Category Drift and How to Spot It

Drift is not miscategorisation. It is the gradual change in what a category means to you, and it makes two periods incomparable without either of them looking wrong.

To connect this idea with an actual work log, consult the complete guide. The useful test is whether the resulting entries remain consistent and understandable during a later review.

How it happens

Your work changes and the label stays. "Writing" meant drafting in 2022 and includes editing, research and correspondence by 2025.

For another established perspective on versioned files, Git provides a useful reference beyond any single tracking product.

A new activity has no home, so it goes in the nearest category and stays.

You become stricter or looser over time about what counts as a break, a meeting, a piece of admin.

And the catch-all absorbs things quietly until it contains three distinct activities.

None of this is visible in the data. The numbers look fine and mean something different.

Why it matters

It produces false trends. "Admin has doubled in three years" may mean admin doubled, or may mean the category widened.

It invalidates exactly the comparisons a long record exists for.

And you cannot detect it afterwards from the numbers alone, which is why detection has to be built in.

Detecting it

Read actual entries, not totals. Take twenty entries from two years ago and twenty from last month, both in the same category, and compare what they describe.

This is the only reliable method and it takes fifteen minutes once a year.

The free-text note is what makes it possible, which is the strongest argument for keeping one.

The other signals

A category growing steadily with no corresponding change in your life.

A category shrinking to near zero, which usually means its content moved elsewhere rather than stopping.

The catch-all growing.

And a sense, when reading old entries, that you would categorise them differently now. That sense is the finding.

What to do about it

Do not retroactively recategorise years of data. It is a large job, it introduces its own errors, and it destroys the record of what you actually thought at the time.

Instead: note the date of the change and treat the series as two segments.

"From March 2025, admin includes invoicing, which was previously in finance" — one line in the definitions file, and every future comparison knows to be careful.

Preventing it

Write the definitions down at the start and reread them annually.

Reread, specifically. Most drift happens because nobody has looked at the definition since it was written.

And when a genuinely new activity appears, create a category for it rather than absorbing it. Absorbing is what starts the drift.

The honest limit

Some drift is unavoidable over a decade because you change.

The goal is not prevention but documentation: knowing where the boundaries moved, so the analysis can account for it.

A ten-year record with three documented redefinitions is far more usable than one that appears seamless and is not.

What to check

When did you last read your own category definitions?

Compare twenty old entries with twenty recent ones in your largest category — do they describe the same thing?

Is anything growing or shrinking without a life change to explain it?

And is there a dated note anywhere recording a definition change?

The point

Do not retroactively recategorise years of data.

Note the date of the change and treat the series as two segments — a documented redefinition is far more usable than a seamless one that is wrong.

Worth adding

Some drift is unavoidable over a decade because you change.

The goal is documentation rather than prevention: knowing where the boundaries moved so the analysis can account for it.

Additionally

Detect drift by reading entries rather than totals: twenty from two years ago against twenty recent, in the same category.

The free-text note is what makes that possible.

Finally

A category growing steadily with no corresponding life change is the clearest drift signal.

So is one shrinking to near zero, which usually means its content moved.

In summary

Drift produces false trends: admin doubling in three years may mean admin doubled or may mean the category widened, and the numbers cannot tell you which..