[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-82175":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":22,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":37,"readmeContent":38,"aiSummary":39,"trendingCount":16,"starSnapshotCount":16,"syncStatus":40,"lastSyncTime":41,"discoverSource":42},82175,"alpha-forge","Liu-Ming-Yu\u002Falpha-forge","Liu-Ming-Yu","Alpha Forge — an agentic AI operating system for systematic trading.","",null,"Python",163,18,12,5,0,80,135,15,73.84,"Apache License 2.0",false,"main",true,[26,27,28,29,30,31,32,33,34,35,36],"algorithmic-trading","alpha-research","backtesting","financial-ml","ibkr","machine-learning","portfolio-construction","python","quantitative-finance","risk-management","trading-systems","2026-06-12 04:01:37","\u003Cdiv align=\"center\">\n\n\u003Cp>\n  \u003Cstrong>Language\u003C\u002Fstrong>\u003Cbr>\n  \u003Ckbd>English · Current\u003C\u002Fkbd>\n  \u003Ca href=\"README.zh-CN.md\">\u003Ckbd>简体中文\u003C\u002Fkbd>\u003C\u002Fa>\n\u003C\u002Fp>\n\n# Alpha Forge\n\n### Introducing Alpha Forge — an agentic AI operating system for systematic trading.\n\n\u003Cp>\n  Built for a world where markets move through price, language, liquidity,\n  positioning, macro pressure, and machine-readable narrative, Alpha Forge\n  unifies research, LLM intelligence, machine learning, governance, and execution\n  into one production architecture that works with IBKR.\n\u003C\u002Fp>\n\n\u003Cp>\n  It turns filings, earnings calls, news, market data, microstructure,\n  ownership, estimates, options, and macro signals into governed alpha\n  candidates. Then it tests them, challenges them, versions them, audits them,\n  and blocks them until evidence is strong enough.\n\u003C\u002Fp>\n\n\u003Cp>\n  \u003Csub>Agentic LLM Intelligence \u002F Representation Learning \u002F Autonomous Research \u002F Governed Execution\u003C\u002Fsub>\n\u003C\u002Fp>\n\n\u003Cp>\n  \u003Ca href=\"https:\u002F\u002Fwww.interactivebrokers.com\u002Fen\u002Ftrading\u002Fib-api.php?menu=A\" aria-label=\"Open the official Interactive Brokers API page\">\n    \u003Cimg src=\"docs\u002Fassets\u002Fibkr-integration-badge.svg\" alt=\"IBKR-ready execution: TWS API, IB Gateway, and governed paper-to-live pathway\" width=\"520\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp>\n  \u003Ca href=\"#command-deck\">\u003Ckbd>Command Deck\u003C\u002Fkbd>\u003C\u002Fa>\n  \u003Ca href=\"#architecture-map\">\u003Ckbd>Architecture Map\u003C\u002Fkbd>\u003C\u002Fa>\n  \u003Ca href=\"#agentic-llm-intelligence-layer\">\u003Ckbd>LLM Layer\u003C\u002Fkbd>\u003C\u002Fa>\n  \u003Ca href=\"#autonomous-research-factory\">\u003Ckbd>Research Factory\u003C\u002Fkbd>\u003C\u002Fa>\n  \u003Ca href=\"#production-execution-fortress\">\u003Ckbd>Execution\u003C\u002Fkbd>\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003C\u002Fdiv>\n\n---\n\n## Command Deck\n\n\u003Ctable>\n  \u003Ctr>\n    \u003Ctd width=\"33%\">\n      \u003Cstrong>Discover\u003C\u002Fstrong>\u003Cbr>\n      \u003Csub>Explore the product surface.\u003C\u002Fsub>\u003Cbr>\u003Cbr>\n      \u003Ca href=\"#showcase\">Showcase\u003C\u002Fa>\u003Cbr>\n      \u003Ca href=\"#architecture-map\">Architecture Map\u003C\u002Fa>\u003Cbr>\n      \u003Ca href=\"#technology-stack\">Technology Stack\u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd width=\"33%\">\n      \u003Cstrong>Investigate\u003C\u002Fstrong>\u003Cbr>\n      \u003Csub>Open the core intelligence systems.\u003C\u002Fsub>\u003Cbr>\u003Cbr>\n      \u003Ca href=\"#agentic-llm-intelligence-layer\">Agentic LLM Layer\u003C\u002Fa>\u003Cbr>\n      \u003Ca href=\"#text-event-v2\">Text-Event-v2\u003C\u002Fa>\u003Cbr>\n      \u003Ca href=\"#hybrid-ml--representation-learning-engine\">Hybrid ML Engine\u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd width=\"33%\">\n      \u003Cstrong>Operate\u003C\u002Fstrong>\u003Cbr>\n      \u003Csub>Understand deployment and safety.