Strict mode is a migration, not a checkbox: MySQL truncation and the 1406 that follows

Nothing in the stack lies about this, which is why it survives review. MySQL treats the statement as complete, so affected_rows comes back 1, so a PDO connection in exception mode has nothing to throw, so the ORM reports a successful save, and an integration test asserting that the row exists is correct to pass ^[4]. Every layer honours its contract. None of those contracts says that the characters you sent are the characters that got stored.

The worked example drops two characters out of twelve ^[14], which is the awkward size for this class of bug: small enough to read as a formatting quirk, big enough to change what the value means.

The number worth computing during an audit is not how many rows sit at a column's limit, but what share of one input class does. In the case the author documents, the count at the limit was every row that had ever been saved with a country code ^[8], a 100 percent truncation rate for that input ^[17]. That is not an edge case leaking past validation. That is the normal write path for internationally formatted input, working as configured.

One trap while running the count: LENGTH() counts bytes and CHAR_LENGTH() counts characters, and VARCHAR(n) limits characters, so on multi-byte text a byte count will disagree with the limit the server actually enforces ^[9].

Strict mode does not narrow the gap between what the application sends and what the column accepts; it changes which side of the write reports the gap ^[11]. The second case in the source material is the one that should set the order of work. Strict mode was on and correct, the column was varchar(150), and the admin form carried neither a maxlength nor a server-side length check, so an editor who pasted a long headline got a blank error page and lost a story they believed they had saved ^[12]. The sql_mode setting decided where the loss landed. It did not remove it.

That puts the config change last, not first. Generate the limits out of information_schema and enforce them at the boundary, since a hand-copied constant drifts the first time somebody runs an ALTER ^[13]. Then change sql_mode, because in that order the user gets a validation message and in the other order they get an error page ^[16]. The audit and the flip are two different pieces of work, and only one of them can be done in an afternoon.