Lender ESG Benchmarking
Produce underwriting-grade ESG screening for capital-markets audiences: lenders sizing a
loan, CMBS underwriters assembling disclosure, insurers pricing risk. The output must be
auditable — every number carries its source and retrieval date.
Design principle: live data first, bundled data as fallback
This skill deliberately ships no benchmarking engine of its own. Peer statistics come
from MCP servers that already exist. Only if none are connected may you fall back to the
bundled approximate table, and the output must then say so.
Priority order for every benchmark number:
- BPD MCP (
bpd) — LBNL Building Performance Database, 1M+ real US buildings.
Query the peer group (building type + state or climate zone + vintage band) and read the
EUI distribution (median, 25th/75th percentile). Compute the subject's percentile
position within that distribution.
- ENERGY STAR Portfolio Manager MCP (
energy-star-pm) — the ENERGY STAR score
(1–100) is a national peer percentile, adjusted for weather and use. Report it as such.
- Bundled fallback —
references/national-median-eui.md (approximate national
medians). Flag clearly: source: approximate national median — connect BPD for peer-group percentile.
Workflow
1. Assemble asset facts
Collect for each asset (from the platform asset record, ESPM, utility data, or the user):
- Property type (map to BPD/ENERGY STAR property-type vocabulary), gross floor area
- Location (city, state → climate zone), year built / last major renovation
- 12-month site EUI (kBtu/ft²·yr) and fuel mix; weather-normalized if available
- ENERGY STAR score if benchmarked in ESPM
- Carbon intensity (kgCO₂e/ft²·yr) — use eGRID subregion factors (Arcadia MCP) or NREL
Cambium for forward years
Missing EUI is a blocker for percentile work — say what's missing and how to get it
(connect utility data via Arcadia/Nectar, or share the ESPM property). Never estimate a
subject building's EUI to fill the gap.
2. Peer benchmark
- Define the peer group explicitly and state it in the output: e.g. "Multifamily,
Colorado + Mountain census division, built 1990–2010, n=412 (BPD)".
- Report: subject EUI, peer median, subject percentile (lower EUI = better = lower
percentile; always label the direction), ENERGY STAR score if available.
- If peer count n < 30, widen the geography one step (state → census division → national)
and note the widening.
3. Lender ESG Card (one page per asset)
Structured card with these sections — omit a section only if no data source is connected,
and list omissions at the bottom:
| Section |
Content |
Source |
| Identity |
address, type, GFA, vintage |
platform |
| Energy |
site EUI, peer percentile + peer group, ENERGY STAR score |
BPD / ESPM |
| Carbon |
kgCO₂e/ft²·yr, grid subregion, CRREM stranding year if available |
utility + eGRID / CRREM |
| Regulatory |
applicable BPS laws, current + projected penalty exposure |
bps-compliance skill |
| Physical risk |
hazard ratings if a risk MCP is connected |
Moody's RMS / physrisk |
| Green financing |
program eligibility summary (see step 4) |
references + asset data |
| Flags |
material ESG risks with one-line rationale each |
derived |
Verdict line at top: Green / Neutral / Flagged with the single most decision-relevant
sentence. Render as markdown by default; offer the soapbox-report template for a branded
PDF when the user wants a client-facing deliverable.
4. Green financing eligibility
Check references/green-financing-programs.md for program criteria (Fannie Mae Green
Rewards, Freddie Mac Green Advantage, C-PACE availability, green bond / CMBS green
tranche framing). For each program output: eligible / not eligible / data gap, and the
requirements gap — what improvement or documentation would qualify the asset.
Program terms change; when a deal decision hinges on eligibility, verify current program
terms with a web search and cite the date.
5. Portfolio roll-up (collateral pools, loan books)
For multiple assets, produce a ranked screening table: asset, type, GFA, EUI, percentile,
ENERGY STAR, carbon intensity, BPS exposure ($), flags. Sort worst-percentile first.
Summarize: distribution of percentiles, total projected BPS penalty exposure, count of
green-financing-eligible assets, and the 3 assets that most drag the pool.
Rules
- Every number carries source + date. No source, no number.
- Never present the bundled fallback table as a peer percentile — it yields only
"above/below approximate national median".
- Percentile direction is stated explicitly every time ("22nd percentile — uses less
energy than 78% of peers").
- This is screening, not an energy audit: recommend
rapid-sustainability-risk for
acquisition-grade analysis and retrofit-advisor for measure-level CapEx planning.
- Run the verifier (if connected) on any card leaving for a client.
