| name | wattlab-assumptions |
| description | Use when building sparse / minimal-info EnergyPlus twins: defaults hierarchy, Ideal Loads vs explicit HVAC, PNNL infiltration/OA candidates, assumption ledger, TMY→AMY ladder, and short/long fuel G14 dial (WWR/glass/ACH then banded SAT/reheat). Triggers on: sparse building, minimal info, defaults, IdealLoads, assumption ledger, autosize, constrain plant, TMY, AMY, prototype, short gas, CVRMSE, calibrate dial, WWR, infiltration. |
WattLab assumptions — agent is the assumption-maker
EnergyPlus simulates; you choose, justify, and log defaults. EnergyPlus-MCP
inspects/patches/sims — it does not pick TMY vs AMY or invent city/area.
Full ladder: ../../docs/SPARSE_BUILDING_PLAYBOOK.md.
Fuel dial depth: ../wattlab-twin-calibrate-dial/SKILL.md
· ../../docs/TWIN_DIAL_PLAYBOOK.md.
Hierarchy (strict order)
- Human / FM facts —
building_type, city, floor_area_ft2, lat/lon,
nameplate tons/hp, known schedules. NEEDS_INPUT if missing. Never invent.
- Evidence — vibe19 dump (schedules, setpoints, FDD, weather_observed),
campus bills,
buildings.json.
- Seed model — nearest archetype IDF (today:
5ZoneAirCooled screening) or
human custom_idf / future DOE–PNNL prototype by type/size/code/climate.
- Code / climate catalog — WattLab
defaults (archetypes, climate.json,
codes). Unknown city keeps user label + honest substitute EPW note (no silent
Madison remap).
- Literature defaults (stamp source) — ASHRAE-style schedules; OA methods;
PNNL infiltration candidate when no measured leakage.
Partition like a good conceptual modeler
| Stage | Focus |
|---|
| Project | Site, area, climate, bills window, rates |
| Design | Simple shell / prototype; perimeter-core OK for screening |
| Schedules | Occupancy / lighting / equipment / HVAC avail |
| Baseline | LPD, EPD, people density, ventilation, infil, setpoints |
| Measures | One hypothesis per runs/<id>/ |
| Results | Monthly vs bills, unmet hours, area scale, weather mode, peak kW, G14 |
After TMY screening: wattlab calibrate-campaign (bill months → AMY → G14) —
../../docs/CALIBRATE_AND_DELIVERABLES.md.
Hard-size nameplate: scale by 1/prototype_area_scale when scale > 1.5; refuse
outside [0.25, 4.0]. City troy → detroit catalog (user label preserved).
Ideal Loads vs explicit system
- Quick model, not all details known → Ideal Loads + rigorous envelope /
schedule assumptions first (when IDF path supports it).
- Compare HVAC pathways / plant ECMs → explicit seed (5Zone VAV today);
always show
prototype_area_scale vs target ft².
- Keep scenarios separate: (A) autosized conceptual, (B) constrained to
nameplate, (C) outage/recovery. Never merge into one “truth” run.
Assumption ledger (publish every campaign)
For each key: value, source (user / dump / default / literature),
confidence (high/med/low), sensitivity note.
Minimum keys: occupancy density, LPD, EPD, OA rate method, infiltration,
thermostat / fan hours, WWR, HVAC family, weather mode
(TYPICAL_YEAR_SCREENING / ACTUAL_YEAR_CALIBRATION /
SUBSTITUTE_CLIMATE_CONCEPTUAL_ONLY), prototype_area_scale.
For ECM-OCC-STANDBY-DCV, record the verified occupied and standby schedules,
minimum-ventilation constraint, sensor calibration evidence, and the source of
the DCV airflow reduction. Never infer zero occupied OA from a screening proxy.
Short / long fuel playbook (Twin G14 dial)
Autosized plant is fine for sparse twins — dial envelope + loads + as-operated
SAT/VAV, not invented nameplate tons. Full playbook:
../../docs/TWIN_DIAL_PLAYBOOK.md (Cursor skill:
wattlab-twin-calibrate-dial). Tools publish chain: ../../docs/AGENT_TOOLS.md.
Gas short annually (model ≪ bills)
- Lock good geometry (e.g. stacked
Floor_1…Floor_N — not DOE mid×4 when
the human wants floor massing).
- Raise WWR toward site (curtain wall often 0.70–0.75).
- Leaky glass: U ≈ 0.80–1.0 IP, not pretty DOE U.
- Infiltration ACH ladder; stop near annual gas ±5%.
- Only then HVAC shape (below).
Elec short annually
- Raise LPD / EPD (W/m²) before plant oversizing.
Annual flat but monthly gas CVRMSE fails (winter high / summer low)
- Read vibe19 AHU discharge-air-temp / SP by month (fan-on).
- Summer dump (often ~50°F when dump shows it) + higher VAV min-flow in
true summer only.
- Warmer winter SAT + shorter winter OA; band months — do not hold cold
dump across long shoulder seasons (blows Oct/Aug gas).
- If bills ≠ HDD (Feb peak / Oct tiny), band aggressively and document residual CV.
G14 pass = both fuels |NMBE|≤5% and CVRMSE≤15%. Annual % alone ≠ calibrated.
Always write calibration_scorecard.json in the Twin-expected nested shape.
Do not
- Invent office / Madison / Chicago for a real dump.
- Claim G14 on TMY vs non-overlapping bills or unscaled ~10k ft² prototype.
- Quietly publish ROI when guardrails hit
INVESTIGATE.
- Call stacked-twin calibrate “done” on DOE mid×4 zoning when the user expects Floor_1–N.
- Start gas-short campaigns with plant oversizing before envelope / infil.
