Vendor Reconciliation and Cross-Checks

Every market data vendor makes its own choices about how to handle corporate actions, how to timestamp trades, which exchanges to prioritize when consolidating a national best bid and offer, and how to treat corrections issued after the fact. Two reputable vendors covering the exact same stock on the exact same day can disagree — sometimes by a fraction of a cent from rounding conventions, sometimes by a meaningful amount because one has already applied a dividend adjustment and the other hasn't yet. Neither vendor is necessarily "wrong"; they're often just making different, defensible choices, or one has simply made a processing error that the other happened not to make. A research pipeline that only ever looks at one vendor's numbers has no way of telling the difference between "this is just how the data is" and "this specific number is broken."

What reconciliation actually looks like

In practice, this means periodically pulling the same field — closing price, dividend amount, shares outstanding — from two independent sources and comparing them, flagging any pair that differs by more than a small tolerance for manual review. The tolerance matters: too tight, and routine, harmless differences (rounding, slightly different adjustment timing) generate constant false alarms that get ignored; too loose, and genuine errors slip through unnoticed. Most operations tune the tolerance per field, since a one-cent difference in a $5 stock's close is a much bigger deal proportionally than the same one cent difference in a $500 stock.

A concrete case

A quant desk maintains both a primary and a secondary price vendor for its equity universe. One week, the secondary vendor reports a stock's split ratio as 2-for-1 while the primary reports 3-for-2. Left unreconciled, any backtest using the secondary vendor's history would show a price series with a spurious jump around the split date, and any live position sizing that referenced share counts would be wrong by a large factor for that name. Because the desk cross-checks every corporate action across both vendors before applying it, the discrepancy gets caught and resolved — usually by checking the exchange's own announcement — before either the research pipeline or the trading system uses the bad number.

What this means in practice

Vendor reconciliation is unglamorous, ongoing work rather than a one-time setup task, because vendors change their processing logic and make new mistakes continuously. Treating a single vendor's feed as ground truth, without ever cross-checking it against an independent source, means data errors are only ever discovered when they cause a visible problem downstream — a bad backtest result, a mispriced position — which is a far more expensive way to find them than a routine automated comparison.