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.
What the campus should be doing at its age, load and weather — from manufacturer and engineering sources — against what the satellites see. Residuals are diagnostic targets with ranked physical hypotheses, never causal findings.
Same-acquisition deviation from the completed-membrane peer median removes sun geometry, atmosphere and reference drift. Landsat thermal deviation from the other roofs is an independent series with mixed 100 m pixels.
| Roof | Role | Albedo (cohort) | Δ vs peer median | Percentile | Landsat ST vs roof peers | n |
|---|---|---|---|---|---|---|
| General Roof 5 · CENTRAL_CORE | visibly soiled | 0 | 47 | |||
| General Roof 4 · DATA_HALL_04 | visibly soiled | 0 | 74 | |||
| General Roof 2 · DATA_HALL_03 | visibly soiled | 5 | 87 | |||
| General Roof 7 · DATA_HALL_02 | visibly soiled | 10 | 96 | |||
| General Roof 1 · DATA_HALL_01 | visibly soiled | 22 | 121 | |||
| General Roof 6 · CENTRAL_CORE | — | 17 | 107 | |||
| General Roof 11 · SUPPORT | — | — | — | 0 | ||
| General Roof 3 · DATA_HALL_02 | visibly soiled | — | — | 143 | ||
| General Roof 8 · DATA_HALL_01 | clean control | 46 | 126 |
69 of 122 clean acquisitions ≥0.03 below peers (mean -0.0686, z -2.47), Nov 14, 2024 → Jun 7, 2026; sentinel-2a, sentinel-2b, sentinel-2c; seasons 2024-Q4, 2025-Q4, 2026-Q1, 2026-Q2.
75 of 135 clean acquisitions ≥0.03 below peers (mean -0.0468, z -1.94), Nov 22, 2024 → Jun 7, 2026; sentinel-2a, sentinel-2b, sentinel-2c; seasons 2024-Q4, 2025-Q1, 2025-Q4, 2026-Q1, 2026-Q2.
| Roof | Direction | Step (Δ albedo) | Window | Confirming | Sensors |
|---|---|---|---|---|---|
| General Roof 7 · DATA_HALL_02 | Step up | +0.263 | Nov 9, 2025 → Nov 12, 2025 | 69 / 433 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 5 · CENTRAL_CORE | Step up | +0.271 | Dec 9, 2024 → Dec 14, 2024 | 86 / 434 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 6 · CENTRAL_CORE | Step up | +0.288 | Jan 23, 2025 → Jan 31, 2025 | 74 / 434 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 2 · DATA_HALL_03 | Step up | +0.309 | Jun 30, 2024 → Jul 2, 2024 | 42 / 443 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 4 · DATA_HALL_04 | Step up | +0.312 | Nov 12, 2024 → Nov 14, 2024 | 41 / 434 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 8 · DATA_HALL_01 | Step up | +0.332 | Mar 4, 2025 → Mar 7, 2025 | 53 / 423 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 10 | Step up | +0.353 | Mar 24, 2025 → Mar 27, 2025 | 46 / 430 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 3 · DATA_HALL_02 | Step up | +0.365 | Oct 15, 2024 → Oct 20, 2024 | 57 / 435 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 9 | Step up | +0.367 | Mar 14, 2025 → Mar 19, 2025 | 51 / 431 | sentinel-2a, sentinel-2b, sentinel-2c |
| General Roof 1 · DATA_HALL_01 | Step up | +0.422 | Oct 3, 2024 → Oct 8, 2024 | 65 / 431 | sentinel-2a, sentinel-2b, sentinel-2c |
On a campus under construction a step-up is usually membrane completion; the Timeline page checks each step against the land-state classifier's completion date and says which it is.
Same-day matches (≤1 day) across landsat-8, landsat-9, sentinel-2a, sentinel-2b, sentinel-2c; both quantities peer-normalized at the same acquisition. The relationship holds in low and high irradiance and both seasons (low ghi r -0.48, n 205; high ghi r -0.51, n 215; cool season r -0.53, n 343; warm season r -0.32, n 77). Roofs that read darker than their peers run hotter. Every roof is a mixed 100 m thermal pixel — agreement across sensors is necessary for a physical reading, not attribution of cause.
visually clean-control candidate (no analyst soiling tag) whose spectral state is below its expected age-adjusted baseline (now/plateau 0.846 vs 0.920); not a verified control. Its Landsat peer temperature is slightly below peers, so the two sensors disagree on this roof. It stays the observation target for a wash test rather than the baseline the experiment is measured against.
