NewTradingSystem

This repository is scaffolded as a broker-agnostic, multi-strategy trading platform.

Start with:

• index.html for the platform redesign brief
• docs/legacy-extraction-portal.html for the published legacy-extraction and moving-parts overview
• docs/legacy-inventory.md for the concrete list of high-value legacy strategies, scripts, workflows, and migration buckets
• docs/current-state.md for the plain-English truth of what is actually built today
• docs/plans/2026-04-03-legacy-extraction-master-plan.md for the migration program from the old trading repo
• docs/plans/2026-04-03-market-intelligence-master-plan.md for the market-intelligence and growth strategy
• tasks/master-roadmap.md for the execution backlog
• loop-system/README.md for the autonomous build/research loop
• .env.paper.local.example for the local paper-broker credential shape

The repo is organized so the trading system, strategies, reusable SDKs, dashboards, schemas, and infrastructure can evolve independently.

Core operating assumption:

• Paper-only is the default operating mode.
• Strategies never place broker orders directly.
• The platform owns execution, risk, reconciliation, persistence, and promotion gates.
• Every strategy must pass replay, internal paper trading, and broker-paper validation before any live capital is considered.
• No strategy moves to real money without explicit live-approval criteria and manual promotion.
• Real broker credentials live only in a local .env.paper.local file, which is git-ignored and should stay machine-local.

First runnable vertical slice:

• src/new_trading_system/services/worker.py: paper worker orchestration
• src/new_trading_system/services/execution_engine.py: intent-to-broker execution path with ledger-backed intraday checks
• src/new_trading_system/services/risk_engine.py: paper-only, opening-vs-closing, whitelist, and cumulative-risk guardrails extracted from the legacy gateway
• src/new_trading_system/services/reconciliation.py: position reconciliation, verification, and stale-order handling
• src/new_trading_system/services/portfolio_ledger.py: canonical SQLite-backed ledger and dashboard summary
• src/new_trading_system/adapters/internal_paper.py: internal simulator broker
• src/new_trading_system/adapters/alpaca_paper.py: Alpaca paper broker and market-data adapter
• src/new_trading_system/strategies/legacy_iron_condor.py: first extracted legacy options strategy plugin with profit/stop/DTE exits, next-available expiry discovery inside the configured DTE window, a basic VIX regime gate when volatility data is available, duplicate-expiry blocking, and broken-structure alerts
• apps/dashboard/index.html: static operator dashboard for the generated summary JSON
• var/worker.lock: overlap protection so two worker runs cannot mutate the same paper state concurrently
• named runtime accounts can isolate their own state DB, strategy state, worker lock, internal paper state, dashboard summary, and paper-broker credentials via --account <id> plus NTS_ACCOUNT_<ID>_* env keys
• run-accounts can execute several named accounts sequentially in one command while preserving per-account isolation
• run-loop and run-accounts can now stay up continuously with --forever for unattended paper cycling
• scorecard turns the canonical ledger into a paper-readiness report so strategy claims stay tied to observed evidence
• replay now runs a first synthetic internal-paper scenario set for the legacy iron-condor strategy so replay coverage starts with deterministic execution evidence instead of historical claims we do not have yet
• replay --scenario-set historical-bars now runs a first honest historical backtest using real SPY daily bars plus modeled option pricing, so the repo can produce bar-driven profitability evidence instead of only synthetic scenarios
• the historical-bars replay now bounds exits to the contract expiry window and skips trades when the future bar data is insufficient, instead of allowing impossible post-expiry holds across data gaps
• the historical-bars replay now also respects the live strategy's max_open_structures and duplicate-expiry constraints, so long-window replay samples are smaller but better aligned with the real runtime behavior
• replay reports now persist under var/replay/, and scorecard consumes them so replay evidence is cumulative instead of session-local
• promotion now evaluates saved replay evidence plus paper scorecard evidence against strategy-defined thresholds and explains exactly why a strategy is or is not ready to advance
• autonomous-runner now adds a market-hours supervisor with a halt file, runtime status/heartbeat files, sequential multi-account execution, and optional post-cycle reconcile/verify hooks
• autonomous-status, autonomous-halt, and autonomous-resume now expose first-class runtime controls around that supervisor without changing trading logic
• when the legacy trading repo is present as a sibling checkout, scorecard and promotion now surface its closed-trade ledger as explicitly labeled legacy reference evidence rather than silently counting it as current-system proof

Quick start:

