RenewOps / Research / EC-WP-1100

Missing Is Not Zero

Six ways a renewable monitoring platform reports something that is not true — and the design rules that stop each one.

A field taxonomy of the ways a fleet monitoring platform reports an absence as a value, drawn from a 2026 assessment of a live production monitoring stack on a multi-gigawatt utility-scale PV and storage fleet. Twenty-plus verified defects, each with a source and a timestamp, sorted into six recurring structural classes and answered with five design rules. Two of the defects were in our own platform.

The six failure classes

• Substitution. An absent measurement is written into the record as a real value, usually zero, because zero is the default of a numeric column.
• Denominator drift. A metric's denominator counts only currently healthy assets, so it shrinks when an asset fails and the metric improves while the plant degrades.
• Clock collapse. Event time and observation time share one column, so neither question can be answered reliably.
• Staleness masquerade. A display drops the date from a timestamp, so an old value looks identical to a fresh one.
• Plausibility void. No bound exists anywhere in the system to check a value against, so an impossible value passes through as live.
• Provenance erasure. Measured, modelled, inferred and issued values sit in one column and render identically, so no one can tell which can be cited.

Contents

1. The failure that looks like health
2. A taxonomy, and why symptom-sorting fails
3. Substitution
4. Denominator drift
5. Clock collapse
6. Staleness masquerade
7. Plausibility void
8. Provenance erasure
9. Five rules that close all six classes
10. What each class costs when it meets a data request
11. A self-assessment protocol
12. What RenewOps does about this

Next in the series: The Board That Deleted Its Own Alarms