How your readiness score works
Updated
Every verdict on this site links here. This page explains exactly how the estimate is built, what it is anchored to, and — just as important — what we do not know yet. When the honest answer is “we don’t have that data,” this page says so.
Calibration status · v1
The current model is a transparent heuristic: it is anchored to the official passing standard (1,400 on a 1,000–1,600 scale) and the published 2026 content outline, not yet to our own outcome data. As users report their real exam results, we publish the comparison — practice estimates vs. actual outcomes, with sample sizes — on this page, quarterly. Until then, treat every estimate as directional with real uncertainty.
Step 1 · The estimate
From % correct to a scaled-score estimate
The real PTCE reports a scaled score from 1,000 to 1,600, with 1,400 to pass; PTCB does not publish its scaling function. We use an explicit piecewise-linear map with three fixed anchors, and we publish it in full rather than pretending to know the official curve:
- 0% correct → 1,000 (the scale floor)
- 72.5% weighted correct → 1,400(the passing score) — our working estimate of the passing standard, sitting between the commonly cited ~70% rule of thumb and the official practice tools’ thresholds
- 100% correct → 1,600 (the scale ceiling)
Between anchors the map is linear, clamped to the scale, rounded to the nearest 10. The table below is generated by the same code that scores your diagnostic — if we change the model, this table changes with it.
| Weighted % correct (scored items) | Estimated scaled score |
|---|---|
| 0% | 1,000 |
| 30% | 1,170 |
| 50% | 1,280 |
| 60% | 1,330 |
| 72.5% · pass line | 1,400 |
| 80% | 1,450 |
| 90% | 1,530 |
| 100% | 1,600 |
Error band: a 90-question sample of an 80-scored-item exam carries meaningful sampling noise — treat the estimate as ±40–60scaled points, wider near the middle of the scale. A single estimate near the pass line means “too close to call,” and the verdict says so.
Step 2 · The verdict
Three bands, no hole
The verdict is a banded reading of the same weighted % correct — contiguous bands, so two people one question apart never get contradictory stories:
The gap is broad — the report points at the weakest domain, and the free drills are the fastest way to move it.
Passing distance is one focused domain away. This band is where targeted drilling pays off most.
Practice performance clears our pass-line estimate. Never a promise — exam-day sampling and nerves are real.
Step 3 · The sample
Why the forms are weighted 32/17/21/20
Every sim form mirrors the real exam’s shape: 90 questions, 80 scored and 10 unscored, on the real 1h50m clock. Domain counts come straight from the official weights (effective 2026-01-06):
| Domain | Official weight | Questions per form | Scored |
|---|---|---|---|
| Medications | 35% | 32 | 28 |
| Federal Requirements | 18.75% | 17 | 15 |
| Patient Safety & Quality Assurance | 23.75% | 21 | 19 |
| Order Entry & Processing | 22.5% | 20 | 18 |
Your % correct is computed on scored items only — the 10 unscored items never move your estimate, exactly like the real exam’s unscored pretest questions. Which items are unscored is never revealed mid-exam.
Item calibration
How items are difficulty-tagged
Every question carries a difficulty from 1 (single-fact recall) to 5 (multi-step application under realistic distractors), assigned at authoring time against written criteria: number of reasoning steps, distractor plausibility, and how deep the fact sits in the outline area. Generated math items compute difficulty from the template’s parameters — step count and unit conversions — so two problems of the same shape always carry the same tag. As attempt data accrues, observed per-item accuracy replaces authored difficulty (that switch will be announced in the changelog).
Sourcing
How questions are written and verified
- Every fact question carries its references. Drug facts are checked against RxNorm and openFDA at writing time; law and regulation items against FDA, DEA, and USP public documents; exam-logistics facts against ptcb.org. The reference and its verification date are stored on the question itself.
- Every math answer is computed, never hand-written. Generated problems come from a tested calculation engine: the answer, each worked step, and every distractor are computed from the problem’s own numbers, and an automated property test re-derives thousands of random cases on every code change.
- What we never use: competitor question banks, shared exam recollections, or “brain dump” sites. Items are written from the official content outline and primary sources, full stop.
- A certified reviewer signs the bank. Questions ship in draft until a CPhT reviewer approves them; the reviewer’s byline appears on every question-bearing page at launch. If a byline slot is empty, the review is not done — we show the status honestly rather than invent a persona.
The loop
Report a question, watch it get fixed
Every question on the site has a “report this question” control. Reports land in a triage queue with a 24-hour review target; anything that changes a live question is published in the public changelog with the date and what changed. After your real exam, the outcome survey (did you pass, and in which score band?) is what turns this v1 heuristic into a calibrated model — that aggregate comparison will be published here.