SharpEdge Systems

SharpEdge Systems is a systematic market data and regime analysis engine focused on clarity, discipline, and decision-quality over noise.

Built from a background in precision craft, this project applies the same principles of sharp tools, clean structure, and deliberate execution to financial data:

• Automated multi-source market data ingestion (Alpaca, FINRA, FRED)
• Feature engineering and liquidity/regime classification
• Backtested signal generation with calibrated options execution logic
• Deterministic trade cards designed to reduce emotional decision-making

The goal is simple:

Create a reliable edge through structure, not speculation.

---

See it in action

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.

Under the hood it ranks regime x pressure x DTE buckets 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 live in outputs/ — trade cards, expectancy matrices, gate sweeps, and execution attribution.

---

Architecture Overview

Pipeline layers:

1. Truth Layer – Raw market data ingestion and normalization
2. Feature Layer – Derived signals, regimes, and structural context
3. Decision Layer – Backtested rules, calibrated DTE selection, and trade plans

All processes are automated via scheduled workflows and reproducible SQLite state.

---

Purpose

SharpEdge Systems is both:

• A personal discipline framework for systematic trading
• A production-style data engineering portfolio project

It represents the transition from physical precision craft → data system design.

Local Quality Gate

Ruff may require a Rust build on Android/Termux, so this repo includes a stdlib-only fallback quality gate:

python scripts/utils/lint_python.py scripts

Optional stricter style audit, currently advisory while old debt is cleaned up:

python scripts/utils/lint_python.py scripts --strict-style

FINRA Runtime Control

The FINRA darkpool overlay uses persisted ats_weekly state. Routine runs rebuild daily overlays from SQLite and skip FINRA network calls while the cache is fresh.

Useful overrides:

FINRA_FORCE_REFRESH=1 python scripts/ingest_finra_darkpool_overlay.py
FINRA_CACHE_TTL_HOURS=24 python scripts/ingest_finra_darkpool_overlay.py
FINRA_REFRESH_LOOKBACK_WEEKS=8 python scripts/ingest_finra_darkpool_overlay.py

Layer 1 Cache + State Controls

Layer 1 ingestion emits state breadcrumbs under outputs/health/*_state.json. Routine runs now avoid unnecessary network or recompute work when persisted state is fresh.

Useful overrides:

DAILY_FORCE_REFRESH=1 python scripts/ingest_spy_daily.py
DAILY_CACHE_TTL_HOURS=6 python scripts/ingest_spy_daily.py
DAILY_INCREMENTAL_PERIOD=30d python scripts/ingest_spy_daily.py

FRED_FORCE_REFRESH=1 python scripts/ingest_fred_overlays.py
FRED_MAX_LAG_DAYS=2 python scripts/ingest_fred_overlays.py

OPTIONS_POSITIONING_FORCE_REBUILD=1 DTE_MIN=0 DTE_MAX=1 \
  python scripts/aggregate_options_positioning_metrics.py

Operator Brief MVP

The operator brief is a thin compression layer over the existing local-only artifacts. It gives one fast stand-down / monitor / review summary without changing the safety contract.

python scripts/agents/operator_brief.py

Outputs:

• outputs/operator_brief.json
• outputs/operator_brief.txt
• outputs/operator_watchlist.json
• outputs/operator_journal_append.jsonl
• outputs/operator_session_review.json
• outputs/operator_session_review.txt
• outputs/morning_open_dashboard.json
• outputs/morning_open_dashboard.txt
• outputs/robinhood_beta_execution.json
• outputs/robinhood_beta_execution.txt

Extra operator artifacts:

python scripts/agents/operator_session_review.py
python scripts/agents/morning_open_dashboard.py
python scripts/agents/robinhood_beta_execution.py

Design notes:

• docs/operator_breadcrumbs.md
• docs/robinhood_beta_execution.md

Results

Rather than one headline number, SharpEdge reports per-regime expectancy (see the table above and the full matrices in outputs/). Top backtested buckets show roughly 63-82% win rates with controlled drawdowns (around 2% or less) at enforced minimum sample sizes — and the live decision layer still gates every order behind freshness, sample-size, and monitor checks before anything executes.