The report Energy Agent wrote on an ordinary school
This is the report the product writes, read on the site instead of as a PDF. Every finding keeps its chart, its dollars and the arithmetic behind them; the faults the 3D school shows open there at the moment they happen; and the last chapter is how a saving gets verified after the fix.
The school is synthetic, built by the Energy Agent demo engine so that the data can be public. Not a customer result. Run of 2026-09-24. Values at the school’s blended rate unless you move the what-if sliders.
Maple Grove Elementary
$30,215a year identified
Six findings on an ordinary school: units and lights that never stand down, weekends run like weekdays, four afternoons that set the demand charge, and a night floor that has been creeping up all year.
At a glance
What the run found
9
findings
8 priced
$30,215
a year identified
365 days of data · ×1.00
138,239
kWh a year
6
active faults
$27,095
high confidence
$3,120
medium confidence
$0
low confidence
The home page’s animation prices the simulator’s planted faults at a flat $0.12 per kWh, which comes to about $43,500. This report is the engine’s own run on the same year, priced at the school’s tariff from its bills: two of the planted faults were evaluated clean, two others fired instead, and the total is what the engine would put in front of the owner.
How this building compares
Against its peers
53.8
kBtu per sq ft per year, electricity
31.7
peer median (education)
Bottom quartile, about the 79th percentile
453,142
kWh a year to reach the peer median
EIA CBECS 2018 electric-intensity percentiles by activity, electricity only, on a full year of data.
Data sources
What the engine was given
Interval meter data, 35,040 readings at 15 minutes
Every priced figure on this page re-prices as you move these: energy at the rate, demand at the demand rate. The time-of-use figure is priced on its own periods and stays. Total at these rates: $30,215
In the report’s own order: the six faults on the meter and the three findings about the tariff, each with its arithmetic and its evidence.
1. MET-04 · Building · high
Weekend load shape looks like a weekday
$13,607a year · High confidence
What it means
Saturday and Sunday demand profiles look almost identical to weekday profiles, despite no occupancy.
Why it matters
An entire empty building running 7 days a week instead of 5 — a 40% energy overhang. Common in tenant-controlled HVAC and BMS schedules that were never set.
Start here
Enable weekend setback. Verify schedule applies to all zones. Look for tenant overrides that defeat the BMS.
Detected 102 events totalling 2,448 h between Jan 4, 2025 and Dec 21, 2025. Estimated savings: $13,607 a year (87,903 kWh), confidence high.
Show the math+
Basis. priced directly from the meter (see assumptions)
Inputs
Input
Value
Source
Days of data
365 days
span of the fault's channels
Blended electricity rate
0.155 $/kWh
from bills
Steps
Step
Result
Formula
Wasted energy over the data window
87,900.6 kWh
priced directly from the meter (see assumptions)
Annualization factor
1 ×
365 / 365 days
Annualized energy
87,903 kWh/yr
wasted kWh × factor
Annual savings
13,607 $/yr
annualized kWh × $0.155/kWh
Assumptions. 102 weekend days ran like weekdays: 125572 kWh above each day's overnight floor × 70% avoidable = 87901 kWh a scheduled weekend would run near the overnight floor; 30% is kept for legitimate weekend use
Likely range $10,886 to $16,329 a year.
Evidence: the main meter around the longest event, the events shaded. Whole-building demand around the longest event (15-minute readings, kW); the shaded spans are the flagged events.
Baseload demand jumped discontinuously up or down on a specific date and didn't return.
Why it matters
Step changes usually mark an event: equipment failure, retrofit, schedule change. Understanding which it was tells you whether to celebrate or fix.
Start here
Identify the date, cross-reference with maintenance logs and commissioning records.
Detected 1 event totalling 168 h between Mar 22, 2025 and Mar 29, 2025. Estimated savings: $6,984 a year (45,114 kWh), confidence high.
Show the math+
Basis. priced directly from the meter (see assumptions)
Inputs
Input
Value
Source
Days of data
365 days
span of the fault's channels
Blended electricity rate
0.155 $/kWh
from bills
Steps
Step
Result
Formula
Wasted energy over the data window
45,112.7 kWh
priced directly from the meter (see assumptions)
Annualization factor
1 ×
365 / 365 days
Annualized energy
45,114 kWh/yr
wasted kWh × factor
Annual savings
6,984 $/yr
annualized kWh × $0.155/kWh
Assumption. baseline stepped up 10.3 kWh/h × 8760 h × 50% assumed reversible
Likely range $5,587 to $8,380 a year.
