A signal without provenance is just an anecdote with a timestamp

Every organisation over about thirty people runs on sentences that no longer have sources. 'Enterprise customers won't accept that.' 'Our users don't read documentation.' 'That segment churns.' Each of these was once an observation made by a specific person, in a specific conversation, on a specific date. By the time it circulates as common sense, all three of those facts have been stripped, and what remains cannot be checked, dated, or overturned.

The mechanism is not carelessness. It is compression. Summarising is how information travels through an organisation, and every summary discards the provenance first, because provenance is the part that feels like overhead. What survives is the interpretation — which is precisely the part most likely to be wrong.

The correction is small enough to be adopted: record the observation verbatim, separately from what you think it means. Quote the sentence. Note who heard it, through what channel, and when. Then, in a clearly separate field, write your interpretation. The separation is the whole intervention. Interpretations get revised as evidence accumulates; observations do not, and mixing them makes both unrevisable.

Provenance also exposes sampling, which is the failure mode nobody notices in qualitative work. Four signals that all arrived through the support inbox are not four signals about your product. They are one signal about the population that writes to support. You cannot see that unless the channel was recorded, and by the time the four have been clustered into a theme, the channel is gone.

This has become more urgent, not less, as synthesis got cheap. A machine will happily cluster three hundred conversations into six confident themes in under a minute. The themes will be plausible. Whether any of them can be traced back to a single sentence a real person actually said is now a choice you have to make deliberately, because nothing in the tooling will make it for you.