\u003C\u002Fsub>\u003Cbr>\u003Cbr>\n      \u003Ca href=\"#v2-shared-account-orchestrator\">V2 Orchestrator\u003C\u002Fa>\u003Cbr>\n      \u003Ca href=\"#production-execution-fortress\">Execution Fortress\u003C\u002Fa>\u003Cbr>\n      \u003Ca href=\"#governance--observability-fabric\">Governance Fabric\u003C\u002Fa>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Mission Control\u003C\u002Fstrong> — Alpha Forge at a glance\u003C\u002Fsummary>\n\n| Mission | System Response |\n| --- | --- |\n| Convert language into alpha | Agentic LLM layer turns narrative into text-event-v2 features |\n| Prevent uncontrolled AI behavior | Manifests, evals, startup assertions, audits, and human-review gates |\n| Search across feature space | Campaigns, walk-forward validation, IC diagnostics, and bootstrap evidence |\n| Merge many strategies safely | V2 account-level orchestrator resolves proposals into one portfolio target |\n| Block unsafe execution | Fail-closed broker path, reconciliation, kill switches, and durable journals |\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Open Alpha Lifecycle\u003C\u002Fstrong> — From raw signal to controlled live deployment\u003C\u002Fsummary>\n\n```text\nLanguage + Market Data\n  -> Agentic Intelligence\n  -> Feature Families\n  -> Campaign Research\n  -> Walk-Forward Validation\n  -> Evidence Package\n  -> Production-Candidate Gate\n  -> Shadow Mode\n  -> Paper Trading\n  -> Paper Soak\n  -> Live Preflight\n  -> Controlled Live\n```\n\n\u003C\u002Fdetails>\n\n## Showcase\n\n| System | Signature Capability |\n| --- | --- |\n| Agentic LLM Intelligence Layer | Multi-agent text-event-v2 alpha generation with tool use, manifests, traceable reasoning, evals, live assertions, and human-review gates |\n| Hybrid ML + Representation Learning Engine | XGBoost, learned embeddings, PCA representations, formulaic alpha mining, microstructure, ownership, estimates, options, macro, and text-event intelligence |\n| Autonomous Research Factory | Campaign-driven alpha discovery, walk-forward validation, IC diagnostics, bootstrap evidence, regime testing, and production-candidate packaging |\n| V2 Shared-Account Orchestrator | Multi-engine portfolio target fusion with governed Shadow → Paper → Live promotion |\n| Production Execution Fortress | Fail-closed broker execution, durable state, reconciliation, kill switches, strict service boundaries, and automated verification |\n| Governance & Observability Fabric | Manifest registry, startup assertions, model provenance, event journals, readiness reports, promotion gates, and operator visibility |\n\n## Architecture Map\n\n![Alpha Forge architecture poster](docs\u002Fassets\u002Falpha-forge-architecture.svg)\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Explore Architecture Layers\u003C\u002Fstrong> — tap into the stack\u003C\u002Fsummary>\n\n| Layer | What It Controls | Why It Matters |\n| --- | --- | --- |\n| Language Intelligence | Agents, text-event-v2 extraction, manifests, evals, assertions | Converts narrative into governed alpha inputs |\n| Research Graph | Feature families, campaigns, walk-forward validation, diagnostics | Turns hypotheses into reproducible evidence |\n| Portfolio Fusion | Multi-engine proposals, account budgets, risk overlays | Prevents strategy modules from fighting over the same account |\n| Execution Safety | Pre-trade checks, broker adapters, kill switches, reconciliation | Stops unsafe action before it reaches the broker |\n| Governance Fabric | Readiness reports, promotion gates, event journals, operator visibility | Makes AI-native trading auditable and controllable |\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Open Service Boundary Console\u003C\u002Fstrong> — how the monolith stays disciplined\u003C\u002Fsummary>\n\n| Boundary | Rule |\n| --- | --- |\n| `core` | Owns contracts, domain models, and events |\n| `application` | Owns use cases and read models |\n| `services` | Own domain logic behind service boundaries |\n| `infrastructure` | Owns durable adapters and external systems |\n| `bootstrap` | Composes concrete runtime dependencies |\n| `cli` \u002F `views` | Stay thin and operator-facing |\n\n\u003C\u002Fdetails>\n\n---\n\n## Agentic LLM Intelligence Layer\n\n### From language to governed alpha.\n\nThe platform treats language as a first-class market data substrate.\n\nFilings, earnings calls, macro commentary, news, transcripts, social narratives,\ncentral-bank language, analyst revisions, and event-driven text are converted\ninto structured text-event-v2 features. These features are not uncontrolled LLM\noutputs. They are manifest-bound, versioned, evaluated, audited, and\npromotion-gated before they can influence a portfolio.\n\nAt the center is a governed multi-agent intelligence layer: a swarm of\nspecialized AI agents that reason over market context, challenge hypotheses,\ninspect risks, and prepare candidate signals for quantitative validation.\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Agent Console\u003C\u002Fstrong> — specialized intelligence roles\u003C\u002Fsummary>\n\n| Agent | Role |\n| --- | --- |\n| Market Oracle | Converts market events, regime changes, and narrative shifts into structured research hypotheses |\n| Sentiment Synthesizer | Extracts sentiment, uncertainty, surprise, and directional pressure from noisy language streams |\n| Strategy Debater | Runs adversarial critique against candidate alpha ideas before they reach formal research |\n| Risk Guardian | Detects overfitting, regime fragility, crowding, stale data, and hidden exposure |\n| Evidence Auditor | Checks whether a signal has enough statistical and operational support for promotion |\n| Execution Examiner | Reviews whether a strategy can survive cash, liquidity, broker, and order-routing constraints |\n\n\u003C\u002Fdetails>\n\nThis is not “prompt engineering.”\n\nIt is governed machine reasoning for alpha research: agent outputs are\ntraceable, reproducible, inspected, and connected to downstream evidence.\n\n## Text-Event-v2\n\n### LLM-native alpha, engineered for production.\n\nThe text-event-v2 family turns unstructured language into quantitative features\nthat can be ranked, backtested, stress-tested, and governed like any other alpha\nsource.\n\nIt supports:\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Text-Event Capability Matrix\u003C\u002Fstrong> — LLM-native features under governance\u003C\u002Fsummary>\n\n| Capability | Description |\n| --- | --- |\n| Event Extraction | Converts raw text into structured market events, catalysts, risks, and directional hypotheses |\n| Narrative Regime Detection | Tracks shifts in market language, macro tone, risk appetite, and sector-level narratives |\n| Filing Intelligence | Extracts management tone, uncertainty, litigation risk, guidance changes, and operating pressure |\n| Earnings Call Reasoning | Identifies surprise, confidence, hedging language, and forward-looking signal changes |\n| Manifest-Governed Features | Every LLM-derived feature is tied to a schema, prompt contract, model version, and audit trail |\n| Live Startup Assertions | Production startup fails if text models, manifests, or required feature contracts are missing or inconsistent |\n| Promotion Readiness | Text-derived signals must survive the same evidence pipeline as price, fundamental, or microstructure features |\n\n\u003C\u002Fdetails>\n\nThe result is a language intelligence layer that behaves less like a chatbot\nand more like a governed research instrument.