Energy units: IP and SI reporting (provisional, 0.92)

CAMBER reports energy in the unit system a config chooses: kBtu for IP or kWh for SI (issue #69). This covers energy, demand and EUI only. Temperatures, pressures and flows are unchanged, and the rules and detectors keep their internal IP units, so no detection result moves.

{"site": "Example office", "units": {"system": "ip"}, "...": "..."}

System	Energy	Demand / power	EUI
`ip`	kBtu	kBtu/h	kBtu/ft2/yr
`si`	kWh	kW	kWh/m2/yr

Without a units block nothing changes. Every output stays in the meter's own unit, byte for byte as before 0.92. units.area (ft2 or m2) states the unit a config's floor areas are in. It defaults to ft2 for IP and m2 for SI. An unknown system or key is an error.

The module is camber.energy_units (provisional: names and signatures may change in a minor release).

Conversion factors

The canonical internal unit is the kWh. Every conversion goes through it, so any pair of units round-trips to float precision. The factors below are exact by definition. The rounded forms are the usual printed values.

Quantity	Factor	Basis
1 Btu (International Table)	1055.05585262 J, exact	NIST SP 811, Appendix B.8
1 kWh	3.6 MJ, exact	definition of the watt-hour
1 kWh	3.41214163 kBtu (printed 3.412142)	the two lines above
1 kBtu	1.05505585 MJ (printed 1.055056)	the IT Btu
1 therm	100,000 Btu = 100 kBtu	the IT therm
1 MMBtu = 1 dekatherm	1,000 kBtu = 10 therms
1 ton-hour	12,000 Btu = 12 kBtu	refrigeration ton
1 ton	12,000 Btu/h = 12 kBtu/h
1 MBH	1 kBtu/h
1 GJ	1,000 MJ = 277.78 kWh
1 ft	0.3048 m, exact	so 1 m2 = 10.7639 ft2 and 1 m3 = 35.3147 ft3
1 lb	0.45359237 kg, exact

The US therm is defined on the 59 F Btu (105.4804 MJ), 0.024% below the IT therm. CAMBER uses the IT therm throughout. EUI converts as 1 kBtu/ft2/yr = 3.15459 kWh/m2/yr.

The 3.412 in `bps.EUI_FACTORS_KBTU`

camber.bps.site_eui has always converted electricity at 3.412 kBtu/kWh. That is the exact 3.412142 rounded, and it is 0.004% low. It is kept, so existing EUI outputs do not move. It is also ENERGY STAR Portfolio Manager's standard multiplier, and so is its 100 kBtu per therm (see Energy conversion factors; a test keeps them equal). The unit-aware bps.site_eui_units uses the exact factors. On a building that is all electric, the two differ by 0.004% of the EUI (108.24 against 108.2428 kBtu/ft2/yr in the test example). Its other rows (100 kBtu per therm, 12 kBtu per ton-hour) are already exact.

Units the parser reads

parse_unit ignores case, spaces and -, _ and ·, and reads the common spellings:

Energy: kWh (kilowatt-hours), MWh, GWh, Wh, Btu, kBtu, MMBtu (dekatherm, Dth), therm (therms, thm), ton-hour (ton-hr, ton·h), kJ, MJ and GJ.
Power: kW, MW, W, Btu/h, kBtu/h (MBH), MMBtu/h and ton (tons).
Gas volume: ft3 (cf), CCF, Mcf and m3. A bare Mcf is a thousand cubic feet, the US gas-utility reading. MMcf is read only through a factor set.
Steam mass: lb, klb and kg.

It refuses:

an unknown unit;
an ambiguous one. MBtu and Mlb are ambiguous because M means a thousand in US utility usage and a million in SI. Write kBtu or MMBtu, and klb;
a unit of the wrong kind, such as a bare ton where energy is expected (it suggests ton-hour).

