The Guarino Infrastructure Dependency Metric, Paper VI. Winter Storm Heather: A Blind Test on the Texas Chain

Abstract

Paper V found that the buffer group ΠB = Ta/Tx, applied to cell sites whose buffers are unknown, predicted worse than exposure alone in two blind storms, and traced the failure to the unknown numerator. This paper moves the test to an edge where the numerator is written down. Since 1 December 2023 every ERCOT generation resource has had to prepare for sustained operation at its weather zone's 95th percentile minimum 72-hour average wind chill, a design value that spans 37.6 °F across ten zones. The group therefore ranks resources by how far a storm pushed their zone toward its design, D = W_d − W72, where exposure alone ranks them by cold, C = −W72. For Winter Storm Heather, January 2024, the two scores computed at the stations that set the standard have a rank correlation of 0.08 across the zones, and every zone stayed inside its design on the 72-hour reading. The test was written, run and hashed before ERCOT's unit outage record for the storm was downloaded, and then scored unchanged on 670 resources, 167 of which lost at least half their seasonal capacity. The two scores do not separate: AUC 0.515 for the design deficit and 0.513 for cold, a difference of 0.002 with a 95 percent interval of −0.120 to 0.122. The group's own claim, that resources inside their design show no excess of losses over quiet weeks, fails: they lost capacity at 0.249 against 0.101, a difference of 0.148 (0.095 to 0.204). ERCOT's records label 21 of the storm's losses icing or cold weather and 15 fuel supply or pipeline problems, labels that no quiet-week loss carries. Paper II's pipeline result does not replicate, while its fuel order, wind first and nuclear last, holds at all six thresholds. With Paper V, the series has now tested the group blind three times as a predictor of which unit or site fails, and it has not beaten exposure; what carries from storm to storm is its ordering of fuels. The protocol was fixed before the outcomes were read but was posted only with the results, and it is published here unchanged. This record carries the typeset paper; the protocol exactly as it was fixed before the outcomes were downloaded, with its PDF, its source, its two deposit zip parts byte for byte as delivered and the SHA-256s of every protocol file; and the complete working deposit: the frozen script, code/heather_blind_test.py (SHA-256 4e5fcebb69063dcb1b36602e658ce5ba999c5875abd4872a308d11d56ac7e326), and the predictions it wrote before the outcomes, results/heather_predictions.csv (SHA-256 e031ac49ee8399fa78c1ff31ac4718a6882c9350cb11210be34de395ea519298); the reader the protocol permits and the descriptive script written after scoring, with every results table and printed output; and the inputs and outcomes as downloaded: ERCOT's 39 daily Unplanned Resource Outages posts from 30 December 2023 to 6 February 2024, the PUCT's Phase II order, ERCOT's Historical Weather Study with its three attachments, ERCOT's July 2023 interpretation of the wind chill standard, ERCOT's December 2020 and December 2023 capacity reports, EIA-860 2023, the Iowa Environmental Mesonet's hourly records for the study's ten stations, and Paper II's copy of ERCOT's Uri outage record. A clean run of the three scripts on the unpacked deposit reproduces every results file byte for byte. The deposit is one zip; its protocol/ folder holds the protocol's own two zip parts, unchanged and renamed to carry this record's DOI.