Aegis — BTC Trading Signal Platform

I trade BTC discretionarily and wanted to stop running the same checklist by hand every session. Aegis is the system version of that checklist: it ingests live market and derivatives data, runs a deterministic regime + signal engine, applies hard risk gates, and proposes trades I confirm manually — it never auto-executes. The design constraint that shaped everything: a trading tool that quietly lies to you is worse than no tool. So the engineering bias is toward honesty — knowing when the data is real, when a signal is trustworthy, and when the whole thing should stay silent.

1. 01

   Domain-driven architecture

   Structured as clean DDD: domains (market_data, derivatives, features, strategies, execution, review) hold the logic; infrastructure (OKX, Coinglass, Telegram, SQLite) holds the adapters; apps (worker, api, backtest) are the entry points. Strategy is a Protocol — feed it a context of daily/4H trend, structure, and crowding features, get back a typed SignalProposal with entry zone, stop, TP1/TP2, R:R, and confidence. Swapping a data source or adding a strategy doesn't touch the core.

2. 02

   Live data + anti-lying guardrails

   The original MVP leaned on Binance futures, which returns HTTP 451 in my environment — so it silently fell back to demo data and produced misleading alerts. I rebuilt derivatives on OKX public endpoints and tagged every snapshot with source and is_live. Snapshot-health checks now run before anything publishes: the daily brief is skipped and auto-published signal alerts are blocked whenever derivatives data is fallback or stale. The system would rather say nothing than say something false.

3. 03

   Signal → confirm → journal loop

   A scheduler runs the scan loop on an interval; signals push to Telegram (rate-limited to 04:00–22:00 PT, de-duplicated, with a once-a-day heartbeat so silence is informative). I confirm or reject from Telegram or the dashboard, which also carries a manual-trade journal, a screenshot-draft review queue, a weekly report, OKX read-only account sync, PnL reconciliation, and a live strategy-parameter panel. A first-pass backtest runner exists too — and I deliberately label its 100% win rate as a smoke test, not an investable number: the window is short and trend-biased, with no slippage, fees, or funding yet.

A 3,500-line, 25-test platform that runs end-to-end on live OKX data and surfaces trustworthy, manually-confirmed BTC signals — with the discipline to go quiet when its inputs degrade. The next milestones are the honest ones: a production-grade backtest with fees / funding / slippage and historical derivatives replay, a public HTTPS endpoint for TradingView webhooks, and richer health observability. The real lesson mirrors my C++ quant work — the math behind a signal is the easy part; the hard part is the engineering that makes the system safe to actually trust with money.