Gas by volume and steam by mass have no default. A volume converts only with an explicit heat_content, such as "10.37 therm/Mcf", "1037 Btu/ft3" or {"value": 1.037, "unit": "MMBtu/Mcf"}. A steam mass converts only with an explicit enthalpy, such as "1000 Btu/lb". Use the heat content printed on the gas bill, or the supplier's value.

from camber.energy_units import convert, convert_rate, UnitSystem

convert(1.0, "therm", "kWh")  # 29.3071
convert(120.0, "Mcf", "kBtu", heat_content="10.37 therm/Mcf")  # 124,440
convert_rate(1.20, "therm", "kWh")  # $/therm -> $/kWh
UnitSystem.of("si").eui_factor("kBtu/ft2/yr")  # 3.15459

Where the system applies

M&V savings, bands and SEP results

A config mv entry's reported quantities are converted. The fits stay in the meter's own unit, so R², CV(RMSE), NMBE, the savings percentage and SEnPI are the same in every system, and the IP and SI savings agree after conversion. What is converted:

the savings, projections, measured totals and bands;
the adjusted figures;
the chain links and their SEP terms (the variance terms by the factor squared);
the ledger's resolved amounts;
the waterfall;
the auto sensitivity table.

The model coefficients and an indicator's per-day rate stay in the meter's unit, which the finding names.

Each converted finding carries energy_unit (kBtu or kWh), meter_unit (the unit the fits are in) and unit_system, and its summary names the unit after every energy number:

Gas meter: avoided energy 554,023 kBtu (16.3%) ± 85,632 kBtu at 90% over 365 reporting days ...

Trended meters. An mv entry on trended equipment must name its meter's rate unit under a unit system, for example "units": "kW" (or Btu/h, kBtu/h, MBH, tons). The daily energy is its hourly integral: kWh, Btu, kBtu or ton-hours. An entry without units is an error once a system is set.
Trended gas by volume (0.93). A gas meter trended as a volume flow names it: "units": "cfh" (or CCF/h, Mcf/h, m3/h). The daily value is then a volume (ft3, CCF, Mcf or m3), and under a unit system it converts with mv[].heat_content ("1037 Btu/ft3"), or opt-in with the units.factor_set heat content of mv[].meter_type (default natural_gas). With neither, the run stops: there is no default. A caveat names the heat content used, and a factor set also adds the energy_factor metric. Without a system the fits and savings stay in the meter's volume, as before. The meter's flow maps to the entry's role (by default energy_rate).
The chain report (0.93). camber mv report follows units.system too. Its page, its JSON (units on each meter) and its CUSUM give energy in kBtu or kWh, the models stay in the meter's unit, and the entry's units is required under a system, as on the run path.
Billing entries. The bills' unit comes from bills.units or the file's units column, and more than one unit in the column is an error. Spellings of one unit (kWh, kwh, kilowatt-hours) count as one unit since 0.93, reported by its canonical name; a spelling that does not parse is compared ignoring case. Under a unit system the unit must parse. bills.heat_content is required for gas billed by volume (Mcf, CCF, m3), and bills.enthalpy for steam billed by mass. Both are validated whenever they are given. Opt-in, a units.factor_set supplies them instead (see Energy conversion factors). Without a system, bills in Mcf are fitted and reported in Mcf, as before. Bills in a bare Mcf carry a caveat under a system (see The "M" problem). BillingSeries.converted(unit, heat_content=...) converts a series in code.

SEP primary energy

Annex B multipliers apply to delivered energy. sep.primary_energy and sep.aggregate_energy_types take units= (the delivered unit of every energy type), heat_content= / enthalpy= (per type, for volumes and steam) and energy_unit= ("kWh" by default, or "kBtu"). Each type is converted to that energy unit first, and the multiplier is applied after. A type without a unit is refused, and so is an ambiguous unit or a volume without a heat content. The result carries unit and delivered_units, and a caveat names each heat content used. Without units the amounts must already share one energy unit, as before.

from camber.mandv.sep import primary_energy

primary_energy(
    {"grid_electricity": 1_000_000, "natural_gas": 1_200},
    units={"grid_electricity": "kWh", "natural_gas": "Mcf"},
    heat_content={"natural_gas": "10.37 therm/Mcf"},
    energy_unit="kBtu",
)

EUI

bps.site_eui_units(energy, units, area, area_unit="ft2" | "m2", system="ip" | "si") returns site EUI in kBtu/ft2/yr or kWh/m2/yr from amounts in their own units. The area unit is required. In a config, report.benchmark takes an optional unit (the unit its EUIs are stated in, kBtu/ft2/yr by default). Under a unit system both EUIs are converted to the system's EUI unit, and the audit report labels them with it.

