To me, that sounds like you have more than one point on the Modbus data source and that there is a gap in the registers between the points you intend to read that the device considers an illegal data address. I would have expected you to have events for point read errors on the events page. The solution is likely to check "Contiguous batches only" such that only registers you have configured are read, and not bundled into larger requests to reduce the total number of requests
i.e. points at holding register 0 and 99, max read registers of 100
w/o contiguous batches you'd send one request to read holding registers, at index 0 for 100 registers. If you do not have permission to read register 50, this will fail with an illegal data address error.
w/ contiguous batches you'd send two requests, one for index 0 length 1 (depending on point locator data type I guess, since it could span a couple registers if it's an eight byte type, for instance) and a second request for index 99 length 1.
But, when you force a point read, you only send a request for that one point, and there is not a chance to bundle other registers into the request. The one minute data you describe as being the same is likely the result of the interval logging going about its business without getting updates, due to the error responses from the device.