The arithmetic, as CMS does it
Every citation carries points by scope and severity — a D is worth 4, an E 8, an F 16, a G 20, and the top of the scale runs to 150. Substandard quality of care is scored higher again. Points are then weighted by how recent the survey was.
Two details do most of the surprising:
- Only two standard survey cycles are rated. The most recent counts for three quarters, the one before it for one quarter, and everything older counts for nothing at all.
- Complaint surveys are scored separately, in their own recency buckets, and do not displace a standard cycle.
Which is why a new annual survey can add a citation and move your score down: it rolls the cycles back one, and whatever was in cycle two stops being rated. That is not a quirk to be explained away — on a real facility it was worth more than forty points.
The what-if shows which part moved
You can vacate a citation, reduce its scope and severity, add one, or simulate an annual survey, and see the score and the star before and after.
A single total cannot show you why, so it does not try. The change is broken into the parts it came from — your standard inspection cycles, your complaint surveys, your revisits — and those parts sum to the headline. When a scenario moves the score for a reason that is not the change you just made, the screen says so in a sentence.
A rejected change is a different question, not a weaker answer
If you build a three-change scenario and one of the citations no longer exists, the server applies what it can and refuses the rest. The figures you then see describe the changes that applied — a smaller scenario than the one you built.
So the refusal is printed above the stars, counting itself, and the star row changes its label. A confident projection for an input nobody asked about is worse than no projection, and it is the easiest mistake in this whole product to ship.
How often we agree with CMS
The only honest test is to score facilities CMS has already rated and compare. Measured on the 1 September 2026 release, across 308 facilities in Alabama, Alaska, Arizona and Arkansas:
| Against CMS's published star | Result |
|---|---|
| Exactly the same star | 89.3% |
| Within one star | 98.7% |
| We rated higher than CMS | 6.5% |
| We rated lower than CMS | 4.2% |
The last row is the one that matters. The limits we know about can only make us rate a facility higher than CMS — so disagreeing downward is the direction that would mean the method is wrong rather than incomplete. It is the number we hold ourselves to, and it is checked on every release.
Dates, because an accuracy figure without one is about nothing
CMS republishes every star monthly and recalibrates the per-state cut points monthly. A facility's rating can be a release behind the table it is being compared against, and comparing the two is comparing two vintages.
So every row is judged only against a yardstick at least as current as the cut points, and anything staler is excluded rather than quietly counted. The first time we ran this, the single facility we scored worse than CMS turned out to be exactly that: our answer was right and the cached comparison was a release old.
What we will not claim
- No CMS file carries waiver status, so a score can never be called exact. It is reported as a floor or as approximate, and which one is stated.
- Citations under dispute are excluded, the way the CMS guide excludes them — but only where CMS has published that a dispute is open.
- Unrateable means two different things, and the reason says which: CMS itself has only one standard cycle on record for you, or we could see only one cycle's worth of citations. The first is nothing to chase. The second is a gap on our side.
Related: Drafting the plan of correction · The 24-hour chart check · The 60-day pilot