The fleet report (0.93) takes build_fleet_report(..., eui_unit="kBtu/ft2/yr", units=None): eui_unit states the unit the buildings' EUIs and the peer median are given in, and units ("ip", "si" or a UnitSystem) reports them in that system's unit. Percentiles and the percentage against the median do not change. FleetReport.eui_unit labels the text and HTML, and the agent context's fleet facts name the same unit. camber fleet passes the configs' shared unit system; configs that disagree keep kBtu/ft2/yr, with a note.

Cost and carbon

Per-unit prices and emission factors convert with convert_rate. For example, convert_rate(5.30, "therm", "MMBtu") gives 53.0 kg CO2e/MMBtu.

A config price block also accepts a rate in any unit:

"price": {"electricity": {"rate": 95.0, "per": "MWh"},
          "gas": {"rate": 8.5, "per": "Mcf", "heat_content": "10.37 therm/Mcf"}}

This is converted to the electricity_per_kwh / gas_per_therm the cost estimators use. The plain electricity_per_kwh / gas_per_therm keys are read as before.

Emission factors have the same per-unit form (0.93). A factor passed to carbon.emissions(..., factors=) may be {"rate": 53.06, "per": "MMBtu"} (with heat_content for a factor per gas volume). carbon.factor_per converts it to the unit the fuel key names: natural_gas_therm gives 5.306 kg CO2e/therm. A key that is not an energy unit (fuel_oil_gal) takes a factor only in its own unit. No greenhouse-gas factor set is bundled (see Not converted). fault_economics.annotate_costs(..., units="ip" | "si") adds waste_energy (both fuels' site energy) and energy_unit beside the existing waste_kwh / waste_therms.

Energy conversion factors

camber.energy_factors (provisional) holds published conversion factor sets: the multipliers that turn a billed quantity (cubic feet of gas, gallons of oil, pounds of steam, tons of coal) into energy. Each set is a JSON file in the package, transcribed from its source and pinned to it by URL, edition, retrieval date and the source file's sha256. Every entry is listed in the factor tables.

The ENERGY STAR set

energy_star_thermal_2015 is the ENERGY STAR Portfolio Manager technical reference Thermal Energy Conversions (U.S. EPA, August 2015), retrieved 2026-09-28 from https://portfoliomanager.energystar.gov/pdf/reference/Thermal%20Conversions.pdf. It is a work of the U.S. Government, and the numbers are transcribed as printed.

Figure 2 (quick reference) gives the standard multipliers between kWh, MWh, kBtu, MMBtu and GJ, the same for both countries.
Figure 3 gives the multiplier to kBtu and the heat content for every unit Portfolio Manager accepts, for 17 meter types: electricity, natural gas, fuel oil No. 1, No. 2, No. 4 and No. 5 & 6, diesel, kerosene, propane, district steam, hot water and chilled water, anthracite and bituminous coal, coke, wood and "other". That is 105 unit rows per region, 210 entries.
Regions. US (U.S. property assumptions) and CA (Canadian property assumptions). The US heat contents come from the EPA Greenhouse Gas Reporting Rule, 40 CFR 98, subpart C, Tables C-1 and C-2. The Canadian fossil-fuel heat contents come from Statistics Canada's Report on Energy Supply and Demand (Text Table 1.1, 2009). District steam uses the International District Energy Association's 1,194 Btu/lb for both. For example, natural gas is 1,026 Btu/cf (US) and 1,031.43 Btu/cf (CA).
Source discrepancies. Five printed multipliers disagree with the table's own heat content beyond rounding: US natural gas per cubic metre (36.303, where 1,026 Btu/cf gives 36.233), the three Canadian propane liquid rows (they follow 0.090809 MBtu/gallon; the printed heat content reads 0.09089), and US wood per tonne (15,857, which is 17,480 divided by 1.10231 instead of multiplied). They are kept as printed, because they are what Portfolio Manager applies. Each is marked in the file, and using one raises a caveat.