\n\n---\n\n## Hybrid ML + Representation Learning Engine\n\n### Classical alpha, modern ML, and learned structure — in one research graph.\n\nThe research engine combines traditional quantitative signals with machine\nlearning and learned representations across the full feature universe.\n\nIt integrates:\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Feature Universe\u003C\u002Fstrong> — explicit domain knowledge plus learned structure\u003C\u002Fsummary>\n\n| Feature \u002F Model Family | Purpose |\n| --- | --- |\n| Price & Momentum Features | Trend, reversal, volatility, distance-to-high, and cross-sectional ranking signals |\n| Microstructure-v3 | Liquidity, spread, volume pressure, intraday behavior, and trading-friction-aware features |\n| Ownership-v1 | Institutional behavior, positioning shifts, insider activity, and ownership structure |\n| Estimates-v1 | Analyst expectations, revisions, dispersion, and forward-looking earnings pressure |\n| Options-v1 | Volatility surface, skew, flow, and implied market expectation signals |\n| Macro-v1 | Rate, inflation, liquidity, sector, and broad regime context |\n| Formulaic Alpha Mining | Evolutionary search over WorldQuant-style grammars with auto-promotion constraints |\n| Learned Representations-v1 | PCA-style embeddings and compressed cross-family structure from existing features |\n| Text-Event-v2 | LLM-derived structured features from filings, calls, news, and narrative sources |\n| XGBoost Ranking Pipelines | Nonlinear interaction discovery, cross-sectional scoring, and feature attribution |\n\n\u003C\u002Fdetails>\n\nThis creates a research system that can discover alpha from both explicit domain\nknowledge and latent structure hidden across feature families.\n\nThe platform does not rely on one model, one signal, or one narrative. It builds\nan ensemble of evidence.\n\n---\n\n## Autonomous Research Factory\n\n### Alpha discovery with memory, discipline, and promotion control.\n\nQuant Platform is designed as an autonomous research factory: generate\nhypotheses, build features, run experiments, evaluate results, preserve\nartifacts, and promote only what survives.\n\nThe research loop is structured:\n\n```text\nHypothesis -> Feature Family -> Campaign -> Walk-Forward Validation -> Diagnostics -> Evidence Package -> Promotion Decision\n```\n\nEach experiment produces reproducible artifacts: feature manifests, model\nconfigs, fold results, IC diagnostics, bootstrap checks, turnover profiles,\ndrawdown curves, and production-candidate reports.\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Research Diagnostics\u003C\u002Fstrong> — how attractive illusions are rejected\u003C\u002Fsummary>\n\n| Research Capability | Description |\n| --- | --- |\n| Campaign-Based Research | Experiments are organized as campaigns instead of ad hoc notebooks |\n| Walk-Forward Validation | Strategies are tested across rolling out-of-sample folds |\n| IC Diagnostics | Signal quality is measured through rank correlation, stability, and negative-streak behavior |\n| Bootstrap Evidence | Statistical uncertainty is measured before promotion |\n| Turnover Analysis | Execution drag and capacity pressure are treated as first-class constraints |\n| Regime Testing | Performance is analyzed across risk-on, risk-off, volatile, and low-liquidity periods |\n| Production-Candidate Reports | Strategies are packaged with the evidence needed for governed deployment |\n\n\u003C\u002Fdetails>\n\nThe system is built to reject attractive illusions: high-return backtests,\nfragile edges, accidental leakage, unstable IC, and strategies that look good\nonly before costs.\n\n---\n\n## V2 Shared-Account Orchestrator\n\n### Many engines. One account. One coherent portfolio.