Construction Costing

Construction Costing

You are estimating construction cost for one or more decarbonization measures — the capex (low/base/high) and opex_delta_yr (annual OpEx change, positive = OpEx rises, negative = OpEx falls) that feed the decarb-plan measure-screening economics, and that quality-review reads back when auditing a plan. This skill does not invent numbers: every base capex and opex_delta_yr comes from the seeded lookup table in references/cost-bases.md, keyed by archetype, climate, and size — not from LLM estimation or memory of "typical" retrofit costs.

Positioning: construction-costing is a narrow, mechanical costing method, not a full economic-feasibility engine. It produces the measure.cost object; decarb-plan owns the downstream NPV/IRR/payback math and landlord-share capture logic. Do not compute IRR or discounted paybacks here — that belongs to the plan's economics step, which consumes this skill's output.

Ground rules

1. No invented cost figures. Every capex/opex number traces to a row in references/cost-bases.md or the electrical service-capacity $/kW curve in that same file. If a measure has no matching row (archetype/climate/size combination not covered), say so explicitly and flag it for a human to add a row — do not guess a number to fill the gap.
2. cost-bases.md is seeded and provisional. Every dollar figure in that reference file is a PLACEHOLDER pending tuning by Christopher against real bids and utility tariffs. When you report a cost estimate to a user, note that the underlying basis is a seeded default, not a verified market price — never present these numbers as authoritative pricing.
3. opex_delta_yr sign convention is fixed: positive means OpEx rises (e.g., more expensive electricity displacing cheap gas), negative means OpEx falls. Never flip this sign when composing or comparing measures.
4. Electrical service capacity is never guessed away. Any fuel-switch / electrification measure that increases electrical demand must carry an explicit electrical service capacity assessment. When the actual service capacity is unknown, the upgrade cost is an honest range, and it must never collapse to a point estimate — see Step 3.
5. Every fuel-switch measure gets a non-switching sibling. Any measure_kind: fuel_switch row must also emit an efficiency_alternative — the standard high-efficiency, non-fuel-switching option (e.g., a condensing boiler instead of an ASHP plant) — so the plan can screen "switch fuels" against "just get more efficient at the same fuel."
6. Validate before handing off. Every measure.cost object this method emits must validate against skills/construction-costing/schema/measure-cost.schema.json via node scripts/validate-measure-cost.mjs before it is considered done.

Method

Step 1: Load the measure roster and the Audette model

Read the measure roster for the engagement (the candidate list produced upstream by decarb-plan's measure generation) and the corresponding Audette building model, specifically: end-use energy breakdown, peak electrical demand by end use, and — where available — existing electrical service capacity (panel schedule, service entrance rating, or utility interconnection headroom). If the roster or model is missing required fields (e.g., no peak demand data for a candidate fuel-switch measure), flag it rather than filling the gap with an assumption.

Step 2: Look up the parametric base per measure

For each measure, determine its archetype (multifamily, office, lab, mixed_use), climate zone band (cold, mixed, hot), and building size band (small, mid, large), then look up the matching row in references/cost-bases.md. That row gives you capex {low, base, high} and opex_delta_yr (again: positive = OpEx rises) scaled to a reference building size — scale linearly by GSF or unit count relative to that reference unless the table notes a different unit basis (e.g., "per 100 hoods" for fume-hood VAV).

If no row matches the archetype × climate × size combination, do not extrapolate silently from an unrelated row — surface the gap explicitly (e.g., "no cost-bases.md row for lab × hot × large; using the nearest mixed climate row as an approximation, flagged for review").

Step 3: Fuel-switch / electrification measures — the electrical service capacity model

For any measure that switches fuel (e.g., gas boiler → ASHP, gas furnace → VRF):

1. Compute demand_increase_kw — the delta between the new electrified end-use's peak electrical demand and the existing electrical peak demand at that panel/service entrance, from the Audette model.
2. Check whether actual electrical service capacity is known for this building (service_capacity_known: true/false):
   • Known (a real electrical survey, panel schedule, or interconnection study confirms available headroom): emit a point upgrade_cost (the single confirmed cost, which may be $0 if headroom covers the increase, or the confirmed upgrade quote otherwise) and flag: 'VERIFIED'.
   • Unknown: emit a range — upgrade_cost.low = 0 (best case: assume existing service has enough headroom, no service upgrade required) and upgrade_cost.high = the full new electrical service cost, computed from the $/kW (or $/A) parametric curve in references/cost-bases.md Section 5.1 as demand_increase_kw × region_$/kW. Set flag: 'UNVERIFIED'.
3. Never collapse an UNVERIFIED range to a point estimate — e.g., never average low/high into a single number, and never silently pick high (or low) as "the" cost when capacity is unknown. The range itself communicates real risk to the plan's economics step and to quality-review; collapsing it hides that risk.

Step 4: Always emit the efficiency alternative

For every fuel-switch measure, also produce the standard non-switching, high-efficiency option at the same archetype/climate/size (e.g., a condensing boiler replacement instead of an ASHP plant conversion — see references/cost-bases.md Section 3 for the boiler-replacement rows that pair with Section 2's electrification rows). Emit it as the efficiency_alternative object with its own capex (a single number — the alternative's base estimate, not a low/base/high spread) and its own opex_delta_yr. This lets the downstream plan compare "switch fuels" against "stay on the same fuel but get more efficient" on equal footing.

Non-fuel-switch measures (LED, DCV, fume-hood VAV, chiller optimization, envelope) do not need an efficiency_alternative — they already are the efficiency option.

Step 5: Emit the measure.cost object and validate

Assemble the full measure.cost object per measure, matching the shape defined in schema/measure-cost.schema.json:

{
  "measure_id": "central-ashp",
  "measure_kind": "fuel_switch",
  "cost": {
    "capex": { "low": 1400000, "base": 2100000, "high": 3200000 },
    "opex_delta_yr": 48000,
    "electrical_capacity": {
      "demand_increase_kw": 420,
      "service_capacity_known": false,
      "upgrade_cost": { "low": 0, "high": 1800000 },
      "flag": "UNVERIFIED"
    },
    "efficiency_alternative": {
      "measure": "high-efficiency condensing boiler replacement",
      "capex": 520000,
      "opex_delta_yr": -14000
    }
  }
}

Non-fuel-switch measures omit electrical_capacity and efficiency_alternative entirely (see skills/construction-costing/example-data.json's led-upgrade entry for the minimal shape).

Before handing the roster's measure.cost objects downstream, validate them by running:

node scripts/validate-measure-cost.mjs

This checks the JSON Schema shape and the contract rules beyond the schema — notably that every fuel_switch measure has both electrical_capacity and efficiency_alternative, and that an UNVERIFIED electrical capacity never has upgrade_cost.low === upgrade_cost.high (i.e., it is never a disguised point estimate). Fix any measure that fails validation before proceeding — do not hand an invalid measure.cost object downstream.

Step 6: Hand off

The measure.cost objects this method produces are consumed by decarb-plan's measure screening step (which combines them with energy-savings estimates to compute NPV/IRR/payback, applying landlord-share capture separately) and are read by quality-review when auditing a plan's cost basis for defensibility. This skill's job ends at producing a validated measure.cost per schema/measure-cost.schema.json — it does not itself rank, screen, or render measures.