Proof · Lot 2026.10
Counts from Medicare records are worth having only if they follow what is real. Each test had its rules written down before its numbers were counted, and the results are shown as they came out.
01 / Listed chains
5 listed companies publish a count every quarter. A company passes when, in at least 75 % of the quarters where its number changed, ours moved the same way, and our count divided by theirs stayed within ±10 % of its usual level in every quarter. The test passes if 3 of 5 do.
Pass · 3 of 5 · 14 quarters
Our count runs at 0.87 of the reported number (different definitions); the gap stayed within ±10 % of that level every quarter.
Our count runs at 0.82 of the reported number (different definitions); the gap stayed within ±10 % of that level every quarter.
Our count runs at 1.04 of the reported number (different definitions); the gap stayed within ±10 % of that level every quarter.
Same direction every quarter, but the gap drifted from 0.44 to 0.95 of the reported number: Medicare enrolled counselors and family therapists only from 1 January 2024, so the records caught up with existing staff. From Q1 2025 on it passes both tests (4 of 4; gap 0.9-0.95).
The gap was steady, but DaVita's number moves by about 0.2 % a quarter, so the direction test is close to a coin flip, as expected before the run.
Shown as decided before the run; none of these changes the result. “After fixes” is the same test once new sites were tied to their owner from evidence public at the time.
| Variant | Official run | After fixes |
|---|---|---|
| Only ties knowable at the time | FAIL: 2 of 5 | PASS: 3 of 5 |
| Quarters after a source change included | FAIL: 2 of 5 | FAIL: 2 of 5 |
| Every counting method, not just the chosen one | FAIL: 2 of 5 | FAIL: 2 of 5 |
Private chains file no list of subsidiaries. Rebuilt from CMS records alone, the same 5 companies pass 0 of 5 counting only confirmed matches; 1 of 5 accepting every organisation found through the chain's officers, unreviewed. So a private chain starts from an outside list of its locations or entities (its own website, a client's target list, state licences), and the records match, count and date them.
02 / Out of sample
3 more listed chains, matched from the locations on their own websites. The rules and the chains were committed before any of their numbers were counted.
Fail · 1 of 3
Its number went 909 → 866 (moves of 2-12 a quarter) while ours stayed between 245 and 248: closed sites stay in the records; our count reached 160 of its 837 listed addresses.
Our count followed every change of the reported number, an acquisition included (6 of 6).
Its number went 253 → 251 (moves of 1-4 a quarter) while ours stayed between 145 and 151: closed sites stay in the records; our count reached 142 of its 277 listed addresses.
A growing chain was followed, an acquisition included. Shrinking or flat chains were not: their closed sites stay in the records. So closures and falling site counts are not part of the data set.
03 / The Texas sample
Each with its bar set before it ran. What failed its check went to an internal file, not the workbook.
Athletico 95 %, Peak 100 %, Vista 91 %, Endeavor 62 %. Bar: 90 % of our sites on the list. The one below the bar bills under an entity that also covers its parent's other brands.
Drawn at random and searched for any outside owner (bars 18 of 20, 45 of 50 and 45 of 50; 108 of 120 together). For 31 nothing was found beyond CMS records, so the claim stays “no platform tie found”, never “independent”.
Held as written when every tie was re-read at its source; the rest were corrected before release.
Drawn “within 25 miles” pairs geocoded street to street were within 30 miles; median error 1.4 miles.
A second program, sharing no counting code with the first, recounts every number in the workbooks and on these pages.
04 / Timing
In 61 of 66 quarters our count for the quarter was public before the company announced its own (median 13-24 days by company). NPI-registry publication dates are estimates.
0 of 9 announced Texas openings were in the records before the announcement; 7 arrived a median 202 days after it.
A billing entity counts for a platform from the day its tie was public. Today's portfolio back in time is shown only as hindsight.
Medicare enrolled counselors and family therapists only from 1 January 2024; physician assistants' billing links appear from 1 April 2026. Every delivery lists all 8 known changes, gaps and repeated files.
Send us one company's location list, or ask for your own platform's book, and judge the counts against what you already know.