TowerWatch x CPUC Evidence Package
Credibility anchor — the boundary correction
The core finding that defines TowerWatch’s technical credibility is not a big number. It is a correction.
The BWFP Parking PV array (260 kW peak, 389 MWh/year) is outside the RMI microgrid boundary. A four-test empirical check — correlation, level, excess, and energy balance — proved it connects to UC Davis campus distribution upstream of the microgrid, not to RMI BWFP. Every storage and dispatch conclusion built on the prior (incorrect) scope had to be retracted.
At 0.54× penetration, naive peak shaving with a 30 kWh battery outperforms all modeled intelligent dispatch strategies. Storage value at RMI BWFP is entirely demand-charge driven; remove SMUD CITS-1 demand charges and NPV is negative at every capacity. Source: docs/cpuc/01_microgrid_results.md.
Claims that will never appear in any CPUC-facing document
- “Battery was observed dispatching.” — Not observable; no telemetry in PI.
- “Battery SOC was tracked.” — Not in PI.
- “Grid export was measured.” — DHWA meter is unidirectional.
- “Storage savings are empirically confirmed.” — They are simulated only.
- “The microgrid has 2.3× PV penetration.” — That is campus scope; microgrid scope is 0.54×.
Live campus status
Pulled directly from the running TowerWatch API. Failed fetches show a muted dash; no values are fabricated.
Evidence deck (full reading package)
Six documents, 969 lines total. Start with the README for the narrative and the empirical-vs-simulated map.
Architecture footprint
TowerWatch runs on four nodes. No institutional compute procurement was required.
What TowerWatch has NOT solved yet
Publishing the gap list is the trust move. Every item is verifiable against a committed source file.
-
Water tower temperature sensor — stuck since 2020DWtr_ET01#FT1131.FPTemp returns 68.0 degF flat from the UTIL-PI-P mirror. Real data likely lives on the Wonderware historian (foa-util-pi-s), which is intermittently disconnected. We do not call the sensor dead; we call it a blind spot in our current query pipeline.docs/cpuc/05_data_integrity_gaps.md §1a
-
JSWB 250 kWh battery — not operationalTwenty second-life Nissan LEAF packs, CEC-funded, physically present on the south side of Jackson Sustainable Winery. The system is not running; reason unknown to Facilities. This is an institutional/operational gap, not a technical one.docs/cpuc/05_data_integrity_gaps.md §2a
-
JSWB telemetry never connected to PI historianJackson building has its own Fusion BMS and a separate fermentation sensor network. Neither feeds the campus PI historian. Data is simply not being collected at the historian level — any JSWB analysis must rely on the DHWA grid-tie meter.docs/cpuc/05_data_integrity_gaps.md §2b
-
Biodigester (READ) — net energy sink5.13 M kWh consumed, 1.59 M kWh produced, net 3.5 M kWh deficit. More gas flared than captured (11,771 MSCF vs 7,570 MSCF). Microturbines failed and were replaced with a conventional natural gas generator. The clearest operated-failed-abandoned example on campus, independently verifiable from the historian.docs/cpuc/05_data_integrity_gaps.md §3a
-
Maintenance reporting — confirmed broken by Facilities Energy leadJosh Morejohn (Facilities Energy) confirmed on 2026-03-31 that “the whole process is extremely flawed.” TowerWatch cannot rely on official maintenance-and-repair records as ground-truth labels for ML anomaly validation — the system must generate labels through student field verification until formal access is granted.docs/cpuc/05_data_integrity_gaps.md §4a
-
No automated vacation scheduling — contributed to the July 2024 cascadeHoliday schedules are maintained manually in Airtable and routinely missed. During the July 2024 chiller plant cascade, buildings drew full cooling while unoccupied, draining TES that should have been available for real loads. Software/process gap with direct dollar and COP impact during heat events.docs/cpuc/05_data_integrity_gaps.md §4b