Failure Modes, Listed

These are the specific things that kill multi-year records. Each has a detection and a remedy, and almost all of them are cheaper to prevent than to repair.

When this principle needs to support project or client work, refer to this implementation guide. It can help test whether the log preserves enough context without turning capture into a separate task.

During the first year

Too many fields. Signal: the entry takes over a minute. Fix: remove a field, permanently.

The broader practice of file synchronization is also covered by Dropbox, which is useful when designing a record that should remain understandable later.

Wrong granularity. Signal: more than ten entries a day for a question that does not need them. Fix: coarsen, and date the change.

No gap convention. Signal: a missed day feels like the record is broken. Fix: decide what a gap means and mark it.

Boredom at month four. Signal: entries thinning with no life change. Fix: schedule a read; the reading is what sustains the recording.

In years two and three

Category drift. Signal: a category growing without a life change. Fix: read old entries against new; redefine with a date.

Dead fields. Signal: a column that is mostly empty. Fix: remove it from the schema rather than leaving it unfilled.

Untracked backfilling. Signal: you cannot tell which entries were written at the time. Fix: a flag, from now on.

Tool dissatisfaction. Signal: an urge to migrate with no reason from the justified list. Fix: usually nothing; the urge is boredom.

In the middle years

Migration damage. Signal: a step in the data at a tool change. Fix: verify old years after any import, and keep the pre-migration export permanently.

Sync duplication. Signal: totals that do not match your sense of a period. Fix: check for adjacent identical entries; prefer one authoritative device.

Untrusted periods. Signal: analyses you quietly avoid. Fix: mark the period and its reason; do not delete or silently correct.

At any point

Reading stops. Signal: you cannot remember the last time you looked. Fix: a date in the calendar, quarterly.

This is the one that precedes most of the others, because maintenance is sustained by consultation and nothing else.

The pattern across all twelve

Nine of them are detected by reading the record rather than by looking at the data.

Eight are prevented by something written down: a convention, a definition, a dated note.

And none of them is prevented by resolving to be more careful, which is the response people actually try.

The order of prevention

If you do one thing: schedule the reading.

If two: write the definitions file.

If three: export monthly and open one occasionally.

Those three prevent or catch eleven of the twelve.

What to check

Which of these has already happened to a record you abandoned?

Which is happening now, quietly?

Are the three preventions in place?

And when did you last read rather than add?

The point

Nine of the twelve failures are detected by reading the record rather than by looking at the data, and none is prevented by resolving to be more careful..

Worth adding

Schedule the reading, write the definitions file, export monthly.

Those three prevent or catch eleven of the twelve failure modes.

Additionally

Nine of the twelve failures are detected by reading rather than by looking at data, and eight are prevented by something written down..

Finally

Reading stopping is the failure that precedes most of the others, because maintenance is sustained by consultation and nothing else..

In summary

Twelve failure modes, each with a signal and a fix.

Almost all are cheaper to prevent than to repair, and none is prevented by resolving to be more careful.

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.