from camber.energy_factors import factor_sets, factor_for, to_kbtu

factor_sets()  # ['energy_star_thermal_2015']
to_kbtu(120, "kcf", "natural_gas", factor_set="energy_star_thermal_2015", region="US")  # 123,120
to_kbtu(500, "gallons", "fuel_oil_2", factor_set="energy_star_thermal_2015", region="CA")
factor_for("klb", "district_steam", factor_set="energy_star_thermal_2015", region="US").describe()

to_kbtu(value, unit, meter_type, *, factor_set, region) takes a meter type by its key (natural_gas), its printed name ("Fuel Oil (No. 2)") or an alias (gas, steam, fuel_oil). A unit the set does not list for that meter type is an error that lists the units it does. factor_for returns the entry with its heat content, footnotes and caveats. to_kbtu issues each caveat as an EnergyFactorWarning.

The "M" problem

Mcf: a thousand or a million cubic feet?

ENERGY STAR writes M for million and k/K for thousand: its Mcf is a million cubic feet (1,026,000 kBtu), its Kcf a thousand, and its MBtu a million Btu (the source's own note 5). Many US gas utilities write M for thousand: their Mcf is a thousand cubic feet, and a million is MMcf. The two readings differ by 1,000x.

CAMBER keeps one reading for bare strings everywhere. A bare Mcf is a thousand cubic feet, as camber.energy_units reads it, and MBtu and Mlb are refused as ambiguous. Each factor set records its own convention (conventions.M), and its entries use unambiguous keys: kcf and MMcf, klb and MMlb, and MMBtu. With a factor set these labels are accepted: kcf, thousand cubic feet, MMcf, million cf, MMBtu, klb, MMlb and million lb.

Whenever a bare Mcf is converted, in to_kbtu or in a billing entry under a unit system, with a factor set or with an explicit heat content, a caveat names the conflict. If the bill follows ENERGY STAR's convention, the energy is 1,000x too low. To remove the caveat, write kcf or MMcf.

In a config (opt-in)

"units": {"system": "ip", "factor_set": "energy_star_thermal_2015", "region": "US"}

factor_set and region go together. Without them nothing changes: a gas volume still needs bills.heat_content, and steam by mass bills.enthalpy. With them, a billing entry in a unit that camber.energy_units cannot convert on its own is converted with the set's multiplier to kBtu, and then to the system's energy unit. Such units are volumes and masses without an explicit heat content, and gallons, litres, tons, tonnes, kcf, MMcf and MMlb.

bills.meter_type names the fuel, for example "fuel_oil_2", "propane" or "coal_bituminous". Without it, a gas volume is read as natural_gas and a mass as district_steam, as camber.energy_units reads them, and a caveat says so. Gallons and tons need it.
An explicit heat_content or enthalpy always wins. The set is not used then.
Energy units (kWh, therms, MMBtu, ton-hours) keep the exact factors above. An electric or therm meter reports the same with or without a factor set.
Every finding of the entry records the factor as energy_factor (the set, region, meter type, printed unit, multiplier, heat content and footnotes), and carries a caveat such as Converted with energy_star_thermal_2015 (..., 2015-08), US, Natural Gas: 1 Kcf (thousand cubic feet) = 1,026 kBtu, heat content 1,026 Btu/cf.

Exact and rounded factors

camber.energy_units converts with exact definitions: 1 kWh = 3.412142 kBtu. A published set carries its publisher's numbers, which are often rounded: ENERGY STAR converts a kWh at 3.412 kBtu and a GJ at 947.817 kBtu. Which one applies depends on the path. to_kbtu with a set uses the set's numbers. The default path, and energy-unit bills under a factor set, use the exact factors. bps.EUI_FACTORS_KBTU keeps its historical values. Its electricity (3.412), gas (100 per therm) and chilled-water (12 per ton-hour) rows equal ENERGY STAR's. Its propane (91.6 per gallon) and fuel-oil (138.7 per gallon) rows are other published values, within 0.5% of ENERGY STAR's 92 and 138, and are unchanged.