Evidence: the main meter around the longest event, the events shaded. Whole-building demand around the longest event (hourly readings, kW); the shaded span is the flagged event.
A measured peak demand interval is far above what historical demand for that calendar position would predict.
Why it matters
Utility demand charges are usually billed on the single highest 15-minute interval of the month. One bad peak can dwarf weeks of careful energy management.
Start here
Investigate what was running during the peak interval — startup sequences, lighting test, HVAC pre-cool. Stagger startups to flatten the peak.
Detected 2,298 events totalling 0 h between Aug 18, 2025 and Nov 17, 2025. Estimated savings: $5,696 a year, confidence high.
Show the math+
Basis. priced directly from the meter (see assumptions)
Inputs
Input
Value
Source
Days of data
365 days
span of the fault's channels
Blended electricity rate
0.155 $/kWh
from bills
Steps
Step
Result
Formula
Wasted energy over the data window
0 kWh
priced directly from the meter (see assumptions)
Annualization factor
1 ×
365 / 365 days
Annualized energy
0 kWh/yr
wasted kWh × factor
Demand-charge savings
5,696 $/yr
priced directly from the meter (see assumptions)
Annual savings
5,696 $/yr
annualized kWh × $0.155/kWh + demand $
Assumption. 40.4 kW of monthly peak attributable to spikes × $11.75/kW-month (demand $/kW from bills) × 12
Likely range $4,557 to $6,836 a year.
Evidence: the main meter around the longest event, the events shaded. Whole-building demand around the longest event (15-minute readings, kW); the shaded spans are the flagged events.
Time-of-use load shifting on the main meter: on-peak runs at a 0.07 $/kWh premium
$2,564a year · Medium confidence
32% of the main meter's energy lands in the on-peak period at $0.144/kWh vs $0.072/kWh off-peak. Shifting a tenth of it (pre-cooling, staggered equipment starts, rescheduling batch loads) is worth about $2,564/yr.
Show the math+
Basis. Move a share of on-peak consumption into the cheapest period (0.072 $/kWh premium).
Inputs
Input
Value
Source
Annual on-peak energy
358,589 kWh/yr
simulated bill
Shiftable share (assumption)
0.1 ×
planning assumption
on-peak premium over cheapest period
0.072 $/kWh
rate schedule GSDT — General Service Demand, Time of Use
Steps
Step
Result
Formula
Energy shifted
35,859 kWh/yr
onpeak_kwh × shift_fraction
Annual savings
2,564 $/yr
shift_kwh × premium
Assumption. Assumes 10% of on-peak kWh can move (pre-cooling, staggered starts, batch loads).
The year as a daily mean of the main meter.
5. MET-11 · Building · medium
Lighting on after hours
$521a year · High confidence
What it means
The lighting submeter runs well above its normal unoccupied level outside the occupancy schedule.
Why it matters
Lighting left on after hours is close to 100 % avoidable — schedules, sweeps and occupancy sensors are the cheapest measures in the building.
Start here
Add or fix the lighting sweep / schedule in the lighting control panel or BAS; check for zones on manual override.
Detected 19 events totalling 65 h between Jan 20, 2025 and Aug 27, 2025. Estimated savings: $521 a year (3,363 kWh), confidence high.
Show the math+
Basis. priced directly from the meter (see assumptions)
Inputs
Input
Value
Source
Days of data
365 days
span of the fault's channels
Blended electricity rate
0.155 $/kWh
from bills
Steps
Step
Result
Formula
Wasted energy over the data window
3,362.8 kWh
priced directly from the meter (see assumptions)
Annualization factor
1 ×
365 / 365 days
Annualized energy
3,363 kWh/yr
wasted kWh × factor
Annual savings
521 $/yr
annualized kWh × $0.155/kWh
Assumption. 3736 kWh of lighting above the unoccupied norm over 65 h × 90% avoidable = 3363 kWh
Likely range $416 to $625 a year.
Evidence: the main meter around the longest event, the events shaded. Whole-building demand around the longest event (15-minute readings, kW); the shaded spans are the flagged events.
Demand ratchet penalty on the main meter: 2 month(s) billed above measured demand
$556a year · Medium confidence
The schedule ratchets billing demand to 80% of the trailing peak. In 2 of 12 months the main meter was billed for more kW than it used (worst: 260 kW billed vs 226 kW measured, 2025-12) — about $556/yr. One high-peak month costs money for a year; peak-limiting during the ratchet-setting season removes it.
Show the math+
Basis. Months billed on the ratchet floor pay for demand that never occurred; trimming the annual peak lowers the floor.
