How the bot works

The bot runs a continuous loop. Every cycle (about every two minutes for the directional scanner) it repeats the same disciplined pipeline for each symbol on your watchlist.

The trade pipeline

1. Scan — walk the resolved watchlist (often 90+ symbols across sectors, no hard cap).
2. Forecast — the Kronos / Chronos-2 model predicts a price path: a direction, a confidence, an expected move, and a p10–p90 cone.
3. Signals — layer in independent evidence: technical indicators (trend/momentum) and a factor model (quality, momentum, value, earnings). The Roundtable requires these — without them the judges reject everything.
4. Build candidates — from the catalog of 43 option structures, assemble the ones that fit, each scored with two probabilities of profit (market-implied and model), expected value, max profit/loss, breakevens, and a margin estimate.
5. Select by regime — match the structure to the forecast: direction plus the volatility risk premium (is the expected move bigger or smaller than the option chain is pricing?). Rank by quality × regime fit.
6. Roundtable vote — the AI judges vote on the top-ranked idea; it needs at least the configured number of approvals (naked/undefined-risk trades require unanimity). The council walks #1 → #2 → #3 until one is approved. Optionally the analyst can veto an approved trade.
7. Size — size against Schwab's real margin (from an order preview), bounded by your per-trade cap and available buying power. A pre-flight buying-power guard skips any order that would over-commit, so you don't get rejections.
8. Place & reconcile — submit the order, then correct the ledger to the actual fill. A position is only managed/closed once its opening order is confirmed filled. A rejected or skipped order never records a phantom position.
9. Manage — a fast poller watches every open position on live bid/ask and runs the exits: take-profit, stop-loss, trailing stops, spread auto-close at ~50% of credit, and Roundtable-gated wheel rolls. Exits keep running even when new trades are paused.

Why premium harvesting, not prediction

In out-of-sample tests, the forecasters landed around or below an always-long ~58% baseline (Kronos 42.8%, Chronos-Bolt 51.2%, Chronos-2 48.8%, a TA ensemble 54.9%). The structural edge is therefore selling premium — the wheel, PMCC, and defined-risk credit spreads — while the forecaster + Roundtable filter out the worst setups and manage risk. See The four engines and Strategy catalog.