Market-data integrity: stale, spoofed, and dropped

Every automated decision rests on one assumption: that the market data feeding it is current, complete, and honest. When that assumption breaks quietly, a strategy keeps trading with total confidence against a picture of the market that no longer exists. Market-data integrity is the unglamorous foundation under every signal, and the failures are rarely loud, they are subtle, and they cost money before anyone notices.

§01Staleness is the quiet failure

The most dangerous data problem is not a feed that stops, that is obvious, but a feed that slows. Prices that are a few seconds old look perfectly valid and drive orders into a market that has already moved. Every consumer of market data should timestamp on arrival and treat data older than a tight threshold as untrustworthy, degrading to a safe state rather than acting on a stale book. If you cannot prove the data is fresh, you must assume it is not.

§02Gaps and sequence integrity

Real feeds drop packets, reorder messages, and occasionally hand you a book with a hole in it. Robust systems track sequence numbers, detect gaps, and refuse to trade on a book they know is incomplete, requesting a snapshot and rebuilding rather than guessing. A single missed delete on the order book can leave a phantom level that a strategy happily trades against. Silence is not the same as no change, and a well-built consumer never conflates the two.

§03Manipulation and sanity checks

Not all bad data is accidental. Spoofing, layering, and momentary dislocations can present a distorted book designed to provoke a reaction. While policing manipulation is the venue’s job, a defensive system still sanity-checks what it receives: prices that cross, sizes that are implausible, or moves too large to be real should trip a guard rather than a trade. Cross-referencing more than one source, where feasible, turns a single manipulated feed from a trap into an anomaly.

§04Trust, but verify the feed

Treating market data as inherently trustworthy is the mistake. Timestamp it, sequence it, sanity-check it, and fail safe when it looks wrong, the same defensive posture you would apply to any untrusted input. This advisory is educational and illustrative and is not investment or security advice for any specific system.