Loading from the repository — every figure will carry its evidence class and lineage.
Loading from the repository — every figure will carry its evidence class and lineage.
Everything on this site was built from orbit and public sources. The roof is observed; the plant is modeled on inferred states; the economics rest on an assumed IT load and facility cap. Each item below names what the model uses today, the owner field that replaces it, and what it resolves. Priority 1 collapses the most uncertainty per hour of the owner's time.
Turn the modeled mechanical opportunities into observed ones, then validated ones. Each rung names the owner fields, what they convert, and which page changes as a result. Rungs are ordered by uncertainty collapsed per hour of the owner's time.
Converts: the three-way cooling-architecture prior and family-level curves.
Converts: IT load, facility cap and therefore whether any recovered kW has capacity or compute value.
Converts: every as-found degradation state (fouled coils, loaded filters, scaled fill, setpoint).
Converts: predicted maintenance uplift → measured uplift.
Converts: wholesale-proxy energy value and the compute marginal-contribution sensitivity.
Collapses: A/B/C configuration prior (±2× on chiller kW, water, PUE)
Collapses: IT load, facility cap, whether compute opportunity is ever non-zero
Collapses: inferred degradation states behind every mechanical counterfactual
Collapses: R-30 vs R-25 vs wet-insulation sensitivity
Collapses: S2->SR scale and re-brightening cause
Collapses: wholesale settlement-point proxy vs actual delivered energy price
Collapses: marginal contribution $/IT-kWh sensitivities
Three examples set the pattern: a cooling equipment schedule collapses the A/B/C cooling-architecture prior; IT load plus the facility cap decides whether recovered kW has any compute value; BAS trends convert mechanical degradation from modeled to observed and then validated.
| Priority | Owner field | What the model uses today | What it resolves | After |
|---|---|---|---|---|
| 1 | Cooling architecture: chiller type, heat rejection, liquid vs air, economizer Mechanical one-line / plant narrative | Assumed Three evidence-weighted priors: A water-cooled + tower (0.30), B air-cooled / dry cooler (0.45), C liquid IT + external rejection (0.25). Equipment farms are unknown_mechanical / unknown_electrical. |
Collapses the A/B/C configuration prior (±2× on chiller kW, water and the PUE band) to one graph. | Available |
| 1 | Equipment make, model and schedules: chillers, dry coolers, towers, CRAH/AHU, CDU, pumps, fans Equipment schedule / nameplate photos | Manufacturer / standard Family-level manufacturer data (Trane / Daikin / JCI-YORK classes) from the RAG; no exact models; unit counts assumed. |
Exact-model performance maps replace family curves (preference rank 1 instead of 3); staging thresholds become known. | Available |
| Priority | Owner field | What the model uses today | What it resolves | After |
|---|---|---|---|---|
| 1 | IT load per hall (kW, utilization) and rack / accelerator inventory DCIM export / monthly IT-kW | Assumed 25 MW per hall (15–40 MW band) at 0.8 utilization, ASSUMED; platform is a generic HGX-class 8-accelerator server. |
Removes the 15–40 MW band; with the interconnect cap it decides whether the facility is ever power-constrained. | Available |
| 1 | Facility / interconnect capacity limit and contracted power Interconnection agreement / utility contract summary | Assumed Facility cap prior = 1.3 × design facility power (ASSUMED); the modeled year is never constrained, so compute value is zero. |
Together with IT load it determines whether recovered kW has compute value at all — currently the biggest unknown in the economics. | Available |
| Priority | Owner field | What the model uses today | What it resolves | After |
|---|---|---|---|---|
| 1 | BAS / BMS trend history (12+ months, 15-minute or better) BAS trend export (CSV / historian) | Inferred No plant telemetry. Every mechanical finding is a counterfactual on an inferred as-found state. |
Converts mechanical degradation from MODELED to OBSERVED / VALIDATED; enables the observed mechanical performance surface. | Available |
| 2 | Filter differential pressure trends and change-out records BAS points or CMMS records | Inferred filter_loaded is an inferred prior (design ΔP × assumed loading). |
Confirms or dismisses the single largest air-side leverage row. | Available |
| 2 | Coil / condenser approach temperatures BAS points (entering/leaving temps) | Inferred coil_fouled and condenser_fouled are inferred priors from manufacturer fouling factors. |
An observed approach above design is the evidence these rows currently lack. | Available |
