Automated Investing — code relocated to GitHub

The code + experiment writeups for this project now live in the canonical private repo, not in the vault. Moved 2026-05-29 to eliminate drift (founder directive: GitHub is canonical, one working copy).

Where it lives now

• Canonical (private): RayDataCo/automated-investing — main is the single source of truth.
• Working checkout (Mac Mini): ~/Projects/automated-investing/ — clone tracking the GitHub remote; where Ray runs/builds. Own .venv (uv, Python 3.12). PR-based workflow per [[feedback_pr_only_workflow]].
• What's in the repo: the autoinv event-driven backtest spine (engine, data, validation, metrics, portfolio, pricing) + Phase-1 additions (guard.py fail-closed no-lookahead guard, oracle.py shadow oracle, backtest.py, report.py, feed.py, strategies/momentum.py) + 58-test suite; early design docs (STRATEGY.md, architecture-vision.md, infrastructure-plan.md); experiments/ writeups + scripts/ analysis code (PEAD, math-foundations levels 1-5, prediction-market Brier/Kalshi/Polymarket — pre-pivot, archived); dead prediction-market modules in autoinv/_archive/.

Why relocated

Nested inside the vault repo with no own remote → true PR review wasn't possible and two local copies risked drifting. Extracted 2026-05-29 via git subtree split (full RED→GREEN test-first history preserved), pushed to the private repo, in-vault copy retired. Full relocation log: working-context 2026-05-29 12:06 ET entry.

Current state of the work

Phase-1 backtest spine built; no-lookahead guard hardened + independently verified (3 checks: incremental causal replay + multi-truncation + future-tail perturbation; fail-closed). Honest momentum result: SPY MA100 underperformed buy&hold (no edge — correctly framed). Phase 2+ (discovery loop, sizing/execution, vectorbt sweep) is a separate founder go.

Original thesis (kept for context)

Always-on AI agents for systematic, data-driven investing: data pipeline → signal → decision → execution → feedback → compound learning, on the same agentic infrastructure RDCO builds for everything else. P&L is the most honest feedback loop. NOT a day-trading bot; NOT real money before a tested system; live trading is founder-gated ([[feedback_paper_trade_deploy_authorization]]). Paper-trading only, founder's personal capital.

Reference material

• [[06-reference/2026-04-04-swing-trading-guide]] — Kevin Xu's swing trading approach (volume + catalysts)
• [[06-reference/2026-04-10-gemchange-quant-from-scratch]] — @gemchange_ltd's zero-to-quant curriculum (probability → stats → linalg → calc → stochastic). The math foundation this sits on.
• [[06-reference/2026-04-10-gemchange-simulate-like-quant-desk]] — @gemchange_ltd's prediction-market simulation blueprint (Monte Carlo → importance sampling → particle filters → copulas → agent-based → 5-layer stack).
• [[2026-05-29-strategy-pipeline-architecture-v0]] — the greenfield pipeline architecture (buy-vectorbt / build-discipline-shell)
• [[2026-05-29-markov-system-requirements-v0]] — Markov as one pluggable candidate, not the project
• [[project_investing_markov_capital_cycle]] — investing style + capital-cycle thesis