Enterprise — sub-metered

Large estates with a BMS or aM&T system already in place
Not built

What this tier can tell you

1 Hz, sub-metered circuits · via BMS / aM&T integration, or dedicated sub-meters

  • Everything the meter-data tier does
  • Attribute load to circuits and plant, not just to a site
  • Event-level diagnostics — “the air handling unit started at 04:12”
  • Demand and power-factor analysis
  • Appliance-level disaggregation (NILM) becomes reachable

What it cannot

Stated, because a capability nobody contradicts becomes an assumed one.

  • Be delivered without hardware and on-site engineering
  • Compete on onboarding friction — that advantage is the other tiers’

What would have to be built

  1. Make data capability a first-class concept. The engine currently runs every detector regardless of resolution. Gating them on declared capability — and annotating each finding with what was visible — is the spine all three tiers hang off. N3rgyAdapter already declares capabilities; nothing consumes it yet.
  2. A sub-metered dataset. One building at 1-minute resolution with circuit-level ground truth. That makes this tier demonstrable without a single device.
  3. Circuit-attribution detectors. Event-level diagnostics and demand analysis, which the half-hourly tier structurally cannot support.

The decision this tier implies

Sites that already have sub-metering usually already have a BMS or aM&T system. So this is not a greenfield upsell — it competes with an incumbent, on their turf, with integration work.

PROJECT_PROPOSAL.md §4 explicitly excludes BMS integrations requiring on-site engineering, and the whole pivot was away from hardware. This tier walks back toward it. That may well be right — owning the upgrade path beats ceding it — but the onboarding-friction advantage the other tiers compete on does not transfer.

A synthetic sub-metered dataset can demonstrate the product surface honestly. It cannot validate disaggregation accuracy — that would be testing a model against data generated from its own assumptions. Any accuracy claim needs real sub-metered ground truth.