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.