1---2name: lender-benchmarking3description: ESG benchmarking for capital markets — screen a property or collateral pool the way a lender, CMBS underwriter, or insurer would. Computes peer EUI and ENERGY STAR percentile within asset class + geography, produces a one-page Lender ESG Card per asset, checks green financing eligibility (Fannie Mae Green Rewards, Freddie Mac Green Advantage, PACE, green bond frameworks), and rolls assets up into a portfolio ESG screening table. Triggers on: "lender ESG card", "ESG for the lender", "CMBS ESG", "underwriting ESG", "peer benchmark", "EUI percentile", "how does this building compare to peers", "green financing", "Green Rewards", "green loan eligibility", "collateral ESG screen", "insurer ESG", "loan book ESG".4---56# Lender ESG Benchmarking78Produce underwriting-grade ESG screening for capital-markets audiences: lenders sizing a9loan, CMBS underwriters assembling disclosure, insurers pricing risk. The output must be10**auditable** — every number carries its source and retrieval date.1112## Design principle: live data first, bundled data as fallback1314This skill deliberately ships **no benchmarking engine of its own**. Peer statistics come15from MCP servers that already exist. Only if none are connected may you fall back to the16bundled approximate table, and the output must then say so.1718Priority order for every benchmark number:19201. **BPD MCP** (`bpd`) — LBNL Building Performance Database, 1M+ real US buildings.21 Query the peer group (building type + state or climate zone + vintage band) and read the22 EUI distribution (median, 25th/75th percentile). Compute the subject's percentile23 position within that distribution.242. **ENERGY STAR Portfolio Manager MCP** (`energy-star-pm`) — the ENERGY STAR score25 (1–100) *is* a national peer percentile, adjusted for weather and use. Report it as such.263. **Bundled fallback** — `references/national-median-eui.md` (approximate national27 medians). Flag clearly: `source: approximate national median — connect BPD for28 peer-group percentile`.2930## Workflow3132### 1. Assemble asset facts3334Collect for each asset (from the platform asset record, ESPM, utility data, or the user):3536- Property type (map to BPD/ENERGY STAR property-type vocabulary), gross floor area37- Location (city, state → climate zone), year built / last major renovation38- 12-month site EUI (kBtu/ft²·yr) and fuel mix; weather-normalized if available39- ENERGY STAR score if benchmarked in ESPM40- Carbon intensity (kgCO₂e/ft²·yr) — use eGRID subregion factors (Arcadia MCP) or NREL41 Cambium for forward years4243Missing EUI is a **blocker for percentile work** — say what's missing and how to get it44(connect utility data via Arcadia/Nectar, or share the ESPM property). Never estimate a45subject building's EUI to fill the gap.4647### 2. Peer benchmark4849- Define the peer group explicitly and state it in the output: e.g. "Multifamily,50 Colorado + Mountain census division, built 1990–2010, n=412 (BPD)".51- Report: subject EUI, peer median, subject percentile (lower EUI = better = lower52 percentile; always label the direction), ENERGY STAR score if available.53- If peer count n < 30, widen the geography one step (state → census division → national)54 and note the widening.5556### 3. Lender ESG Card (one page per asset)5758Structured card with these sections — omit a section only if no data source is connected,59and list omissions at the bottom:6061| Section | Content | Source |62|---|---|---|63| Identity | address, type, GFA, vintage | platform |64| Energy | site EUI, peer percentile + peer group, ENERGY STAR score | BPD / ESPM |65| Carbon | kgCO₂e/ft²·yr, grid subregion, CRREM stranding year if available | utility + eGRID / CRREM |66| Regulatory | applicable BPS laws, current + projected penalty exposure | bps-compliance skill |67| Physical risk | hazard ratings if a risk MCP is connected | Moody's RMS / physrisk |68| Green financing | program eligibility summary (see step 4) | references + asset data |69| Flags | material ESG risks with one-line rationale each | derived |7071Verdict line at top: **Green / Neutral / Flagged** with the single most decision-relevant72sentence. Render as markdown by default; offer the `soapbox-report` template for a branded73PDF when the user wants a client-facing deliverable.7475### 4. Green financing eligibility7677Check `references/green-financing-programs.md` for program criteria (Fannie Mae Green78Rewards, Freddie Mac Green Advantage, C-PACE availability, green bond / CMBS green79tranche framing). For each program output: eligible / not eligible / data gap, and the80**requirements gap** — what improvement or documentation would qualify the asset.81Program terms change; when a deal decision hinges on eligibility, verify current program82terms with a web search and cite the date.8384### 5. Portfolio roll-up (collateral pools, loan books)8586For multiple assets, produce a ranked screening table: asset, type, GFA, EUI, percentile,87ENERGY STAR, carbon intensity, BPS exposure ($), flags. Sort worst-percentile first.88Summarize: distribution of percentiles, total projected BPS penalty exposure, count of89green-financing-eligible assets, and the 3 assets that most drag the pool.9091## Rules9293- Every number carries source + date. No source, no number.94- Never present the bundled fallback table as a peer percentile — it yields only95 "above/below approximate national median".96- Percentile direction is stated explicitly every time ("22nd percentile — uses less97 energy than 78% of peers").98- This is screening, not an energy audit: recommend `rapid-sustainability-risk` for99 acquisition-grade analysis and `retrofit-advisor` for measure-level CapEx planning.100- Run the verifier (if connected) on any card leaving for a client.
soapboxbuild