The short version. Before you call a well-to-well spread “drift,” rule out two normal causes: the statistical noise floor of the count (worse the fewer counts you collect) and the small, healthy movement of a photomultiplier’s gain. A real fault is progressive and repeatable; normal jitter is random and stays within a band. The single best test is to count more — more counts, or a higher-yield isotope — and see whether the spread shrinks. If it does, it was statistics, not the instrument.

What well-to-well spread is

On a multi-well counter, the system efficiency check counts the same calibrated source in every well and reports each well’s efficiency, then the spread — the gap between the highest and lowest well. A small spread is expected and healthy. The question is never “is there any spread” — there always is — but “is this spread telling me something.”

Cause one: the statistical floor

Every count carries an irreducible statistical uncertainty set by the number of events you collect. Collect 120,000 counts and the one-sigma noise on that measurement is about 0.3%. Collect only 10,000 counts and it climbs to about 1%. Collect a couple of thousand and you are at 2–3% before the instrument has done anything at all.

This matters enormously for low-yield isotopes. A tracer like Cr-51 emits a counted gamma in under 10% of its decays, so at clinical activities the effective count rate is low — and a 4–5% well-to-well spread can be entirely statistical noise, with a perfectly healthy instrument underneath. The fix is not service; it is more counts: a longer count time, or a higher-yield source such as Cs-137 for the stability check.

Cause two: normal photomultiplier jitter

A healthy photomultiplier does not hold its gain to the last decimal. Its photopeak drifts a little from day to day, and a detector settles over its first year of service as the tube finds its long-term operating point. That movement is normal and within specification. It only becomes visible as “efficiency drift” when something amplifies it — most often an energy window that has been narrowed below the factory default, which turns a small, healthy peak shift into a larger swing in counts. (See Energy windows — why a properly set window beats a tight one.)

The trap: judging every well against your quietest one. Occasionally one detector is not just stable but exceptionally stable — holding to a fraction of a percent, better than a photomultiplier normally performs. It is tempting to treat that well as the standard and conclude the others are “drifting” against it. That is backwards. The quiet well is the outlier; the others may be moving completely normally. Read the spread as “one remarkably quiet tube and several normal ones,” not “one good tube and several bad ones.”

How to tell real drift from normal scatter

Four checks separate a fault from the noise floor:

  • Count more. Repeat the check with far more total counts — a longer count time, or a higher-yield isotope. If the spread shrinks toward a fraction of a percent, it was statistics. If a high-statistics check still shows several percent, now you have something real to look at.
  • Look at the trend, not the snapshot. Plot each well over weeks. Random scatter inside a stable band is noise. A monotonic walk in one direction, or a clear step that then persists, is the signature of a genuine change.
  • Check the photopeak position. If a well’s efficiency moves together with its photopeak channel, you are watching gain move. Whether that matters depends on size and persistence — a few keV that wanders is normal; a steady march is not.
  • Rule out the window. Confirm the energy windows are at the factory default before trusting the spread at all. A narrowed window manufactures apparent drift from normal motion.

When a spread is actually worth acting on

Act when the spread is real (it survives a high-statistics count), localized (one well, not the whole array), and progressive (it grows or steps over time rather than scattering around a stable mean). A single well that walks steadily away from the group, or throws repeatable excursions a high-count check still shows, has earned a closer look. A spread that is none of those things — that shrinks when you count longer and otherwise drifts randomly within a band — is the instrument behaving exactly as it should.

What to do

  • Use a high-yield source for stability checks. Cs-137 gives you the counts to push the noise floor down to a fraction of a percent, so a real change stands out.
  • Confirm your windows are at factory default before reading anything into a spread.
  • Trend each well over time. The shape of the history tells you far more than any single reading.
  • Don’t benchmark against your quietest tube. Judge each well against specification, not against the best-behaved detector in the array.

Further reading