| 2 | Cooling-tower approach and range BAS points (CWS/CWR, wet bulb) | Inferred fill_scaled is an inferred prior; design approach from engineering standards. |
Confirms whether tower performance is below design at all (only applies under the water-cooled prior). | Available |
| 2 | CHWS / CHWR setpoints and trends BAS trend export | Assumed CHWS 60 °F and supply air 24 °C, ASSUMED. |
Replaces assumed setpoints with measured operating points in the chiller model. | Available |
| 2 | Maintenance history: filter changes, coil / tube cleaning, tower treatment, roof washing CMMS export / work orders | Inferred Degradation states are inferred priors with no dated events. |
Turns inferred as-found states into dated known states; enables before/after validation of every intervention row. | Available |
| 3 | Chilled- and condenser-water flows; pump and fan speeds BAS points / VFD logs | Manufacturer / standard Affinity-law priors from engineering standards. |
Pins the auxiliary term, which is the second-largest overhead component in the model. | Available |
| Priority | Owner field | What the model uses today | What it resolves | After |
|---|---|---|---|---|
| 1 | 15-minute facility and cooling-plant power (or utility interval data) Utility interval data / BAS meter points | Assumed No facility telemetry. Facility power is modeled from IT priors and expected plant curves. |
Turns the whole facility layer from MODELED to OBSERVED and lets residual diagnostics run on the plant. | Available |
| 2 | Measured PUE (monthly or better) Monthly report | Modeled Modeled PUE 1.12–1.24 across priors, evidence-weighted 1.20; no observation. |
A single measured PUE trajectory discriminates between the three cooling priors. | Available |
| Priority | Owner field | What the model uses today | What it resolves | After |
|---|---|---|---|---|
| 3 | Roof assembly: membrane product, insulation R, cover board, deck, install dates Submittals / warranty file / permit set | Inferred TPO candidate (visually inferred); 60-mil white TPO configuration prior; R-30 code prior with R-25 / R-18 sensitivities; install dates from public documentation or the albedo step. |
Collapses the manufacturer reflectance band and the R-30/25/18 spread; removes the 'different product than peers' hypothesis. | Available |
| 3 | 2025 roof wash / coating records and a test-patch wash with before/after Sentinel-2 passes Coordinate a wash date; work orders | Inferred The Aug–Dec 2025 re-brightening is unexplained; cleaning recovery fraction 0.75 (0.55–0.90) is an empirical class value. |
One ~$2k wash of a test patch observed across 3–4 clean acquisitions validates or falsifies the roof hypothesis. | Available |
| Priority | Owner field | What the model uses today | What it resolves | After |
|---|---|---|---|---|
| 3 | Energy contract / retail tariff and the campus settlement point PPA summary / tariff schedule / bills | Peer derived Wholesale settlement-point proxy; exact node pending; PPA or tariff terms unknown. |
Replaces the wholesale proxy with the delivered $/MWh the owner actually pays. | Available |
| 3 | Compute utilization, contract price, capital recovery Utilization report / contract summary | Assumed Public on-demand list price as a gross ceiling; marginal contribution $/IT-kWh sensitivities ASSUMED. |
Turns the compute lens from a sensitivity range into a number. | Available |
This is the surface BAS data unlocks. Each row compares an asset's expected performance (from the baseline engine at its configuration, age, load and weather) with its actual normalized performance from trends, attributes the residual to a probable degradation signature, predicts the maintenance uplift from the matching counterfactual, and — after a maintenance event — records the measured uplift. Until trends arrive every cell is owner-data-required; nothing here is estimated.
| Asset | Expected performance | Actual normalized performance | Residual | Confidence | Probable degradation signature | Predicted maintenance uplift | Measured post-maintenance uplift |
|---|---|---|---|---|---|---|---|
| Chillers / dry coolers | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required |
| CRAH / AHU filters and fans | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required |
| Condenser / heat rejection | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required |
| Cooling tower | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required |
| Pumps | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required | Owner data required |
Expected backend contract: rpc/campus_mechanical_performance(p_campus_slug, p_token) fed by BAS trends or utility interval data in the org-private tier — documented in docs/FRONTEND_ARCHITECTURE.md §9. The modeled counterfactuals are the predicted-uplift column of this table.