Taiwan Point-in-Time Backtest Data — snapshot_date on every row
Backtesting on restated data silently uses information not knowable on signal day — look-ahead bias, and it inflates results. This bundle stamps snapshot_date on every row, ready to feed straight into a backtest engine.
Why look-ahead bias kills backtests
A company announces EPS 3.2 on Aug 10, restated to 3.8 on Nov 12. A conventional source overwrites in place, so a backtest querying Sep 1 gets 3.8 — but the market saw 3.2. Your strategy silently used future data.
What we deliberately do not do
We do not ship strategies, signals or parameter suggestions, and we do not claim this bundle makes backtests perform better — only that it makes them trustworthy. Once look-ahead bias is removed, many pretty equity curves get uglier. That is the point.
We also do not retroactively patch history: when an official figure was wrong at the time and corrected later, we keep the as-published version and store the restatement separately. A backtest needs the world as it was visible then, not the truth we learned afterwards.