Overview
Session equity, order stats, source P&L, tournament state and signal history.
Execution Engine
Small, compiled, and built to stay hot.
The trading process is a single C++23 binary with paper/live parity, idempotent reconnects, reseed logic, and hard boundaries around sizing and timing.
Shared paper/live pathThe demo, paper sessions and live orders all exercise the same strategy code path.
Feed fan-outfeed_proxy keeps one upstream socket per venue and rebroadcasts to every bot with bounded queues.
Persistence firstOrders, fills, feature snapshots, strategy versions and run results persist to Postgres.
Architecture
The data loop is simple by design.
Ingest, fan-out, execute, journal and command. Each layer has one job, so failures are easier to isolate and latency remains predictable.
01
Ingest
Venue streams and oracle state enter once.
02
Fan-out
Bounded queues distribute the firehose.
03
Execute
C++ bots scan windows and route orders.
04
Journal
Postgres captures fills and features.
05
Command
Cockpit and PolyTron read the same state.
Real cockpit screenshots
The cockpit is already built.
These are live product captures from the PolyPulse dashboard: overview, market charts, fleet control and strategy hot reload. No decorative mockups, no stock UI.
Operator Surface
Everything important is visible and editable.
The cockpit exposes live fills, positions, market charts, fleet control, strategy hot reload and AI apply cards without exposing infrastructure internals.
Markets
Candles, active windows and live edge bars by symbol.
Fleet
Bot cards, model selectors, wallet state, history and start/stop controls.
Strategy
Range-validated safe knobs with save and hot-reload controls.
Logs
Unified event trail across engine, cockpit and AI Advisor decisions.
PolyTron
Apply cards and decision history from the embedded AI Advisor.
Put the engine in front of live books.
Launch a paper bot, open the cockpit, and watch how the system reads windows, proposes adjustments, and manages positions.