Tradesight
Self-hosted Python strategy research and Alpaca paper trading with broker reconciliation, evidence gates, and live execution disabled.
Install / Use
npx skills add rmbell09-lang/tradesightInstalls into whichever agent you are using.
README
TradeSight
TradeSight is a self-hosted strategy-research and paper-trading application. It evaluates strategy families against historical data, preserves champion/challenger evidence, and connects to an Alpaca paper account for broker-simulated orders and reconciliation.
TradeSight is research software. It is not financial advice, does not promise returns, and does not contain an enabled live-money execution path.
What is real in the current release
- Alpaca market data is labeled
REALonly after source verification. - Orders, positions, and fills from the Alpaca paper endpoint are labeled
PAPER. - Broker/local accounting is reconciled from a defined paper-account epoch.
- Historical local P&L that lacks complete broker proof is preserved as
LEGACY / UNVERIFIEDand excluded from trusted performance. - Strategy Lab reads the persisted optimizer, tournament database, champion, challengers, and rejection evidence.
- The Live Readiness screen shows mandatory evidence gates but cannot activate live trading.
- Polymarket records are archived and hidden when stale; the scanner is not currently maintained.
Requirements
- Python 3.11 or newer
- macOS or Linux
- Alpaca paper credentials for broker-backed market and trading evidence
Supported installation
Extract the verified TradeSight source release, then run:
python3 install_tradesight.py
./launch_tradesight.sh
The installer:
- selects Python 3.11 or newer;
- creates an isolated
.runtimeenvironment; - installs declared runtime dependencies;
- installs the local extracted source without cloning another repository; and
- verifies the CLI in
PAPER_ONLYmode.
Release verification may use --no-register so an extracted test copy cannot
replace the canonical CLI pointer.
The dashboard opens at http://127.0.0.1:5001.
For development from the canonical source tree:
python3 install_tradesight.py
.runtime/bin/python -m pip install -r requirements-dev.txt
.runtime/bin/python -m pytest tests -q
.runtime/bin/tradesight --no-browser
Data labels
REAL: verified external market evidence.PAPER: broker-simulated trading evidence.DEMO: explicit development-only synthetic data.STALE: retained data too old to support a current decision.UNVERIFIED: local evidence that has not met broker-proof requirements.
TradeSight never silently substitutes generated prices for unavailable market data.
Strategy lifecycle
Candidate → Backtest → Out-of-sample → Shadow paper → Qualified → Champion → Retired
Daily evaluation continues with new real data. Full searches are evidence-triggered. A challenger cannot promote automatically and must pass out-of-sample, walk-forward, multiple-testing, Monte Carlo, and forward-paper evidence gates.
Live-money status
This release is technically locked to paper trading. The readiness dashboard requires, among other evidence:
- 30 verified accounting sessions with zero unexplained differences;
- 60 forward-paper sessions and 100 broker-confirmed forward trades;
- a frozen strategy with a passing qualification receipt;
- protected controls, working outbound safety alerts, and passed failure drills; and
- the operator's explicit strategy approval and hard per-trade, daily, and total-pilot loss limits.
Passing those gates would only support a separate canary-build review. It would not automatically enable trading or guarantee profitability.
Release integrity
Build a local release archive with:
.runtime/bin/python scripts/build_release.py
.runtime/bin/python scripts/verify_release.py dist/releases/tradesight-1.2.0-paper-only.zip
The builder uses an allowlist, rejects runtime/private state and common credential patterns, writes a SHA-256 manifest, and never publishes the archive.
Tests
Run the complete canonical suite instead of relying on a hardcoded README count:
.runtime/bin/python -m pytest tests -q
The dashboard displays the latest test receipt and marks it stale when it ages out.
Safety boundaries
- Paper endpoint only.
- No live activation endpoint.
- No automatic strategy promotion.
- No automatic scaling.
- New paper entries fail closed on stale signals or unverified accounting.
- Exits remain available during operational suspensions.
License
MIT. Third-party data and brokerage services retain their own terms.
Related Skills
node-connect
385.5kDiagnose OpenClaw Android, iOS, or macOS node pairing, QR/setup code, route, auth, and connection failures.
python-debugpy
385.5kDebug Python with pdb, breakpoint(), post-mortem inspection, and debugpy remote attach.
skill-creator
385.5kCreate, edit, audit, tidy, validate, or restructure AgentSkills and SKILL.md files.
claude-opus-4-5-migration
140.6kMigrate prompts and code from Claude Sonnet 4.0, Sonnet 4.5, or Opus 4.1 to Opus 4.5