\n\nReal trading is not a collection of independent strategy demos. Multiple\nengines compete for the same cash, risk budget, exposure limits, and broker\nconnection.\n\nThe V2 Shared-Account Orchestrator solves this by merging independent strategy\nproposals into a single account-level portfolio target.\n\nStrategy engines can include:\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Engine Rack\u003C\u002Fstrong> — proposal sources for one account-level target\u003C\u002Fsummary>\n\n| Engine | Role |\n| --- | --- |\n| Cross-Sectional Equity Ranker | Selects and weights securities based on alpha scores |\n| ETF Macro Allocator | Adjusts broad exposure based on macro and regime signals |\n| Risk Overlay Engine | Reduces exposure when drawdown, volatility, or signal quality deteriorates |\n| Text-Event Engine | Incorporates LLM-derived event signals into portfolio intent |\n| Future Strategy Modules | Additional engines can submit proposals through the same governed interface |\n\n\u003C\u002Fdetails>\n\nThe orchestrator applies cash constraints, exposure limits, stale-state checks,\nthrottles, kill switches, reconciliation, and broker-readiness checks before any\ntarget becomes executable.\n\nDeployment is staged:\n\n```text\nShadow Mode -> Paper Trading -> Paper Soak -> Live Preflight -> Controlled Live\n```\n\nNothing becomes live by accident.\nNothing bypasses governance.\nNothing trades without a validated account-level view.\n\n---\n\n## Production Execution Fortress\n\n### Designed to stop before it fails.\n\nThe execution layer is built around a conservative principle:\n\nIf the system cannot prove it is safe to act, it does not act.\n\nOrders pass through a fail-closed execution path with broker abstraction,\npre-trade validation, state reconciliation, event journaling, and\npromotion-state checks.\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Safety Matrix\u003C\u002Fstrong> — proof before action\u003C\u002Fsummary>\n\n| Safeguard | Purpose |\n| --- | --- |\n| Fail-Closed Execution | Blocks action when state is stale, invalid, incomplete, or unverifiable |\n| Durable Portfolio State | Persists snapshots, parent-child provenance, and account state across restarts |\n| Event Journal | Records execution events, promotion events, and operational state transitions |\n| Broker Reconciliation | Compares intended portfolio state with broker\u002Faccount reality |\n| Kill Switches | Stops trading when operational or risk conditions become unsafe |\n| Order Throttling | Prevents runaway order submission and broker abuse |\n| Paper\u002FLive Separation | Keeps simulation, paper, and live paths explicitly separated |\n| IBKR Gateway\u002FTWS Integration | Supports broker-backed paper and live execution through controlled adapters |\n\n\u003C\u002Fdetails>\n\nThe system is not optimized to appear active.\nIt is optimized to be correct, observable, and controlled.\n\n---\n\n## Governance & Observability Fabric\n\n### The control plane for AI-native trading.\n\nModern AI systems are expected to be observable, testable, and governable,\nespecially when agents use tools and interact with external systems. OpenAI’s\n[Agents](https:\u002F\u002Fplatform.openai.com\u002Fdocs\u002Fguides\u002Fagents),\n[Agents SDK tracing](https:\u002F\u002Fopenai.github.io\u002Fopenai-agents-python\u002Ftracing\u002F),\nand [human-in-the-loop](https:\u002F\u002Fopenai.github.io\u002Fopenai-agents-python\u002Fhuman_in_the_loop\u002F)\ndocumentation directly emphasize tools, integrations, observability, guardrails,\nand human review as core production concepts.\n\nQuant Platform applies those ideas to systematic trading.