Administrative teams in hospitals produce a large amount of data about their own work: admission lists, referral trackers, scheduling logs, billing queues. Every one of those artefacts mixes things that were measured with things that were inferred. The questions below take an analytical look at how to tell the two apart. They concern office and records processes only. Nothing here is clinical guidance or medical advice, and none of it should be applied to decisions about anyone’s health.
What does it mean to separate known facts from guesses?
It means classifying each statement by the strength of its support. A known fact is backed by an artefact you can inspect: a timestamped entry, a signed form, a system log. An assumption is a statement that is probably true because similar cases usually were, but that nobody has looked at in this instance. A guess is a statement with little or no support that someone offered to fill a gap. The classification is not about how likely the statement is to be true. A highly probable assumption remains an assumption until it is observed.
Why do teams mix them up so easily?
Three forces are at work. First, repetition: a statement passed along a chain gains apparent weight without gaining evidence. Second, tone: confident phrasing is read as accurate phrasing. Third, time pressure: checking costs minutes, and at the end of a long shift a plausible sentence is cheaper than a verified one. Notice that none of these requires carelessness. They are structural, so the cure must be built into the process too, such as labels in the template, rather than a reminder to “be more careful”.
How do you label statements in practice?
Add a short tag to each line in a handoff note or tracker. Many teams use three: confirmed, assumed and guessed. For confirmed lines, add the source and the time you saw it. For assumed lines, add the reason you believe it. For guessed lines, add what would settle it. A tracker with these three extra fields takes a few seconds longer to fill in and saves considerably more time downstream, because the reader no longer has to reconstruct how sure the writer was.
Which statements deserve checking first?
Rank uncertain statements by two factors: how costly it is if the statement is wrong, and how cheap it is to check. A statement that is expensive if wrong and cheap to check goes first. One that is cheap if wrong and expensive to check can often be left as a labelled assumption. This is a simple decision matrix, and it keeps teams from the twin errors of checking nothing and checking everything.
Can a guess be tested without causing harm?
Often, yes. The safest tests are read-only: look up a record, search a queue, compare two lists, ask a colleague a closed question. Avoid tests that change state, such as editing a record to see whether a downstream system updates. A good rule of thumb is that a test should be reversible in seconds and observable by someone else. If a test fails both conditions, it is probably an action rather than a check, and it should go through whatever approval your organisation uses for changes.
How should external information be treated?
Public articles and general guides are useful for context, but they describe typical situations, not your own site’s records. Material such as a patient-facing guide to finding hospitals for digestive care may help staff anticipate the questions families ask, yet every specific claim about local processes still needs confirmation from your own documents. The same caution applies to technical writing that crosses into other fields, for example a plain-language piece on running automated machines without losing your composure, whose general lessons about calm monitoring are worth reading but which does not tell you anything about how your records behave.
Is there a parallel in other professions?
Yes, and it is a useful one for analysts. Responders handling a live software incident keep a running ledger of confirmed readings, shared beliefs and open hypotheses, and they trial a hypothesis with a low-risk step before making a change. The people behind DevOpsSchool run an online learning platform in this space; its pages mention courses, certification tracks and study resources, and the DevOpsSchool site is where readers can see how those habits are framed for students. The context differs from a hospital office, but the structure of the reasoning is shared.
What does a good week of data look like after adopting the labels?
Expect three observable changes. The proportion of lines tagged as confirmed will rise, slowly, as people find that the cheap checks are worth doing. The number of rework incidents traced to an unlabelled assumption should fall, although you should track that carefully and avoid claiming a precise effect. And the quality of questions in handoffs improves: instead of “is this done?”, people ask “which part is confirmed?”. If you keep a simple tally of the three labels over several weeks, the trend itself becomes evidence about whether the habit is taking hold.
How should a team introduce the labels without friction?
Start with one document type and one team. A referral tracker or a daily handoff sheet is ideal, because it is read by several people and edited often. Explain the three labels in a single paragraph and show two worked examples from the team’s own material, so nobody has to translate from a generic illustration. Resist the urge to audit past entries; the aim is to change what is written from now on, not to relitigate old notes.
After two weeks, review a handful of entries together. Ask which labels were easy and which were awkward, and adjust the template accordingly. Teams usually discover that the “assumed” category needs a short reason field, and that the “guessed” category quietly disappears once people learn that a guess they cannot test is better left out of the record altogether.
What are the limits of the approach?
The labels capture your certainty, not the truth. A confirmed line can still be wrong if the underlying record is wrong, and an honest guess can turn out correct. Labelling also does nothing to resolve disputes between sources; when two records disagree, both statements are known to exist while the underlying fact remains unknown, and someone with authority over the data has to decide which one prevails. Finally, the approach is not a policy. Use your organisation’s own procedures for anything with regulatory or patient-facing consequences.
A compact summary
- Sort every statement into confirmed, assumed or guessed before acting on it.
- Attach a source and a time to confirmed lines, and a reason to assumed ones.
- Check first what is costly if wrong and cheap to verify.
- Prefer read-only tests that someone else can observe.
- Treat outside articles as context, not as records of your own site.
None of this requires new software. A spreadsheet column, a template line or a shared habit is enough. What it does require is a small, steady willingness to say “I have not checked that yet”, and a workplace where that sentence is welcomed rather than punished.
