As often as the position can be derived without people assembling it. Which, if the system is built correctly, is continuously. The frequency question is usually asked backwards: teams pick a cadence they can operationally sustain, when the right approach is to make derivation automatic so that cadence becomes a policy decision instead of a limit. Whatever the regulatory minimum is in your jurisdiction, a report you can only produce monthly is close to worthless in a week when confidence is moving hourly.
Attestation exists to make a claim checkable by someone who does not trust the issuer. Its value therefore decays with age in exactly the way the underlying risk does not. A figure that was true six weeks ago tells a holder nothing about a redemption they want to make today, which is the only moment they actually care.
The engineering consequence is that reserve position, liabilities outstanding and the reconciliation between them should be queryable on demand. If producing them requires a person to pull custodian statements and assemble a spreadsheet, the effort determines the frequency. And the effort is highest exactly when the demand for the number is highest.
Frequency is not the only axis. Granularity matters as much: a single total tells a holder far less than a breakdown by instrument and maturity, because the question during stress is not how much you hold but how fast it converts.
Verify the current requirement in your jurisdiction against the primary regulation rather than against secondary summaries, which lag amendments. Treat that requirement as a floor and the system's own capability as the target.