Inputs
Input
Value
Source
Σ (billed − measured) kW across floor-bound months
47.3 kW·mo
simulated bill (conservative: measured proxied by the month's peak)
Demand rate
11.75 $/kW
rate schedule GSDT — General Service Demand, Time of Use
Steps
Step
Result
Formula
Observed ratchet penalty
556 $
penalty_kw_months × demand_rate
Annualized penalty
556 $/yr
penalty_usd × 12 / 12
Assumption. Fully avoiding the penalty requires holding the annual peak below the ratchet-setting month — see the demand chart.
The year as a daily mean of the main meter, the months billed on the ratchet floor shaded.
7. MET-03 · Building · low
After-hours operation
$188a year · High confidence
What it means
Building demand is well above baseload during hours the schedule says it should be off.
Why it matters
Every hour over baseload is paid-for conditioning of an empty building. Together with MET-01, the single biggest controllable energy story in most commercial buildings.
Start here
Check BMS schedules, tenant override panels, after-hours request logging, and HVAC zone schedules.
Detected 7 events totalling 27 h between Jan 20, 2025 and Aug 18, 2025. Estimated savings: $188 a year (1,217 kWh), confidence high.
Show the math+
Basis. priced directly from the meter (see assumptions)
Inputs
Input
Value
Source
Days of data
365 days
span of the fault's channels
Blended electricity rate
0.155 $/kWh
from bills
Steps
Step
Result
Formula
Wasted energy over the data window
1,216.6 kWh
priced directly from the meter (see assumptions)
Annualization factor
1 ×
365 / 365 days
Annualized energy
1,217 kWh/yr
wasted kWh × factor
Annual savings
188 $/yr
annualized kWh × $0.155/kWh
Assumption. 1521 kWh above the normal unoccupied load over 27 h × 80% avoidable = 1217 kWh
Likely range $151 to $226 a year.
Evidence: the main meter around the longest event, the events shaded. Whole-building demand around the longest event (15-minute readings, kW); the shaded spans are the flagged events.
Baseload power during unoccupied hours is higher than expected for the building's size and type.
Why it matters
Baseload runs every hour of the year. Even a small fraction of avoidable baseload (lighting circuits, plug loads, HVAC override) adds up to a measurable annual bill.
Start here
Walk the building during unoccupied hours. Check for lighting circuits left on, plug-load schedules, and HVAC overrides that never expired.
Detected 5 events totalling 8,787 h between Jan 1, 2025 and Jan 1, 2026. Estimated savings: $99 a year (642 kWh), confidence high.
Show the math+
Basis. priced directly from the meter (see assumptions)
Inputs
Input
Value
Source
Days of data
365 days
span of the fault's channels
Blended electricity rate
0.155 $/kWh
from bills
Steps
Step
Result
Formula
Wasted energy over the data window
642.5 kWh
priced directly from the meter (see assumptions)
Annualization factor
1 ×
365 / 365 days
Annualized energy
642 kWh/yr
wasted kWh × factor
Annual savings
99 $/yr
annualized kWh × $0.155/kWh
Assumptions. 1285 kWh above the normal night load across 4 flagged nights × 50% achievable the nights are compared with the same weekday/weekend nights of the surrounding four weeks
Likely range $80 to $119 a year.
Evidence: the main meter around the longest event, the events shaded. Whole-building demand around the longest event (15-minute readings, kW); the shaded spans are the flagged events.
29% of the main meter's bill is demand, riders, fixed charges and taxes
Only 71% of the simulated bill is energy at the metered rate — the rest is demand ($37,275), riders ($0), fixed charges and taxes. kWh-reduction measures alone cannot touch that share; demand management and tariff options matter as much as efficiency here.
The year as a daily mean of the main meter.
After the fix
How a saving gets verified
A report prices what is wrong. After the fix, the same meter proves what changed: a baseline model of the year before, the actual year after, and the gap between them with its confidence band. The full method is on the verified-savings page.
After the fix
Then the meter proves it.
A baseline model learns how the building uses energy from its own meter and the weather. Once a fix is made, the model says what the building would have used and the meter says what it did. The gap is the saving, reported with its uncertainty.
15,604 kWh verified over 100 days. 95% confident: 14,627 to 16,581 kWh. The saving built into this example was 15,620 kWh.
A report like this takes a year of interval data and the bills; the BAS trends, if you have them, add the equipment chapters. Your first building is free, and an engineer reads the result before you do.
Screening-level estimates; confirm with measurement and verification before capital commitment. Rate 0.1548 $/kWh (blended rate from 14 electric bills); demand 11.75 $/kW-month; timezone America/New_York.