Adding or updating a set

A new set is a new JSON file in camber/energy_factors/. It needs no code change, unless it uses a physical unit outside camber.energy_factors.UNIT_KEYS. Examples are an EIA or Statistics Canada update, an ASHRAE table, site-to-source factors, or emission factors with their own kind and output_unit. A new edition of an existing source is a new file, so that results citing the old one stay reproducible.

scripts/energy_factors_refresh.py validates a draft (--validate file.json) and checks a set's source against its pinned sha256 (--check). It re-pins the sha256, retrieval date and edition (--pin ... --edition ... --write) and regenerates the factor tables (--docs --write). Its docstring is the checklist. On load, every set is validated: the schema, the unit keys, the footnote references, and that each multiplier equals heat content x unit size within the printed precision. A row the source prints inconsistently must be marked a source_discrepancy.

Unit-scale plausibility

camber.unit_scale (provisional, 0.92, #71) asks of every meter whether its billed quantities are right as given, or 1,000x off. A unit prefix read the wrong way passes every schema check and moves every result by a factor of 1,000: a steam utility that prints MLb for thousands of pounds, a bill in MMcf entered as Mcf, an export in MWh labelled kWh. It usually surfaces only when an EUI comes out impossible and someone works backwards.

For the three candidate scales (the quantities x0.001, x1 and x1000) the check collects evidence, combines it by a fixed rule and returns a UnitScaleCheck: the most likely scale, a confidence (high, medium, low), a status and a readable explanation. It never corrects anything. A correction is an explicit scale_override with a reason, recorded on every result.

from camber.unit_scale import check_bills

chk = check_bills(bills, unit="MMlb", fuel="steam", area=200_000, property_type="office")
chk.status, chk.scale, chk.confidence  # ('implausible', 0.001, 'high')
chk.explanation

bills is a BillingSeries or a table with start, end (exclusive unless end_inclusive=True) and quantity, and optionally cost (the invoiced $), demand (billed kW), read_start / read_end / multiplier and hdd. check_series does the same for an interval meter (energy per interval, summed by month), and check_eui for a stated EUI.

Evidence

Item	Weight	Compares
`tariff`	4	The bill recomputed at each scale under a `Tariff` or a URDB rate (`urdb=`, e.g. from `interop.openei.fetch_urdb_rate`), per bill, with `tariff.validate_bill` (15% tolerance). The fixed charge does not scale. Electricity only.
`price`	3	The implied $/MMBtu (all charges divided by the energy) against the state's EIA commercial average (opt-in) or the bundled price bands.
`eui`	2 or 3	The annualised site EUI of the meter, and of the building when the other meters' kBtu are given, against the property type's ENERGY STAR median x policy factors (2), or against hard bounds no building crosses (3).
`weather_intensity`	2	Heating fuel per HDD65 per ft2 (plausible 0.3-150 Btu/ft2/degF-day), and the peak bill's average heating power per ft2 (plausible up to 150 Btu/h/ft2; for steam also shown as lb/h).
`load_factor`	2	kWh against billed kW. A load factor above 1 is impossible.
`meter_reads`	2	The billed quantity against (end read - start read) x multiplier.
`continuity`	--	A ~1000x step in the series (see below).

The combination rule

A scale is ruled out when the evidence against it includes an item of weight 3 or more, or adds up to 4 or more, and the heaviest item against it outweighs the heaviest item for it. So a tariff match (4) beats a hard EUI bound (3). A price band (3) beats any EUI policy verdict (2). Equal weights on both sides rule nothing out. EUI and weather intensity together (2 + 2) rule a scale out only when no price or tariff evidence supports it.
The chosen scale is the one not ruled out with the highest net score (support minus against). A tie goes to x1, the reading as billed.
The status is implausible when x1 is ruled out, or the series has a ~1000x step. It is uncertain when another scale scores higher but x1 is not ruled out, plausible when x1 is chosen, and insufficient when no item had an opinion.
The confidence is high when every other scale is ruled out and either the tariff decides it or two items of weight 2 or more support the chosen scale. It is medium when every other scale is ruled out, and low otherwise.

