A strong answer. “My first question would be which of three quite different things they mean, because they get reported identically.

Missing. A gap, a sequence break, an outage. This is the easiest to check and the least common cause of a broken backtest — and it's visible: compare bar counts against the exchange's own end-of-day file.

Wrong. A bad print, a vendor error, a mis-mapped instrument. Real, and usually detectable by cross-checking two sources on the disputed day.

Correct but used wrongly. This is nearly always the actual answer and it's the one worth being good at.

The specific traps, in the order they occur: adjusted prices. If they used a split- and dividend-adjusted series, the adjustment was applied retroactively to the whole history — so the price they 'saw' on a date in 2019 embeds information from a corporate action in 2023. That is lookahead, and it is invisible because the series looks perfectly clean. For India this bites constantly with bonus issues and splits.

Timestamp semantics. Is this the exchange timestamp, the vendor's receipt timestamp, or our capture timestamp? They can differ by milliseconds intraday and by a whole bar at the boundary. And is a bar labelled with its open or its close? An off-by-one-bar error is the single most common source of a spectacular fake backtest.

Instrument identity over time. A symbol is not an identity. Companies rename, merge, get delisted and re-list; a futures contract's symbol rolls monthly. If they joined on symbol rather than on a stable instrument ID with validity dates, they have silently stitched two different instruments into one series.

Survivorship in the universe. If the universe list came from today, everything that failed is missing.

What I'd build so this stops recurring: a point-in-time API where every query takes an as-of timestamp and returns exactly what was knowable then — including the reference data, not just the prices. Once researchers cannot accidentally get the future, most of this class disappears. And a daily reconciliation against the exchange's official files with an alert, so I find the problem before a researcher does.”