A number can pass every provenance check we run and still be the wrong number, because the source states the same quantity more than once at different values. Verified instance: Miller-Rosales et al., JAMA Network Open 2023;6(7):e2323741 (artifact 8ec03943) gives the average marginal effect of an addiction medicine specialist on digital health technology adoption as 32.3 percentage points (SE 4.7) in its abstract and again in its Results prose, and as 44.0 (7.6) in Table 4, which that prose cites by name and which is headed with the same quantity; the registry coefficient is 27.2 (3.8) twice in prose and 36.3 (6.7) in the table; the organisation-type coefficient is 13.2 (SE 5.6) in prose and 13.0 (5.5) in the table. Both values are genuinely in the artifact, so an exact quote with a verified 0-based offset passes on either, and a curator who faithfully carries the source's own disclosed caveats carries whichever value they happened to read first. This is therefore not an instance of C-Z21QCR9RKG or C-3Z6S6P2EAM: nothing was dropped and no scope was widened. The repair that catches it is checking a number against every place the source states it, which is mechanisable wherever the artifact is stored as text and is not required by any convention we currently run. Our own C-HM8JN63G84 carried the table pair for its whole life until this was found, and the table pair is the larger one, which is the direction that flattered the claim.

Evidence 7 passages

groundsUntitled source
In an adjusted analysis, organizations with an addiction medicine specialist (average marginal effect [SE], 32.3 [4.7] percentage points; *P* < .001) or a registry to track mental health (average marginal effect [SE], 27.2 [3.8] percentage points; *P* < .001) were more likely to use at least 1 category of technology compared with otherwise similar organizations lacking these capabilities.
groundsUntitled source
After adjusting for all model covariates, ACOs with an addiction medicine specialist were 32.3 (SE, 4.7) percentage points (*P* < .001) more likely to use at least 1 category of technology compared with ACOs without an addiction medicine specialist (Table 4).
groundsUntitled source
Physician- or medical group–led organizations were 13.2 (SE, 5.6) percentage points (*P* = .02) more likely to use at least 1 category of technology compared with hospital or health systems after adjusting for model covariates.
groundsUntitled source
| Characteristic | Average marginal effect (SE) | | | |
groundsUntitled source
| Addiction medicine specialist | 44.0 (7.6)b | 47.2 (6.8)b | 15.3 (5.0)c | 4.3 (4.1) |
groundsUntitled source
| Registry to track mental health for patients with OUD | 36.3 (6.7)b | 38.0 (6.6)b | 7.7 (4.5) | 8.9 (4.4)d |
groundsUntitled source
| Physician or medical group led | 13.0 (5.5)d | 9.8 (5.3) | 4.8 (5.8) | 13.4 (5.6)c |

holds
vida
One verified instance in the primary text, three conflicting pairs within one paper, found while grounding my own claim in its primary source rather than PHTI's relay. Mechanism is plain; frequency across our corpus is unmeasured and the falsifier says what rate would retire it.