Decisive price evidence is enough on its own. The bands are built so that a correctly scaled bill priced anywhere inside its band puts both 1,000x readings beyond the implausible bounds. When every item against x1 depends on the floor area, the explanation says to check the area too.

Continuity. A step between two bills counts when both sides are steady (two or more bills within 10x of each other, not one repeated placeholder value) and about 1,000x apart (10^2.5-10^3.5), and when the bills after it are also ~1000x their year-earlier bills. When there is no earlier year, it counts only for electricity, because a heating or chilled-water meter can drop 1,000x every season; the check notes that it saw the change. A run of year-over-year ratios near 1,000 marks a step on its own. The series is split at each step and each segment is judged separately.

References (CAMBER screening policy)

The bands and factors are CAMBER screening policy, not a standard. They are wide on purpose, because the error they look for is a factor of 1,000, and a correctly billed laboratory, hospital or data centre must stay inside them. Both sets are camber.energy_factors files, validated on load (get_reference_set, factor_sets("price_band")).

camber_price_bands_2024 (kind price_band), implied $/MMBtu:

Fuel	Plausible	Implausible below / above
electricity	5-300	0.35 / 3,000
natural gas	1.5-150	0.2 / 1,000
district steam, hot water	4-200	0.25 / 3,000
district chilled water	3-200	0.25 / 2,500
fuel oil (all grades), propane	5-150, 5-200	0.2 / 4,000, 0.25 / 4,000

Sources: the EIA Open Data API's 2024 state commercial electricity prices (7.19-38.18 cents/kWh) and natural gas prices (5.94-36.73 $/Mcf), EIA weekly heating oil and propane prices (2024-25), and CBECS 2018 Tables C1 and C2 (sha256 pinned), whose all-building averages are $29.2 per MMBtu for electricity, $7.1 for gas, $16.3 for fuel oil and $14.1 for district heat. No public national series exists for district chilled water, so its band is policy around the cost of chilled water from an electric plant, and is marked unverified. - energy_star_us_median_eui_2024 (kind eui_reference): the national median site and source EUIs of every Portfolio Manager property type, transcribed from the ENERGY STAR technical reference U.S. Energy Use Intensity by Property Type (August 2024, mostly CBECS 2018; sha256 pinned). The policy block around it: - A building's total EUI is plausible within median / 8 to median x 6, and implausible outside median / 60 to median x 25. - A single fuel is implausible above median x 25. - Hard bounds hold for any building: electricity 0.1 to 10,000 kBtu/ft2/yr (10,000 is a continuous 330 W/ft2, beyond any data centre), heating fuels up to 3,000, and the total 0.3 to 10,000. - Data centres (printed as a PUE), parking and manufacturing have no median, so only the hard bounds apply. The table gives medians, not percentiles; the CBECS percentile distributions are not used.

Regional prices (EIA, opt-in)

price_source="eia" with state="NY" compares the implied price with the state's average commercial price from EIA Open Data API v2 (camber.interop.eia.fetch_state_price), as a band of average / 4 to average x 4 (implausible beyond / 40 and x 40). It uses electricity/retail-sales for electricity and natural-gas/pri/sum (process PCS) for gas. It needs a free key in EIA_API_KEY, or an injected eia_transport. The request carries only the state and the months. The key is added by the live transport and never enters the cache (eia_cache_dir, keyed by the key-free URL) or a result. eia_offline=True reads only the cache. With no key, no network or a cache miss, the check falls back to the bundled bands and says so in a note.

Where it runs

Billing M&V (mv entries with bills). The check runs whenever the bills name a unit. It reads the bills file's cost, demand, read_start, read_end and multiplier columns when they are present, and pairs each bill with the entry's outdoor temperature. An optional bills.scale_check block adds the rest:
```
"bills": {
  "file": "steam.csv", "energy": "usage", "units": "MMlb",
  "scale_check": {"fuel": "steam", "area": 200000, "property_type": "office",
                  "cost": "amount", "price_source": "eia", "state": "NY",
                  "tariff": {"urdb_file": "rate.json"}, "on_implausible": "decline"},
  "scale_override": {"factor": 0.001, "reason": "the utility's MLb is thousands of pounds"}
}
```
The other keys are area_unit, demand, read_start, read_end, multiplier, eia_cache_dir, eia_offline, other_site_kbtu_per_year and enabled ("scale_check": false turns the check off). A tariff is a Tariff's fields, {"urdb_file": ...} or {"urdb_label": ...} (fetched from OpenEI: opt-in, and only the rate label is sent).