• python3 -m pip install -e ".[dev]"
• bash scripts/verify.sh
• PYTHONPATH=src python3 -m new_trading_system.cli run-once --broker internal-paper
• PYTHONPATH=src python3 -m new_trading_system.cli run-loop --account primary --broker internal-paper --execute --forever --interval-seconds 60
• PYTHONPATH=src python3 -m new_trading_system.cli run-once --account primary --broker internal-paper --execute
• PYTHONPATH=src python3 -m new_trading_system.cli run-accounts --account primary --account swing-two --broker internal-paper --execute
• PYTHONPATH=src python3 -m new_trading_system.cli run-accounts --account primary --account swing-two --broker internal-paper --execute --forever --interval-seconds 60
• PYTHONPATH=src python3 -m new_trading_system.cli autonomous-runner --account primary --broker internal-paper --execute --forever --interval-seconds 300 --verify-after-cycle
• PYTHONPATH=src python3 -m new_trading_system.cli autonomous-status
• PYTHONPATH=src python3 -m new_trading_system.cli autonomous-halt --reason "manual operator halt"
• PYTHONPATH=src python3 -m new_trading_system.cli autonomous-resume
• PYTHONPATH=src python3 -m new_trading_system.cli scorecard --account primary --broker internal-paper
• PYTHONPATH=src python3 -m new_trading_system.cli replay --strategy legacy-iron-condor
• PYTHONPATH=src python3 -m new_trading_system.cli replay --scenario-set historical-bars --days 90 --strategy legacy-iron-condor
• PYTHONPATH=src python3 -m new_trading_system.cli promotion --account primary --broker internal-paper --strategy legacy-iron-condor
• PYTHONPATH=src python3 -m new_trading_system.cli run-once --broker internal-paper --execute
• PYTHONPATH=src python3 -m new_trading_system.cli reconcile --broker internal-paper
• PYTHONPATH=src python3 -m new_trading_system.cli verify --broker internal-paper
• PYTHONPATH=src python3 -m new_trading_system.cli run-once --broker alpaca-paper
• PYTHONPATH=src python3 -m new_trading_system.cli verify --account swing-two --broker alpaca-paper
• PYTHONPATH=src python3 -m new_trading_system.cli verify --broker alpaca-paper
• PYTHONPATH=src python3 -m new_trading_system.cli dashboard
• pytest -q

Worktree note:

• paper credentials stay in the git-ignored root .env.paper.local
• worktree checkouts automatically fall back to that root env file, so feature branches can verify Alpaca paper behavior without copying secrets into each worktree

Multi-account note:

• the default runtime account keeps the original paths and commands unchanged
• named accounts use --account <id> and isolate state under var/accounts/<id>/
• named dashboard summaries are written to apps/dashboard/data/accounts/<id>.json
• fleet runs also update apps/dashboard/data/accounts/summary.json with the latest cross-account iteration summary
• named broker credentials can be supplied with NTS_ACCOUNT_<ID>_BROKER, NTS_ACCOUNT_<ID>_ALPACA_PAPER_API_KEY, and NTS_ACCOUNT_<ID>_ALPACA_PAPER_API_SECRET
• run-accounts runs named accounts sequentially in one process and respects each account's own worker lock and state paths
• add --forever to run-loop or run-accounts when you want the process to keep cycling until you stop it
• replay reports are written under var/replay/ and show up in scorecard once they exist for a strategy
• historical-bars replay uses real historical SPY bars from Alpaca, Black-Scholes option pricing, and a realized-volatility proxy for the VIX gate; it is stronger than the demo replay, but still not a quote-by-quote options-chain replay
• the live Alpaca paper path now uses the same realized-volatility proxy when direct VIX quotes are unavailable, so the entry filter stays active instead of silently degrading to a warning-only path
• promotion is an evidence gate for advancing strategy stages; it is not a profitability guarantee and it does not bypass manual live approval
• autonomous-runner honors var/runtime/AUTONOMOUS_HALTED, writes var/runtime/autonomous_status_latest.json, var/runtime/autonomous_status_history.jsonl, and var/runtime/autonomous_heartbeat.json, and has a service-wrapper example at infra/systemd/new-trading-system-autonomous.service.example
• autonomous-status reads those runtime artifacts, while autonomous-halt and autonomous-resume manage the halt file directly for operator control
• the current legacy iron condor runtime now allows up to 8 concurrent structures, giving the runtime room for a very near weekly expiry rung without degrading the constrained replay profile
• the live DTE entry window now spans 11-52 DTE, matching the enforced rule that new entries stay outside the 10-DTE forced-exit zone
• the live risk engine now merges broker-backed same-day order history into its intraday structure/fill guardrails, so repeated entries cannot bypass the policy just because the local ledger started fresh
• the live weekly ladder now permits up to 8 same-day structures, with cumulative-risk limits still providing the primary exposure cap
• the current iron condor minimum credit floor is now 0.40, which allows the near-expiry weekly rung without degrading the constrained replay profile
• the current iron condor take-profit target is now 2.5%, chosen because it preserved the clean constrained replay profile while shortening holding time further and bringing the live paper book closer to real exits
• the live DTE exit threshold is now 10 DTE, which left the constrained replay unchanged while allowing the near-expiry live condor to qualify for exit sooner
• live reconciliation now tolerates tiny option mark drift (for example a $1 market-value difference on a single contract) so harmless quote jitter does not mark the autonomous runtime degraded
• legacy trade evidence, when found at the sibling trading/data/trades.json, is shown separately as calibration/reference only and does not satisfy current-system readiness checks by itself

Verification note:

• .github/workflows/ci.yml now runs bash scripts/verify.sh on pushes and pull requests
• the verification script checks patch hygiene, compiles Python, parses schema JSON files, runs pytest -q, and exercises the internal paper worker plus reconcile and verify commands

Top-level layout:

• apps/: dashboard and operator-facing interfaces
• services/: platform runtime services
• packages/: reusable SDKs, schemas, and shared types
• strategies/: pluggable strategy families
• schemas/: versioned JSON contracts
• infra/: deployment and environment assets
• tests/: replay and integration verification harnesses

Documentation governance:

• AGENTS.md tells future LLM and engineering work to keep docs synchronized with the codebase.