\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Governance Console\u003C\u002Fstrong> — the control plane for AI-native trading\u003C\u002Fsummary>\n\n| Governance Layer | Function |\n| --- | --- |\n| Feature Registry | Tracks all feature families and their schemas |\n| Model Manifest System | Records model versions, feature contracts, prompts, configs, and artifacts |\n| Startup Assertions | Prevents production startup when required contracts are missing |\n| Readiness Reports | Summarizes whether a strategy is eligible for paper or live progression |\n| Promotion Gates | Blocks strategies that fail evidence, risk, or operational requirements |\n| Traceable Agent Actions | Makes LLM\u002Fagent behavior auditable instead of opaque |\n| Human Review Hooks | Allows sensitive transitions to require explicit approval |\n| Operator API | Provides controlled visibility into system state and readiness |\n\n\u003C\u002Fdetails>\n\nThe platform does not trust AI output because it sounds intelligent.\nIt trusts only what can be measured, replayed, audited, and governed.\n\n---\n\n## Technology Stack\n\n\u003Cdetails open>\n\u003Csummary>\u003Cstrong>Open Stack Matrix\u003C\u002Fstrong> — production components behind Alpha Forge\u003C\u002Fsummary>\n\n| Area | Stack |\n| --- | --- |\n| Language & Tooling | Python 3.11, uv, pyproject.toml, pre-commit, ruff, mypy, pytest |\n| ML \u002F Representation Learning | XGBoost, learned representations, PCA embeddings, feature attribution, rank modeling |\n| LLM \u002F Agentic AI | Text-event-v2 pipelines, manifest-governed agents, tool-using research workflows, audit trails |\n| Data & Research Storage | Parquet, object-store-ready artifacts, campaign outputs, reproducible experiment persistence |\n| State & Infrastructure | PostgreSQL, Alembic, Redis\u002Fevent-bus patterns, durable event journals |\n| Execution | Simulated backend, paper backend, IBKR ibapi, controlled live-adapter pathway |\n| Deployment | Docker, docker-compose, Makefile orchestration, FastAPI operator API |\n| Verification | make verify, import-boundary checks, module-size guards, type-debt enforcement, regression tests |\n| Governance | Readiness reports, production-candidate gates, startup assertions, promotion evidence, paper-soak validation |\n\n\u003C\u002Fdetails>\n\n---\n\n## Operating Surface\n\nThe installed console script and module entrypoint are equivalent:\n\n```bash\nquant-platform --help\npython -m quant_platform --help\n```\n\n## Setup\n\nUse Python 3.11 for project verification.\n\n```bash\npython -m venv .venv\n.venv\\Scripts\\activate\npython -m pip install --upgrade pip\npython -m pip install -e \".[dev,api]\"\n```\n\nOn macOS\u002FLinux\u002FWSL:\n\n```bash\npython3.11 -m venv .venv\nsource .venv\u002Fbin\u002Factivate\npython -m pip install --upgrade pip\npython -m pip install -e \".[dev,api]\"\n```\n\nOptional extras:\n\n```bash\npython -m pip install -e \".[ml]\"       # XGBoost research\npython -m pip install -e \".[backtest]\" # vectorbt backtests\n```\n\n`ibapi` is not a default dependency. Install it from the IBKR TWS API\ndistribution when using real IBKR paper\u002Flive paths.\n\n## Configuration\n\nSettings are loaded by `PlatformSettings` from `.env` and environment variables\nwith the `QP__` prefix.\n\nStart from:\n\n```bash\ncopy infra\\config\\settings.example.env .env\n```\n\nMinimal local in-memory development:\n\n```bash\nset QP__STORAGE__POSTGRES_DSN=\nset QP__STORAGE__REDIS_URL=\nset QP__STORAGE__EVENT_BUS_BACKEND=in_memory\nset QP__BROKER__PAPER_TRADING=true\n```\n\nDurable paper\u002Flive requires at least:\n\n```bash\nQP__STORAGE__POSTGRES_DSN=postgresql+psycopg:\u002F\u002Fuser:pass@host:5432\u002Fquant_platform\nQP__STORAGE__REDIS_URL=redis:\u002F\u002Flocalhost:6379\u002F0\nQP__STORAGE__EVENT_BUS_BACKEND=redis_streams\nQP__API__OPERATOR_API_KEY=\u003Cstrong random