Utility Split Estimation

Utility-Split Estimation

Estimate what fraction of each utility (electricity, gas, water) is paid by the landlord (owner) versus the tenant (resident) for one asset. Produce a per-fuel split with the evidence for each, labeled PRESUMED vs CONFIRMED, a confidence note, and the single question to ask the owner to confirm. Record the estimate as a verifier finding so it is adjudicable, and — when Audette is in play — flag that Audette's landlord-share settings must be updated to match the confirmed split.

Why this matters (do not skip, do not default)

Retrofit savings accrue only to the party that pays the bill. If residents pay 90% of the electric load, an electric-efficiency measure returns ~10% of its dollar savings to the owner — and its IRR collapses accordingly. The landlord share is therefore the savings basis for every measure's economics. A wrong or defaulted split silently mis-prices the entire measure plan. Never assume 100%-owner, never assume a round number, never carry a prior asset's split forward. Estimate it from this asset's evidence.

This is frequent, load-bearing work: it feeds decarb-plan P3 economics and any retrofit IRR. Run it whenever an asset's owner/tenant utility split is unknown, stale, or unconfirmed.

Core method (all asset types)

Work these four evidence sources in order, stopping to record what each yields. Sources lower in the list refine or confirm what higher sources presumed; a document or a bill that states responsibility outranks an inference from building form.

1. SEARCH THE ASSET'S OWN DOCUMENTS FIRST. Before any inference or web search, look in the asset's uploaded files — leases and lease abstracts, the Offering Memorandum (OM), PCAs, energy/water audits, operating statements (T-12), and utility bills frequently state utility responsibility outright ("resident pays electricity", "owner pays gas heating and hot water", a RUBS line item, an expense-stop clause). Use list_files / search_files then read_file / search_documents with queries like "utilities", "resident responsible", "tenant pays", "RUBS", "master metered", "individually metered", "expense stop", "recoveries". A document statement is the strongest evidence short of the owner confirming — mark those fuels CONFIRMED with the document cited.

2. Building form (asset-type-specific — see the module for this asset's type below). Metering configuration and central-plant layout set the presumed split when documents are silent. Example (multifamily): garden-style, individually-metered units → most electricity/gas billed directly to residents → low landlord share; mid/high-rise with central plants (central boiler/chiller, house-metered common systems) → higher landlord share. Building form yields a PRESUMED split, never a confirmed one.

3. Jurisdiction pass-through / RUBS regulations. Whether and how a landlord may bill utility costs back to residents is regulated locally (submetering rules, Ratio Utility Billing System allowances and caps, prohibitions on certain pass-throughs, tenant protections). This bounds what the split can be regardless of building form — a master-metered building in a RUBS-permitted jurisdiction may still pass most cost through; the same building where RUBS is barred leaves the cost with the owner. Search the reference library first, then web search; cite every rule with its source and URL.

4. Live leasing evidence. Pull the property's current listings/brochures (apartments.com, the property's own site, Zillow rentals, etc.) via web search. Listings routinely state "utilities included" or itemize "resident pays electric/gas/water" — this is real, current, market-facing evidence of what residents actually pay and often confirms or corrects the building-form presumption. Cite the listing and its URL.

Resolve per fuel, not per building. Electricity, gas, and water frequently split differently (e.g. residents pay in-unit electric, owner pays central gas heat + common-area water). Estimate each fuel independently against the evidence above.

Asset-type modules

Read the module matching this asset's property type before finalizing the building-form step. Each lists the determining factors and how they map to a presumed split.

If the asset type has only a stub, use the listed determining factors plus the core method to produce a best estimate, mark the result PRESUMED with low confidence, and be explicit in the owner question that the split is unconfirmed. Do not fabricate precision the module does not yet support.

Granularity rule (critical)

Consumption allocation (when distributing whole-property utility data across building models)

Never split evenly by default. Allocate in this order: (1) carve out identified common/amenity loads first (pool heater, clubhouse, exterior/corridor lighting — use the audit's end-use breakdown) and assign them to the building/line-item where they belong; (2) allocate the remainder across residential buildings weighted by GFA (adjust for known differences: floors, vintage, equipment); (3) state the allocation method with the upload. An even split is only acceptable when buildings are genuinely identical in GFA and use — and must still be labeled "GFA-weighted (identical buildings)".

Apply the split at the FINEST granularity the model supports — per building and per end-use — never as one blended property-wide percentage. A blended % misattributes savings both ways: in-unit measures get phantom landlord credit, and fully-landlord amenity measures get under-credited.

• Residential buildings with tenant-metered fuel: landlord share for that fuel = 0% in those building models. House-meter loads (corridor/exterior lighting, common HVAC) that live inside residential buildings keep a per-building landlord share for that fuel (typically electricity).
• Amenity/clubhouse/pool buildings (no tenants): landlord share = 100% for their fuels.
• Master-metered ≠ 100% landlord. A master-metered residential load is only 100% owner if the owner ABSORBS it. If the jurisdiction permits RUBS, assume the owner rebills up to ~90% → net landlord share ≈ 10% for that load, unless documents show a true gross lease / no rebill. Never set a master-metered residential load to 100% by default. **The RUBS assumption is CONDITIONAL on the jurisdiction permitting RUBS — confirm it in step 3 (cite the statute); if RUBS is BARRED, the owner bears the cost (100% on master-metered), not ~10%.**
• Solar under Virtual Net Metering (VNM): assume 80% of solar output value flows to the landlord (owner-captured) — but ONLY after confirming the state/utility actually offers VNM / aggregated NEM / community-solar export (check the reference library → PUC/utility tariff → cite the rule + URL). If only behind-the-meter (BTM) net metering is available (no virtual/export aggregation), solar value = BTM self-consumption offset only — owner-share on the owner-paid loads it displaces — NOT the 80% VNM credit. Never assume VNM without the jurisdiction check.
• If the model has NO separate amenity building (e.g. a property modeled as N identical residential buildings), set the tenant-metered fuel to 0% landlord across all modeled buildings and evaluate common-load measures (pool heater, clubhouse equipment) as standalone owner-paid line items outside the building models. Say explicitly which loads were handled this way.
• Only fall back to a blended % when the modeling tool cannot express per-building shares — and label the blend and its composition.

