If PharmSee samples 200 NHS Jobs pharmacist postings and finds zero explicitly titled "FCS Pharmacist" or "Olezarsen Prescribing Specialist", can we claim those specialties do not exist in the UK workforce? No. The honest answer comes from the Rule of Three, a well-established result in statistics that gives an approximate 95% confidence upper bound for events that were not observed in a finite sample.

This article explains what the Rule of Three says, when PharmSee uses it, and how to read PharmSee's zero-hit claims so you don't over-interpret them.

What the Rule of Three says

For a binary event (present / not present) that is not observed in a sample of size n, the approximate 95% confidence upper bound on the event rate is 3/n. That is the rule, and it is derived from the Poisson approximation to the binomial distribution when the event count is zero.

For PharmSee's standard NHS Jobs sample of n=200, 0 events gives a 95% upper bound of 3/200 = 1.5%. For a population of 516 live NHS pharmacist postings, that translates to roughly 8 postings that could exist undetected in the full sample. For n=400 (a combined two-cycle sample), 0 events gives 3/400 = 0.75%, or ~4 postings in the 516-posting population.

The Rule of Three is a rough but widely-used approximation. More precise binomial confidence bounds (Clopper-Pearson) give very similar numbers at these sample sizes — 0/200 has a Clopper-Pearson 95% upper bound of 1.83%, close to but slightly wider than the 1.5% Rule of Three estimate. PharmSee uses the Rule of Three because it is simple, communicable, and close enough for workforce-signal reporting.

PharmSee's standard zero-hit quoting format

Past PharmSee cycles reported findings like "0 postings found for olezarsen, inclisiran, or FCS pharmacist across the NHS Jobs sample". That claim is factually correct but is often misread as "this specialty does not exist". It should be quoted as:

PharmSee observed 0 "olezarsen" specialty postings in a 200-item NHS Jobs sample (38.5% of 516 live postings). By Rule of Three, the 95% upper confidence bound is ~1.5% of the population, or approximately 8 postings. The signal is specialty is rare or invisible in external NHS Jobs postings, not specialty does not exist.

That quoting style is the cycle-14 formalisation of a habit several earlier cycles stumbled into. Zero-hit claims should always carry the Rule of Three upper bound, and the interpretation should always distinguish between rare-in-sample and rare-in-reality.

Why the distinction matters for rare specialties

Many rare pharmacy specialties are hidden in metadata rather than absent from the workforce. A specialty lipid pharmacist might appear in the job title as "Highly Specialist Clinical Pharmacist" and only mention olezarsen in the role description. A PharmSee text search against the job title returns zero; a full-text search against the description might return several. The zero hit is a metadata artefact, not a workforce truth.

This is especially acute for:

• Haematology / haem-onc pharmacy — usually titled "Clinical Pharmacist – Haematology" or "Specialist Pharmacist – Cancer Services" rather than by the drug or indication
• Biologic / advanced therapeutics — often rolled into rotational descriptions
• Rare metabolic / lipid specialties — usually attached to tertiary centre clinical pharmacy roles
• Homecare pharmacist roles — frequently hidden because the employer is not in PharmSee's job source set at all

For each of these, a zero hit in a 200-sample is only weakly evidentiary. It bounds the upper limit of the specialty's presence in the externally-titled NHS Jobs sample, not the workforce at large.

How to combine two-sample zero hits

PharmSee occasionally runs sequential samples — two 200-item NHS Jobs pulls in different cycles — and finds zero events in both. A naïve reader might think this makes the zero claim 2× stronger. The correct calculation is to combine the sample sizes: two independent 200-samples with 0 events gives n=400, upper bound 3/400 = 0.75%, translating to ~4 postings in the 516-posting population.

That 4-posting upper bound is how PharmSee's cycle-15 "Boots has zero externally-advertised Pharmacy Technicians" finding is quoted. The combined n=400 from cycles 13 and 15 lets us say "Boots does not visibly advertise technician roles under that title" with defensible strength-of-claim, while still leaving room for up to ~4 such postings to exist in the full Boots vacancy set unseen by our sample.

When the Rule of Three does not apply

Three caveats:

• The samples must be independent. Two pulls of the same 200 postings at different times are not two samples of size 200 — they are one sample of size 200 with partial refresh. PharmSee's sequential samples are pulled at least one full cycle apart and are effectively independent because the NHS Jobs feed refreshes weekly.
• The specialty label must be consistent. If a specialty is labelled differently between samples (e.g. "FCS" in one cycle, "Familial Chylomicronaemia Syndrome" in the next), combining the zero counts inflates apparent confidence without actually increasing coverage.
• Population boundaries must match. If cycle 13's population was 547 Boots postings and cycle 15's was 540, the combined population isn't exactly 1,087 — it's the union, minus whatever overlap exists.

Use PharmSee to run your own NHS Jobs rare-keyword audit

The PharmSee jobs explorer exposes NHS Jobs postings filtered by source and keyword. For any rare-specialty claim you want to verify, pull the same 200-sample, count zero / positive hits, apply Rule of Three, and quote the bounded result — not the raw count. That is the minimum defensible reporting standard PharmSee uses internally.

Takeaway

The Rule of Three is simple: if you didn't see it in n samples, the 95% upper bound on its rate is 3/n. For PharmSee's standard n=200 NHS Jobs audit, 0 events means ≤8 postings in the 516-population. Any "X specialty does not exist in the UK pharmacy workforce" claim sourced from a finite NHS Jobs sample should be quoted with the Rule of Three upper bound, and the zero hit should be interpreted as a metadata or sampling signal, not a workforce verdict.

PharmSee's methodology page — cycle 14 formalisation, re-quoted here for cycle 16. Every PharmSee zero-hit claim across the blog should be read with this framework in mind.