key>\n```\n\nCommon IBKR ports:\n\n| App | Mode | Port |\n| --- | --- | ---: |\n| TWS | Paper | `7497` |\n| TWS | Live | `7496` |\n| IB Gateway | Paper | `4002` |\n| IB Gateway | Live | `4001` |\n\n## Common Commands\n\nSchema:\n\n```bash\npython -m quant_platform migrations-check\npython -m quant_platform migrate\npython -m quant_platform verify-schema\n```\n\nSingle paper cycle:\n\n```bash\npython -m quant_platform run-cycle --initial-cash 50000\n```\n\nBounded engine runs:\n\n```bash\npython -m quant_platform run-engine --mode shadow --cycles 5\npython -m quant_platform run-engine --mode paper --execution-backend ib-paper --contracts-file infra\u002Fconfig\u002Fpaper_contracts.json --cycles 1\npython -m quant_platform run-engine --mode live --contracts-file .\u002Fcontracts.json --cycles 1\n```\n\nV2 multi-engine proposal\u002Forchestration path:\n\n```bash\npython -m quant_platform run-multi-engine ^\n  --engines cross_sectional_equity,etf_macro_allocator ^\n  --budgets-file .\u002Fbudgets.json ^\n  --mode paper ^\n  --contracts-file infra\u002Fconfig\u002Fpaper_contracts.json ^\n  --cycles 1\n```\n\nData and research:\n\n```bash\npython -m quant_platform ingest --start YYYY-MM-DD --end YYYY-MM-DD --contracts-file .\u002Fcontracts.json\npython -m quant_platform maintain --interval 900 --contracts-file .\u002Fcontracts.json\npython -m quant_platform compute-features --contracts-file .\u002Fcontracts.json\npython -m quant_platform features backfill --contracts-file .\u002Fcontracts.json --start YYYY-MM-DDT00:00:00+00:00 --end YYYY-MM-DDT00:00:00+00:00 --feature-set-version paper-alpha-composite-v1 --date-policy nyse-sessions\npython -m quant_platform boosting gpu-check\npython -m quant_platform research-campaign run --help\n```\n\nOperator API:\n\n```bash\npython -c \"import secrets; print(secrets.token_urlsafe(32))\"\nset QP__API__OPERATOR_API_KEY=\u003Cgenerated key>\npython -m quant_platform serve-api --host 127.0.0.1 --port 8000\ncurl -H \"X-API-Key: %QP__API__OPERATOR_API_KEY%\" http:\u002F\u002F127.0.0.1:8000\u002Fhealth\u002Fready\n```\n\n## Verification\n\nFast local checks:\n\n```bash\npython scripts\u002Fcheck_import_boundaries.py\npython scripts\u002Fcheck_service_coupling.py\npython scripts\u002Fcheck_module_size.py\npython scripts\u002Fcheck_type_debt.py\npython scripts\u002Fcheck_lint_debt.py --skip-ruff-probe\npython -m pytest -q tests\u002Funit\u002Ftest_engine_loop.py\n```\n\nFull offline gate:\n\n```bash\nmake verify\n```\n\nDurable and live gates are opt-in:\n\n```bash\nset QP_VERIFY_DURABLE=1\nset QP_VERIFY_LIVE_IBKR=1\nset IBAPI_PACKAGE_PATH=\u003Cpath-to-TWS-API\u002Fsource\u002Fpythonclient>\nmake verify\n```\n\n## Documentation Map\n\nStart with [USEME.md](USEME.md) for operator commands and\n[CONTEXT.md](CONTEXT.md) for project vocabulary.\n\nArchitecture:\n\n- [System overview](docs\u002Farchitecture\u002Fsystem-overview.md)\n- [Service boundaries](docs\u002Farchitecture\u002Fservice-boundaries.md)\n- [Production roadmap](docs\u002Farchitecture\u002Fproduction-roadmap.md)\n- [Risk register](docs\u002Farchitecture\u002Frisk-register.md)\n- [Source audit checklist](docs\u002Farchitecture\u002Fsource-audit-checklist.md)\n- [V2 execution flow](docs\u002Farchitecture\u002Fv2-execution-flow.md)\n- [Core contracts](docs\u002Finterfaces\u002Fcore-contracts.md)\n\nRunbooks live under [docs\u002Frunbooks](docs\u002Frunbooks\u002F). Use them for operations,\nincident response, paper promotion, backups, IBKR recovery, data recovery, and\nproduction readiness.\n","Alpha Forge 是一个面向系统化交易的代理型人工智能操作系统。它通过整合研究、大语言模型智能、机器学习、治理和执行等功能，利用IBKR平台实现从文件、财报电话会议、新闻、市场数据等多源信息中提炼出具有潜在价值的投资信号，并对其进行测试、挑战、版本管理和审计，直至证据充分可靠。项目采用Python编写，支持算法交易、金融机器学习、投资组合构建及风险管理等场景，适用于需要将复杂市场信息转化为可操作投资策略的专业投资者或机构。",2,"2026-06-11 04:07:57","CREATED_QUERY"]