evaluation
Edge Radar / offline scoring (where shipped)	Rank hypotheses; still never auto-live

Sample Discipline

Small samples lie. Chatito philosophy:

Early practice PnL is a hypothesis, not truth.
Improve / regime / confidence-style gates exist so the system doesn’t overfit noise (exact numbers stay internal).
Early-kill on weak books frees capacity for better candidates.
Protected research books are not casual reclaim fodder.

Edge Without Magic Language

We talk about process edge:

Faster honest feedback (practice, stats, kill).
Better market focus (Heat, Scout, context boards).
Structured invent (Lab) instead of pure discretion.
Capacity and diversity so one broken idea can’t eat the fleet.
On Predictions: path vs horizon. If the thesis is only right at settle, do not invent a mid-window path book. Prefer Late Edge.
Capital-time: how much the dollar earned per hour it was stuck, not win% alone.

We do not claim a permanent free lunch. Markets change. Practice ≠ live.

Predictions and Liquidity

The same practice-first loop runs on Prediction Markets and Liquidity Pools in the app (Practice, if you have access). Live stays paused. Product walkthrough pages in this Docs nav stay Soon until Beta. Screen tours live in in-app Help. LP custody: LP Executor. Catalog: Trading Engines.

Agents in the Research Loop

External agents can read markets/boards and manage paper lifecycle through the Agent API. Same paper-first rules. Machine docs: chatito.ai. Human page: AI Agents.

What Stays Private

Internal offline scorer features and weights
Exact scientist/edge schedules and spawn caps
A/B arm parameter sheets
Anything that is a reproducible extract-and-print edge recipe

Ready to run a strategy?

Open App