How Later SEC Comparatives Leak Into Historical Backtests: A CoreWeave Case Study

This reproducible case study examines how later SEC comparative values can leak into an earlier historical backtest when a financial-data warehouse destructively retains only the latest value for each issuer, taxonomy concept, unit and fiscal period. The study matches 14 CoreWeave facts across two Form 10-Q accessions. At a decision time in August 2025, a naive latest-value query substitutes a comparative value accepted in August 2026 for every matched fact. Ten differences are compatible with reduced display precision, four are classified as presentation changes, one changes sign, and none is identified by the issuer as a formal restatement. Five deterministic calculations measure the downstream effect on reported revenue, other investing activities, gross margin, an operating-cash-flow-less-capital-expenditure proxy and stock-based-compensation intensity. The clearest example changes Other investing activities from positive USD 26.109 million in the first-reported filing to negative USD 6 million in the later comparative filing. The correct point-in-time query contract filters source vintages by SEC acceptance time before calculating a feature. Each accession-level observation remains immutable, missing values remain missing, and rounding-compatible differences are distinguished from presentation changes and issuer-identified restatements. This is a source-vintage contamination study, not a portfolio-return backtest or performance claim. It does not estimate alpha, information coefficient, return, Sharpe ratio or drawdown. AhaSignals is an independent research publisher. Company names and ticker symbols identify the relevant SEC filer only. AhaSignals is not affiliated with, endorsed by or sponsored by the issuer or the SEC. Research and educational use only; not investment, trading, legal, accounting or tax advice.