The Records You Stop Trusting

Every long record eventually contains a stretch the owner does not quite believe. Handling it explicitly is better than the usual approach, which is to quietly avoid analyses that touch it.

A concrete way to test the method is outlined in the detailed explainer. Use a short pilot and keep only the fields that still make sense when comparing the record months later.

How a period becomes untrusted

Heavy backfilling that was not marked at the time.

For another established perspective on project communication archives, Basecamp provides a useful reference beyond any single tracking product.

A tool change whose import was never verified.

A period of drift where the categories meant something you can no longer determine.

Or simply a bad few months where you know the entries were guesses.

Why it matters more than a gap

A gap is honest: the analysis excludes it and says so.

An untrusted period is included by default and distorts the result while looking exactly like the reliable years.

And the doubt spreads: once you know one stretch is unreliable, the whole record feels less solid, which is the mechanism by which people stop consulting records they still maintain.

Marking it

A flag on the entries, or a dated note in the definitions file: this period is unreliable, for this reason.

Do not delete it. The data may still be roughly right, and deletion is irreversible.

Do not silently correct it either, which replaces a known-bad record with an invented one.

Marked is the honest state, and any analysis can then include or exclude it deliberately.

Deciding how bad it is

Read the entries. Frequently a period you distrust turns out to be fine for coarse questions and unreliable only for fine ones.

Check against an external anchor: invoices, a calendar, messages sent. A week that matches the calendar is more trustworthy than your impression of it.

And ask what specifically is wrong — durations, categories, coverage — because usually only one dimension is affected and the others are usable.

Analysing around it

Exclude and state the exclusion.

Or include with a caveat recorded alongside the finding.

What not to do is include silently, which is what happens by default and which produces conclusions nobody would defend if the qualification were visible.

Preventing the next one

The backfill flag, applied at the time.

Verification after any tool change.

And the annual read, which catches drift while you can still remember what the categories meant.

All three are cheap and all three are about preserving the ability to trust the record later.

The uncomfortable case

Sometimes the untrusted period is most of the last two years.

Then the honest options are to treat the record as ending at the last reliable point, or to restart with the old data kept as an archive.

The starting-over note covers this. What does not work is continuing to add to a record you no longer believe.

What to check

Is there a period in your record you quietly avoid?

Is it marked, or only known to you?

What specifically is wrong with it — durations, categories, or coverage?

And do your analyses currently include it?

The point

A gap is honest and an untrusted period is not.

The second is included by default and distorts the result while looking exactly like the reliable years.

Worth adding

Read the entries before writing a period off.

A stretch you distrust is frequently fine for coarse questions and unreliable only for fine ones.

Additionally

Do not delete an untrusted period and do not silently correct it.

Marked is the honest state, and any analysis can then include or exclude it deliberately.

Finally

Analyse around an untrusted period and state the exclusion.

Including it silently produces conclusions nobody would defend if the qualification were visible.

In summary

Every long record eventually contains a stretch the owner does not quite believe.

Handling it explicitly beats quietly avoiding analyses that touch it.

Last word

The practice this points at is unglamorous: a small daily act, a short weekly correction, an annual read, and a written page of conventions beside the data. None of it is difficult and all of it is easy to skip, which is why most records do not reach the point where they start paying.