Can I just pick the format that parses the most rows?

No, because the minority rows parse too, and picking the majority format files them under the wrong months.

Where do mixed dialects usually come from?

From an export that was appended to, assembled from two systems, or edited by hand after it was produced.

What should the page say?

Which format was proved, from which rows, and that the proof was applied per row rather than once over the file.

Failure catalogue · The file lies · 11 of 51

OBSERVED FAILURE MODE

Two date formats in one column.

One column, two dialects: ISO values and US values interleaved, usually because an export was appended to or assembled from two systems. Every row parses, and the months mix silently. A file-level format proof makes it worse, since the format it proves is right for one part of the column and wrong for the rest.

What we saw

A subscription export at Meridian Journal Group carried ISO dates for rows written by one system and US dates for rows appended by another, in a single column, with no marker separating them. Every row parsed. The months mixed, and a monthly chart drawn from the column was smooth enough that nobody questioned it. A format proof run once over the whole file would have confirmed a format correct for most of the rows and wrong for the rest, which is worse than no proof at all, because it produces a disclosure that reads as evidence.

Why it passes a glance

Nothing fails. A sample of rows read from the top of the file all agree with each other, because the appended rows sit at the bottom. Totals tie, since a value is unaffected by which month it was filed under. A page that discloses a proven date format looks more trustworthy than one saying nothing, and the disclosure is true of the rows it was proved on.

What addresses it

Doctrine principle 3, refuse to guess, applies the format proof per row rather than per file, so a column that changes dialect mid-stream is an error to raise, not a parse to accept. The hostile-exports skill supplies the wider habit this belongs to: silent drift in units, sign, timezone, and encoding is found by checking each of them, never by trusting that a vendor's export is internally uniform.

Check your own file in two minutes

Group your date column by its written pattern, counting how many values use each separator and digit layout.
If more than one pattern appears, split the rows by pattern and prove the format for each group separately.
Check whether a pattern clusters at the end of the file, which usually means rows were appended from a second system.

What this does not catch

Pattern grouping finds dialects that differ in shape. Two dialects that look identical, such as two systems both writing slash-separated dates in opposite orders, are separable only by rows whose day part exceeds twelve, and a group holding none of those stays unproven.

Quick answers

Can I just pick the format that parses the most rows?: No, because the minority rows parse too, and picking the majority format files them under the wrong months.
Where do mixed dialects usually come from?: From an export that was appended to, assembled from two systems, or edited by hand after it was produced.
What should the page say?: Which format was proved, from which rows, and that the proof was applied per row rather than once over the file.

Nearby failures

The date that parses two valid waysThe file lies The encoding that crashes the pipelineThe file lies The filename that lies about the periodThe file lies

Last updated 2026-09-02 · Dylan, founder · one of 51 observed failure modes, every one seen in a real build or in our own audits, none invented.

Get the free skill → The doctrine All 51 failure modes