The three configuration priors the engine carries for this campus, each evaluated hourly on the same weather, load prior and prices. Water-cooled plants trade electricity for water; dry heat rejection trades water for electricity — no single metric is optimized in isolation. Weights are the evidence-weighted prior, not a finding.
| Prior | Weight | Annual PUE | Overhead MWh/yr | Peak overhead kW | Overhead cost/yr | Economizer hours | Water m³/yr |
|---|---|---|---|---|---|---|---|
| B ac chiller dry | 0.45 | 1.241 | 355,016 | 59,685 | $13.0M | 11 % | 0 |
| A wc chiller tower | 0.3 | 1.209 | 308,235 | 44,690 | $11.4M | 28 % | 3,128,614 |
| C liquid it external | 0.25 | 1.121 | 178,797 | 22,335 | $6.7M | 28 % | 2,902,123 |
Evidence-weighted PUE 1.202 on an assumed IT load. An equipment schedule collapses this table to one row.
Residuals from engine run 78a2537c on graph B_ac_chiller_dry. Roof residuals compare the last-60-day observed reflectance ratio (vs each roof's own plateau) with the age-adjusted expectation; thermal residuals are peer-normalized Landsat ST. Residuals are diagnostic targets, not causal findings.
| Asset | Metric | Expected | Observed | Residual | z | Persistence | Top hypothesis |
|---|---|---|---|---|---|---|---|
| General Roof 1 · DATA_HALL_01 | Peer normalized surface temp c | -1.01 | 109/186 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 7 · DATA_HALL_02 | Peer normalized surface temp c | -0.75 | 111/181 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 8 · DATA_HALL_01 | Peer normalized surface temp c | -0.74 | 128/183 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 9 | Peer normalized surface temp c | -0.42 | 133/179 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 8 · DATA_HALL_01 | Relative solar reflectance now vs plateau | -0.53 | 7/7 · 60 d | surface soiling / dust deposition (reversible) | |||
| General Roof 3 · DATA_HALL_02 | Relative solar reflectance now vs plateau | 0.10 | 5/15 · 60 d | surface soiling / dust deposition (reversible) | |||
| General Roof 9 | Relative solar reflectance now vs plateau | 0.12 | 1/8 · 60 d | surface soiling / dust deposition (reversible) | |||
| General Roof 6 · CENTRAL_CORE | Relative solar reflectance now vs plateau | 0.13 | 0/5 · 60 d | surface soiling / dust deposition (reversible) | |||
| General Roof 3 · DATA_HALL_02 | Peer normalized surface temp c | 0.07 | 113/184 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 10 | Peer normalized surface temp c | 0.11 | 107/175 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 6 · CENTRAL_CORE | Peer normalized surface temp c | 0.12 | 95/180 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 11 · SUPPORT | Peer normalized surface temp c | 0.33 | 114/186 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 2 · DATA_HALL_03 | Peer normalized surface temp c | 0.43 | 142/186 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 4 · DATA_HALL_04 | Peer normalized surface temp c | 0.64 | 143/184 · 700 d | lower reflectance than expected (soiling / weathering) | |||
| General Roof 5 · CENTRAL_CORE | Peer normalized surface temp c | 1.19 | 143/186 · 700 d | lower reflectance than expected (soiling / weathering) |
The engine evaluates the whole system hourly on real weather. The observed column stays owner-data-required until BAS trends or 15-minute facility power arrive; that dataset turns every mechanical counterfactual from a modeled scenario on an inferred state into an observed residual with a measured uplift. See the observed-mechanical-performance surface.
| Metric | Expected (30-day mean) | Observed | Residual |
|---|---|---|---|
| Facility power | |||
| IT load | |||
| PUE | |||
| Chiller power | |||
| Auxiliaries (pumps, fans) | |||
| Roof heat gain (all roofs) | |||
| Cooling demand | |||
| Available compute power |
What a white TPO roof should reflect as it ages (manufacturer CRRC initial and 3-year aged anchors, interpolated; the band is the manufacturer spread, never averaged), with each roof's satellite-measured reflectance at its actual age. Dots below the band are roofs reflecting less sunlight than their age explains.
8 of 8 roofs measure below the manufacturer's expected range for their age.
Version dc-baseline v1.2.0 · 203 metrics · 56 curves · membrane is a visually inferred candidate, not a verified product.
How a change anywhere on the campus moves through cooling, power and compute. Every line carries its physics: function, units, uncertainty and source.
How to read this: left to right is the order energy moves. Weather and sun heat the roofs and the IT equipment makes heat; the cooling plant removes it, rejects it outdoors and draws electricity; that electricity shows up in facility power, PUE and the compute capacity left over. Pick an intervention to light up every part of the system it changes.
Which assets and interactions have the most leverage under this cooling prior, from the engine's design-day sweeps. Every label is prefixed MODELED because no plant observation exists.