Approach
Integration problems in utility metering are rarely novel. They recur, they are usually invisible until something downstream is wrong, and they tend to sit in the assumptions systems make about one another rather than in any single system. A few of the patterns we look for first.
The corrected re-send that disappears
A head-end re-issues a reading after a correction. The natural key — meter, channel, timestamp — is unchanged; only the value differs. Any de-duplication rule keyed on identity alone will treat the second record as a duplicate and drop it. The correction is lost silently, and nothing downstream reports a problem. Handling this properly means deciding, explicitly, what constitutes a distinct record and what constitutes a revision to an existing one.
The silent gap and the declared gap
There are two kinds of missing interval. One carries a status code saying the read failed — a system that knows what it does not know. The other is simply absent: no row, no flag, nothing to notice. Downstream estimation treats both identically. The distinction matters, because a declared gap can be reasoned about and a silent one cannot.
Zero that means “no reading”
When a meter loses power, some systems record the interval as zero. Zero is also a legitimate reading — a vacant service consumes nothing. Once both are stored the same way, no downstream process can separate them, and every average computed across that period is diluted by an unknown number of non-observations.
Timestamps that claim to be UTC
Some systems emit local time under a UTC label. Some emit a fixed standard-time offset year-round, which drifts by an hour for half the year. Some emit genuine UTC. Assuming rather than confirming which produces errors that are invisible for ten months and inexplicable for two.
Register and interval that do not agree
Over any period, a cumulative energy register should advance by the sum of the intervals. Where it does not, energy was measured that the interval record does not account for. The direction of the mismatch distinguishes a genuine data loss from an outage — and the check works without depending on per-interval status codes, which is what makes it consistent across vendors.
Assumptions that stopped being visible
Voltage compliance requires a nominal to compare against. Where nominal voltage is not recorded, most systems assume 120. The assumption was made once, sensibly, and then became invisible — and a compliance figure computed across meters where it does not hold contains an error nobody can see.