Output contract

Produce, for the asset:

Per-fuel split table — one row each for electricity, gas, water:

• Landlord share is the fraction of that fuel's annual cost the owner pays (this is the savings basis). If only a metering configuration is known and not a dollar split, state the presumed configuration and the share it implies, and mark it PRESUMED.
• Label each fuel independently: CONFIRMED = a lease/OM/audit/bill or the owner states it; PRESUMED = inferred from building form, jurisdiction, or listings.
• Evidence cites the exact source for that fuel (document + page/line, or listing URL, or the building-form + jurisdiction rule). No uncited splits.

Confidence note — one short paragraph: how strong the overall evidence is, which fuels are weakest, and what would move a PRESUMED fuel to CONFIRMED.

The single question to ask the owner — one precise, closed question that, once answered, confirms the presumed fuels. Example: "For each of electricity, gas, and water — are these individually metered and billed directly to residents, master-metered and paid by ownership, or billed back via RUBS? If RUBS, what allocation formula?" Tailor it to what is actually still open.

Persist to asset metadata (the CANONICAL record — read this FIRST on every run)

The resolved split MUST be written to a machine-readable field that survives across threads — otherwise every new session re-derives it and re-defaults to Audette's 100% landlord-share (the exact bug this skill exists to prevent). The verifier finding (below) is a QA/adjudication artifact and the register file is a human doc; neither is read by the economics. The canonical store is metadata.utility_split.

• WRITE it via update_asset_metadata(asset_id, { utility_split: {...} }) (an allowlisted key):

  utility_split = {
    elec: <owner fraction 0-1>, gas: <0-1>, water: <0-1>,   // net owner capture per fuel
    source: "gig-cad | lease | listing | building-form | audette-default",
    confidence: "confirmed | presumed",
    basis: "<one line: RUBS X% CAD / individually-metered / gross-master-absorbed / all-electric→gas 0 / VNM 80>",
    resolved_at: "<YYYY-MM-DD>"
  }
  

  If the tool response reports a REJECTED / not-in-allowlist key, the split did NOT persist — STOP and surface it (the platform allowlist must include utility_split).
• READ it FIRST. Before deriving a split — in this skill, in portfolio-analysis, and in decarb-plan — read metadata.utility_split; if present, USE it. Only run the four-source derivation when it is absent. NEVER default to 100% owner, and never read the split from Audette's landlord-share, when a stored utility_split exists (Audette re-defaults to 100% on model re-pull, which is why it can't be the source of truth).
• WRITE the split to BOTH stores on every run, where Audette is in play: metadata.utility_split (the canonical Soapbox record) AND Audette's per-building landlord-share (section below). Audette's own engine applies its landlord-share to decide which measure savings accrue to the owner, so a split that lives only in metadata while Audette sits at its ~100% default produces overstated owner savings, IRRs, and value creation (the exact failure this skill exists to prevent). Keep the two in sync — metadata.utility_split is the authoritative human record — but the split is NOT 'applied' until Audette reflects it. (You still never read the source-of-truth split from Audette — it re-defaults to ~100% on re-pull — but you MUST write the confirmed split into Audette.)

Record as a verifier finding (adjudicable)

Record the estimate so it can be adjudicated (mirrors decarb-plan P2 conflict handling). Call verifier__record_finding with:

• asset_id = the Soapbox asset id (not the Audette uid).
• kind: data-quality.
• severity: by materiality to the measure economics — high when the split is unconfirmed and drives a large electric or gas load (i.e. it swings IRRs materially), otherwise medium.
• evidence[]: the per-fuel splits with their labels and the source behind each.
• sources[]: every document, listing URL, and jurisdiction rule cited.
• A summary that states the per-fuel landlord shares and their PRESUMED/CONFIRMED labels.

Store the returned finding_id. In a decarb-plan engagement this finding is adjudicated at Gate 1 alongside the other baseline conflicts — do not duplicate an existing open split finding; reference it (see the Cortland gas-split finding pattern in decarb-plan P1/P2). Resolve it via verifier__resolve_finding once the owner confirms.

Audette landlord-share settings — MANDATORY write + read-back (not a 'note to self')

When Audette is in play, the confirmed per-fuel split MUST be written INTO Audette, because Audette's engine — not the Soapbox economics alone — computes which measure savings accrue to the owner. A split that lives only in metadata.utility_split while Audette sits at its ~100% default yields overstated owner savings, IRRs, and value. This is not optional and is not a flag-for-later.

Per building (Audette is building-level; a multi-building asset needs EVERY building set):

1. Look up the building model uid(s) for the asset (not the property uid).
2. Write default_landlord_share_electricity = and default_landlord_share_natural_gas = via the Audette building-edit tool.
3. Editing a building triggers a synchronous re-model that often exceeds the tool timeout; the write still commits in the background. Always read the building back and confirm the value persisted (retry any that rolled back). For many buildings, fire all edits first, then do one batch read-back pass rather than verify-each-inline.
4. Record in the engagement state which buildings were written AND verified.

Do NOT run or trust the decarb-plan / portfolio economics for an asset until its Audette landlord-share equals the confirmed split. If Audette is unreachable (connector down/expired), STOP and tell the user to reconnect — never fall back to Audette's default and never silently compute at 100%.