A clinician sees fourteen patients. At seven in the evening, they sit down to write up the ones they did not finish during the day. They are not retrieving those appointments. They are reconstructing them, and the distinction is one psychology has spent a century establishing. Memory does not work like a recording. Retrieval is an act of rebuilding, and what gets rebuilt is shaped by expectation, by what usually happens, by what happened in the appointments on either side of it. The gaps get filled with the plausible rather than left empty, and the person doing the filling cannot feel the difference between the part they retrieved and the part they supplied.
Why clinical notes are the worst case for this
Several factors stack up badly here. Volume first. Fourteen appointments about broadly similar problems create interference, and similar episodes blur into a composite. Ask which patient mentioned the dizziness, and the honest answer is often a reconstruction dressed as a recollection.
Then expertise, which helps and hurts. An experienced clinician has strong schemas for how a presentation usually goes. Schemas speed up recognition, and they also fill gaps with the typical case. The more expert the clinician, the more confident and more detailed the reconstruction, without being more accurate.
And the delay. Between the appointment and the write-up sit hours and a dozen other patients, which is precisely the condition under which reconstruction replaces retrieval. None of this is carelessness. It is ordinary human memory doing what it does, under conditions designed to maximise the effect.
What it produces, measured at scale
American healthcare has the most thoroughly audited records in existence, because a federal agency samples and publishes. The US Office of Inspector General reports that across diagnoses it considers high risk, roughly 70 per cent were not supported by the associated medical records, some categories above 90 per cent.
In one audit of 97 records carrying an acute stroke diagnosis, none were supported as acute strokes by the documentation. In 68 of those cases, the patient genuinely had had a stroke, recorded in their history, and the entry described it as current. That specific error is what reconstruction looks like. The fact was right. The temporal frame, the detail most dependent on context that was present at the time and absent later, was the part that did not survive.
Why reading it back does not fix it
Here is the uncomfortable part for anyone who assumes review solves this. A reconstructed memory does not feel reconstructed. It carries the same sense of knowing as an accurate one, sometimes more, because the plausible version is tidier than what actually happened. So a clinician re-reading their own note finds nothing wrong, and a reviewer reading it months later has only the document and no access to what was omitted. Review catches contradictions and missing fields. It cannot catch a confident, coherent, slightly wrong account, which is exactly what reconstruction produces.
The only intervention that works
If the problem is the gap between the event and the record, the solution is closing the gap. Not better memory, not more careful review. Less delay. In healthcare, the move is called concurrent hcc coding, documenting and checking during the episode of care rather than after it. The clinician still has the patient in front of them, so a prompt for a missing detail retrieves something rather than generating it.
The principle generalises past medicine. Any account written from memory hours later is a reconstruction, whether it is a clinical note, an incident report, a research observation or a conversation you are recalling for someone else. The confidence you feel about it is not evidence of its accuracy, and that is the finding people find hardest to accept about their own recall.
Leave feedback about this