Feed Quality Validator

No. The checks run in the browser on the page itself, so the pasted content is not transmitted to a server and nothing is stored between visits. That design keeps proprietary or licensed data from leaving the machine, which matters because market data is usually subject to redistribution terms. It also means the tool works without an account and that reloading the page discards whatever was entered.

A consistency failure means the bar is internally impossible: a high below the open or close, a low above them, or a non-positive price. No market condition produces that, so it is always a pipeline defect, typically a column mapping error, a bad parse, or a merge that combined fields from different bars. An outlier flag is statistical and can be either an error or a genuine large move, so it needs a human judgement that a consistency failure does not.

A full-sample score is simple, deterministic, and needs no warm-up period, which suits a tool run on a pasted extract. The tradeoff is that it assumes a roughly stable level: a series with a strong trend will flag its own recent values as extreme simply because they sit far from the overall mean. For a trending series, running the check on returns rather than on price levels, or on shorter segments, gives a more meaningful result.

A lower threshold catches more genuine errors and produces more false positives, which costs review time. The sensitive setting suits a first pass over unfamiliar data where missing an error is worse than reviewing extra bars. The conservative settings suit a series known to be volatile, where flagging every large move buries the real problems. Running the same data at two thresholds and looking at what appears between them is often more informative than choosing one.

A calendar-day rule needs no exchange calendar, so it works on any market including instruments that trade continuously, and it does not go wrong when a calendar is out of date. The cost is precision: a multi-day holiday can exceed the threshold without any bar being missing, and a single missing bar in the middle of a normal week will not trigger it. Treat a gap flag as a prompt to check against the venue calendar rather than as a confirmed missing bar.

Most often an unadjusted corporate action, where a split or a large distribution produces a step change in an otherwise continuous series. Other frequent causes are a decimal or currency unit error, a stale bar repeated from the previous day, and a genuine market event. The distinguishing feature of an unadjusted split is that the ratio between the two closes is close to a simple fraction, which is worth checking before assuming the move was real.

No. These checks cover internal consistency, statistical outliers, obvious gaps, and large jumps. They cannot detect survivorship bias in the universe, silently restated values, a subtly wrong timestamp convention, prices adjusted with the wrong factor but still continuous, or volume reported on a different basis than expected. A clean result removes a class of obvious defects, which is a starting point for trusting a dataset rather than a conclusion.

The consistency, outlier, and large-return checks apply to any bar frequency, since none of them assumes a daily interval. The gap check is the exception: it reasons in calendar days, so it will not identify a missing five-minute bar. For intraday data, the practical approach is using the tool for the other three categories and checking interval continuity separately against the expected number of bars in a session.

The same four categories translate directly into a validation stage that runs before data is written to a store, with each check producing a severity rather than a pass or fail for the whole file. Consistency violations usually justify rejecting a file outright, while outliers and gaps are better routed to review with the affected rows quarantined. Recording the thresholds used alongside the result keeps a later reader from having to guess why a bar was accepted.