Designing NAV controls that catch problems before publish

A NAV control that only runs after publish is not a control. It is a post-mortem. The point of a control is to catch a problem before an investor ever sees the number.

Three checks worth automating

A day-over-day movement check flags any position or NAV move outside a set tolerance band, using simple statistical bounds rather than a fixed percentage that stops working as the fund grows. A price-source check compares the price used against a second, independent source and flags a gap. A completeness check confirms every position, cash balance and corporate action expected for the period has actually landed before NAV calculates.

None of these are exotic. What matters is that they run automatically, on every NAV cycle, against the same ledger the rest of the fund operates on, rather than as a manual checklist someone works through under time pressure.

Where machine-learning models help, and where they do not

A statistical or machine-learning model is useful for flagging what looks unusual against historical patterns. It is not a substitute for the deterministic checks above. Use models to prioritize what a human looks at first, not to replace the checks that decide whether NAV is right or wrong.

Funds that build these controls into the ledger itself, rather than as a spreadsheet run before publish, catch the same problems earlier and with less manual effort every single cycle.