What a laboratory reference interval actually is

Your veterinarian hands you a laboratory report. There is a column of values, a column of units, and a column of numbers in brackets that somebody has told you is the normal range. One of your dog's values sits just outside it, and the rest of the afternoon is spent worrying about a number in brackets.

That column deserves a better explanation than "normal range", because it is not one.

The interval belongs to a laboratory, not to a species

A reference interval is built from a population of healthy animals, measured on a particular analyser, with a particular laboratory's reagents, by that laboratory. It describes what that laboratory sees in animals it considers healthy on the machine it actually owns.

Two accredited laboratories can therefore publish intervals for the same analyte that do not overlap, and both of them can be correct. The numbers differ because the populations differ, the analysers differ and the reagents differ, not because one of them is careless.

This has a practical consequence that is easy to miss. A value is only interpretable against the interval that came with it. If you write down "ALT 96" and throw away the report, you have not saved the result. You have saved half of it.

So what does "outside range" mean

It means the printed value sits outside the interval that the reporting laboratory printed next to it. That is a factual statement about two numbers on one page. It is not a verdict, and on its own it is not a finding.

A value can sit outside an interval for reasons that have nothing to do with disease: the animal was stressed on the table, it had eaten, it is a young growing dog measured against an adult interval, it is a breed whose normal sits away from the population mean, or the sample sat for an hour longer than it should have. A value can also sit comfortably inside an interval and still be the thing your veterinarian wants to talk about, because it has moved a long way since last year.

That is why Talli reports the laboratory's own flag rather than inventing one. Only a flag printed by the laboratory that ran the sample can call a result abnormal. The app shows you the value, the interval the report supplied, and the flag the report carried, and it does not add a judgement of its own.

Colour is not a diagnosis either

A red chip on a screen is a design decision, not a clinical one. Talli holds itself to the rule that colour never asserts danger on its own. Every status is written out as words next to the colour, and the wording is deliberately flat: within range, below range, above range, outside range, worth watching, not enough data.

"Not enough data" is a real answer and appears often. It is more useful than a confident-looking chart built from two readings.

The intervals that vary

An interval is not always one pair of numbers per species. Depending on the analyte, it can differ by species, by life stage, by sex, and by an age band in months. A puppy's alkaline phosphatase is not an adult dog's, and treating them as one interval produces a scary-looking result from a normal growing animal.

Units matter for the same reason. The same analyte is printed in different units by different laboratories, and a number that looks alarming in one unit is unremarkable in another. When Talli stores a result it keeps the unit that was reported, and it knows the common alternates for each analyte so a converted value can be compared honestly rather than silently.

What Talli does with all of this

When a report supplies its own interval, that interval is the one the result is drawn against. Always. The app does not replace it with a tidier number.

For the cases where there is no interval on the page, because you typed the value in from a printout or the report simply omitted it, there is a fallback catalogue of 56 measurements: 49 laboratory analytes plus seven vitals and clinical scores. It exists so that a value can still be drawn on a range bar with an honest label, and it is described as a fallback wherever it is used.

Three rules govern that catalogue, and they are worth stating because they are the difference between a reference and a guess:

• Every interval names its source, and the source is shown next to the interval in the app. The sources are named veterinary references: university diagnostic laboratories, the major reference laboratories, the Merck Veterinary Manual, IRIS staging guidance, ACVIM consensus statements.
• Where mainstream sources disagree, the competing figures are recorded next to the interval rather than averaged into a single house number. An average of two disagreeing intervals is a third interval that nobody published.
• Missing data stays missing. Nothing in that file is invented to fill a gap.

There is one more rule that only shows up when something goes wrong. Critical thresholds are held for triage, but they deliberately do not decide how a result is classified on your screen. Only a laboratory's own flag can promote a result to critical. A threshold that Talli chose is not allowed to put the word "critical" next to your animal's name.

Why Talli will not read a photograph of your report

This is the part that connects a dry catalogue decision to something you will notice.

When the reading was tested, a vision model was given deliberately degraded photographs of veterinary laboratory reports with known correct values, and was told to write a placeholder rather than guess when a character was not legible. On the worst image it returned 32 fully populated analyte rows, 30 of them wrong, and used the placeholder zero times.

Then the same page was re-read asking only for a plain transcription. Of the 19 reference intervals it printed, 16 were invented, and the invented ones were the standard human reference ranges rather than the ones on the page.

Read that again with this article in mind. The model did not merely get a value wrong. It replaced the laboratory's own interval with a textbook interval from a different species, which is the one substitution that changes whether a correct value looks high, low or fine.

So Talli reads text out of a PDF's text layer, where the characters are the actual characters, and anything that came from a photograph or a scan goes to a person before it is filed. Not at a high confidence threshold. At any confidence at all.

What to do with your own reports

• Keep the whole report, not the value. The interval, the units and the laboratory's name are part of the result.
• Ask for the PDF rather than a photograph when a clinic offers you a choice. A PDF with a real text layer can be read exactly. A photograph cannot.
• When a value is outside an interval, ask what the laboratory's flag said and whether the interval was the right one for the age and life stage of your animal.
• Look for the series rather than the point. Talli keeps every result for a single analyte across every panel on file, so the history is there when somebody asks for it.

None of this replaces the conversation with your veterinarian. It makes the conversation shorter, because the person you are talking to can see the same page you can.