The Guarino Infrastructure Dependency Metric, Paper V. Hurricane Helene: The Communications Edge

Papers I to IV of this series built the buffer group ΠB = Ta/Tx for the edge between two infrastructure nodes, tested it on the gas-to-power-to-water chain in two winter storms, and priced its hours with a cost group ΠC = ca/(v Cd). This paper takes it to a third hazard and a fourth node. Hurricane Helene, 26 September to 19 October 2024, put 4,562 cell sites out of service at its peak, the largest count in the public record of the FCC's Disaster Information Reporting System, and the reporting form itself separates the two upstream edges of a cell site: the commercial power edge, whose buffer is a battery of two to eight hours or a generator with two to three days of fuel, and the transport edge, whose buffer is a second backhaul path or nothing. The exposures on the power edge, bounded in advance from the utility's releases and measured afterwards from the Department of Energy's EAGLE-I county record, ran to days in the mountain counties of North Carolina against Winter Storm Elliott's hours and the Texas freeze's seventy; the file shows little outage in the Tennessee counties, where its coverage is thin. The framework predicts that a battery site cannot hold, ΠB of 0.04 to 0.17 against a two-day exposure, that the realised buffer fraction of the site population falls by the second day to the share of sites with a fuelled generator and then decays at the refuelling rate set by road access, and that the transport edge, which has no stock behind it, dominates the outage in terrain where fibre runs with the rivers and outlasts the power edge because it takes longer to restore. The one daily report read before the hypotheses were fixed, 29 September, two days after the storm crossed the mountains, has 18 percent of North Carolina's power-lost sites still up on backup, 12 percent in Buncombe and none in Yancey, and transport level with power as the attributed cause; the other twenty-three reports are the test. All of them and the Department of Energy's file were opened the next day. Where half a county's customers lost power, the median exposure was 270 hours. On the second day the realised fraction passed the hypothesised ceiling of 0.4 in one county, Rutherford, so H1 fails by its own clause, though the state's 0.18 is as predicted. By 18 October, 28 of the 31 attributed North Carolina sites still out were out for transport, as H3 predicts. H2, run on the nine counties with a road-closure count and enough power-lost sites to read, fails: across days four to ten the correlation between the realised fraction and closures is near zero, on a test too small to find the predicted one reliably. An exploratory contrast added afterwards finds the lowland counties of Georgia and South Carolina holding three to six times the mountain counties' odds of keeping a power-lost site on backup at matched exposure, on every report from 28 September to 7 October. Read through the exposure fraction that Paper I's third revision names, the transport edge cost North Carolina's sites 2.3 days each against 2.1 on the power edge, the larger term in 23 of 29 counties. Priced from the public record, an hour of autonomy at a site costs, in 2024 dollars, about 4,800 in an eight-hour battery, 2,300 to 2,900 in a battery of two to seven days, 390 to 960 in a diesel generator with its tank, and at most 5,700 on an operator's programme figure for a fixed generator. Against a value of lost service of 0.06 to 0.77 dollars per person-hour and about 755 people per site, the generator pays back over 119 to 1,496 site-hours of outage: within one Helene exposure at the top tier, a little over one at the middle tier and five or six at the bill, where H4 predicted two or three. A battery takes four or more events that drain it to pay back at the top tier and fifty or more at the bill. A second transport path, a microwave hop at 36,300 to 71,000 dollars, pays back in 62 to 1,533 site-hours of transport outage. The ranking the series found on the power and water chains travels as a computation: the first dollar on the power edge goes to fuel, not to batteries, at every tier. Whether the operators bought in that order after Helene the public record does not say. What is new is an edge whose only buffer is a second path, transport, which the cost group prices against a cut fibre, and on which the tail of the event again belongs to the time to rebuild and not to autonomy. Section 9 then runs a blind test, fixed before any report on the next hurricane, Milton, was opened: the framework's prediction with nothing fitted, that once the batteries are spent the fraction of power-lost sites still up sits at the published generator share of 0.57, against a curve in exposure hours fitted to Helene. Under the rule stated in advance the buffer group predicts worse, a log loss of 0.717 against 0.657 with the interval of the difference wholly above zero, and worse than a constant: Milton's fraction ran well below the generator share for two days, even on its upper bound, and above it by the fifth, even on its lower bound, and the curve from Helene's exposure hours carried over almost unchanged. A second blind test, on Hurricane Beryl in Texas in July 2024, fixed the same way before its reports were opened, gives the same answer, a log loss of 0.719 against 0.650 with the interval again wholly above zero, and by the reading of the pair stated in advance the paper concludes that on the public record the buffer group describes the realised fraction at cell sites and does not predict it. This record carries the typeset paper and the complete working deposit: the sixteen Python scripts that compute every table, ratio and figure in the paper, each with its printed output; every results table; the three figures with their pgfplots sources; and the sources the paper reads, as downloaded (the twenty-four FCC Communications Status Reports for Hurricane Helene and the two notices that shrank the reporting area; the five FCC status reports for Hurricane Milton and the six for Hurricane Beryl, with the Commission's notices on both; three cuts of the EAGLE-I 2024 county outage file, with scripts that remake them byte for byte from the full release; the Census Bureau's 2023 county estimates; NCDOT's release, its saved reopening dashboard, the closure counts read from it and screenshots of the readings; the cost, regulatory and price documents and the web pages the paper cites, except the paywalled ones, which are cited by URL). A clean run of the scripts on the unpacked deposit reproduces every results table byte for byte. The full EAGLE-I file, the cause-decomposition preprint and the journal articles cited are not redistributed.