Define point-in-time stock universe foundation

[dvp2004]

Summary

• restore the concise project research story and frozen seven-strategy benchmark ladder in README.md
• define the minimum viable point-in-time stock-universe foundation in docs/point_in_time_stock_universe_plan.md
• establish the terminal gate that no model training begins until the universe contract passes and is frozen

Benchmark narrative

• SPY Buy & Hold remains the raw-return and terminal-wealth benchmark
• SPY 3D Overlay + deep-drawdown guard + loose relief remains the best execution-realistic risk-adjusted ETF candidate built so far
• the future stock model must attempt to beat SPY after comparable costs while also being evaluated against the overlay's drawdown and risk-adjusted profile

The README figures were checked against the frozen full-period table in docs/research_history.md; no historical strategy research was rerun and no result was changed.

Boundary

Design documentation only. This PR does not fetch data, train a model, run a backtest, create a portfolio, or add executable research behavior.

Changed files

1. README.md
2. docs/point_in_time_stock_universe_plan.md

Validation

• README benchmark figures reconciled to docs/research_history.md
• bounded Markdown structure/content check: passed
• git diff --check: passed
• path/credential/research-result safety scan: zero hits
• changed manifest: exactly two files

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ba940514d9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Remove the coverage-limitation bypass

If the coverage audit finds an unresolved gap, this wording permits training once someone accepts the limitation, directly contradicting the plan's fail-closed rules and terminal requirement that the contract actually pass before training. That exception can allow survivorship-biased or incomplete dates into the next experiment; require a passing, frozen contract here as well, or define a formal contract version that excludes the affected period and then passes all gates.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Fail qualification before filtering missing observations

When an active constituent lacks its cutoff close or volume, this predicate makes the security ineligible, while the later fail-closed rule only rejects missing prices for an already "eligible" security. An implementation following these definitions can therefore silently drop the missing-data constituent instead of failing the date, reintroducing data-availability or survivorship bias; define membership-qualified candidates first and make missing required observations an explicit qualification failure before applying liquidity and price filters.

Useful? React with 👍 / 👎.