When the bills are implausible as given, the entry gets a unit_scale finding (warn, with the whole check in metrics["unit_scale"]). By default its M&V is declined; with "on_implausible": "warn" it is fitted as billed. A plausible, uncertain or unjudgeable check adds nothing, so existing outputs do not change. - The override. bills.scale_override multiplies every billed quantity before the fit. Each finding records it as metrics["scale_override"] and carries a caveat naming the factor and the reason. The check then judges the corrected quantities. - EUI and BPS. bps.site_eui_plausibility(energy_by_fuel, area, ...) judges each fuel behind an EUI as a year-long bill (with cost_by_fuel for price evidence). A report benchmark whose site_eui is implausible as given adds a unit_scale finding for benchmark; the optional benchmark.property_type names its ENERGY STAR type. - Catalog ingest. The BDG2 adapter judges every electricity, chilled-water, steam, hot-water and gas meter, using the metadata's floor area and primary use and the site's weather. An implausible meter is an ingest warning. The data are never rescaled.

Validation

Synthetic bills (tests/test_unit_scale.py): each injected 1000x error is caught, in the expected direction. The cases are steam billed MLb for thousands of pounds and entered as million pounds, gas in MMcf for Mcf and the reverse, MWh labelled kWh (with a tariff recompute and a URDB rate), and fuel oil. None of the correct cases is flagged: a 1.2 million ft2 hospital, a 2,000 ft2 shop, a laboratory at 420 kBtu/ft2/yr of steam, a restaurant at 900 kBtu/ft2/yr of gas, a data centre at 2,500 kBtu/ft2/yr, a small office, a warehouse and a mild-climate hot-water meter.
BDG2 cleaned meters (2,835 meters; the metadata floor area and primary use, the site weather):

Meter Meters Flagged

electricity 1,572 4 (0.25%)

chilled water 550 84

hot water 185 74

steam 351 2

gas 177 4
- Eagle's chilled water. The check catches the documented 1,000x error at Eagle (eagle-chilled-water-1000x in the catalog) on its own, with no knowledge of it: 82 of the 87 meters are implausible with x0.001 most likely. Four more are uncertain leaning x0.001, and one reads zero.
- Eagle's hot water. 58 of its 60 hot-water meters are flagged the same way, x0.001 most likely, at a median 830x the same building's electricity. The catalog records it as eagle-hot-water-1000x (annotated, not rescaled).
- Other flags. 28 of the 2,539 meters outside Eagle are flagged. Most are thermal meters at thousands of kBtu/ft2/yr (15 at Fox, all hot water). Fox's floor areas are consistent and the same buildings' electricity is ordinary, so the hot-water energy scale is at fault, but no single factor fits the site (fox-hot-water-scale in the catalog, annotated). The four electricity flags are meters that stopped reading or read a near-zero constant.

Not converted

Factor sets apply only to config billing entries, trended gas meters (0.93) and camber.energy_factors itself, not to SEP, EUI or price conversions, which still take an explicit heat content.
Temperatures, pressures, flows, and every rule's own metrics.
Greenhouse-gas factor sets (a follow-up). A camber.energy_factors kind for emission factors, such as EIA's CO2 coefficients per fuel and EPA eGRID's subregion output rates, was considered for 0.93 and deferred. eGRID is a yearly edition of per-subregion tables, and a site needs its subregion, which CAMBER does not yet record. Pinning that honestly (edition, sha256, the choice of subregion, and location-based against market-based reporting) is its own change. Until then, factors are given per unit to carbon.emissions, as above.

Meter	Meters	Flagged
electricity	1,572	4 (0.25%)
chilled water	550	84
hot water	185	74
steam	351	2
gas	177	4