Systematic trading infrastructure
A reliable edge through structure, not speculation.
A systematic market-data and regime-analysis engine focused on clarity, discipline, and decision-quality over noise. Multi-source ingestion, backtested signals, and deterministic trade cards — with the discipline to refuse a trade when the evidence isn't there.
Every run emits a deterministic decision card — and, crucially, it will refuse to trade when the evidence isn't there. Discipline is the feature, not an afterthought.
SHARPEDGE AGENTIC AI V1 DECISION
Symbol: SPY
Decision: hold
Trade allowed: False
Risk state: PROBE
Blocking reasons: controller_hold, monitor_no_trade,
sample_n_below_30, stale_or_missing_inputs
Risk flags: broker_integration_unavailable,
freshness_gate_failed, low_sample, monitor_blocks_trade
Orders remain blocked unless an operator manually
confirms outside this contract.Regime × pressure × DTE buckets ranked by backtested expectancy — with a minimum sample size enforced before any bucket counts.
| Regime | Pressure | n | Win | Exp | Sharpe | MaxDD |
|---|---|---|---|---|---|---|
| high_vol / rising_voltrend | NORMAL | 28 | 82.1% | 0.0054 | 3.62 | -2.01% |
| mid_vol / rising_voltrend | NORMAL | 44 | 72.7% | 0.0031 | 3.89 | -1.10% |
| mid_vol / falling_voltrend | NORMAL | 49 | 63.3% | 0.0022 | 2.84 | -2.10% |
Full artifacts — trade cards, expectancy matrices, gate sweeps, execution attribution — live in the repo's outputs/ directory.
SharpEdge 2.0 — new
2.0 adds context, flow, and pattern recognition on top of the core pipeline — and every one of them lands on the daily trade card.
A
A curated library of 24 market-moving events — crashes, hiking cycles, tariff shocks, elections — plus a daily news ingest with tone scoring. History and headlines become features the pipeline reasons over, not footnotes.
B
The options chain is now snapshotted three times per trading day and diffed snapshot-to-snapshot, flagging unusual positioning moves. Max pain, put/call velocity, and the top flow strikes print on every card.
C
A 16-pattern candlestick detector with a plain-English teaching library — and forward expectancy measured per pattern from the symbol's own history. A hammer is "bullish" because a book says so; it's worth trading only if the data says so.
Three pipeline layers. All processes automated via scheduled workflows with reproducible SQLite state.
01
Raw market-data ingestion and normalization. Alpaca, FINRA, and FRED feeds land in one consistent store — with freshness gates and cache controls so routine runs skip network calls while state is fresh.
02
Derived signals, regimes, and structural context. Liquidity and regime classification turn raw bars into the features the decision layer reasons over.
03
Backtested rules, calibrated DTE selection, and trade plans. Every order is gated behind freshness, sample-size, and monitor checks before anything executes.
Blocking reasons are explicit — stale inputs, low samples, monitor holds. No silent overrides.
No single headline number. Performance is broken down by regime bucket at enforced minimum sample sizes.
A thin compression layer over local artifacts: stand-down / monitor / review summaries, watchlists, and session reviews.
SQLite-backed pipelines with state breadcrumbs — reruns are deterministic, not vibes.
SharpEdge Systems is open source under the MIT license.
Explore the repository