Reliability, measured

No one would size a generator from a rumor about the load. Yet resilience investments are sized from rumors about outages every week. The measured record exists. Start there.

Section 01The vocabulary, in one table

Utility reliability reporting uses a small set of indices, defined consistently enough that you can compare years and, with care, systems. The California Public Utilities Commission requires the state's investor-owned utilities to file annual electric system reliability reports built on them, at system and division level, each July.¹

Two reading disciplines matter. First, system averages are not your site: reliability varies enormously by division, circuit, and even position along a circuit. Second, the definitions split "sustained" from "momentary" at a threshold measured in minutes, so a facility whose real exposure is momentary events can look fine in SAIDI terms while bleeding money at MAIFI frequencies.

Section 02Where the actual data lives

Three layers of public record cover a California site, from general to specific:

• The CPUC annual reliability reports. Filed each year by PG&E, SCE, SDG&E, and the smaller investor-owned utilities under standing Commission requirements, with system and division indices, worst-performing circuit lists, and major-event documentation.¹
• The utility's own reliability publications. PG&E publishes electric reliability reporting, including performance by area, on its public site.² These are the same underlying indices, presented closer to the customer.
• Your circuit and your meter. The decisive layer. Utilities maintain outage history by circuit, and your own operations log plus interval data record what actually reached your equipment, including momentary events and voltage excursions the public indices summarize away. Ask your utility representative, in writing, for the interruption history of your specific circuit for the last several years.

Public Safety Power Shutoffs sit alongside these as a separate, planned category with its own public event record: dates, areas, durations, and customer counts for each event. Treat PSPS exposure as its own line in the analysis, driven by your circuit's fire-risk geography rather than by ordinary equipment performance, and pull its history from the documented event records rather than memory.

The average is the marketing. The circuit is the truth.

Section 03From record to price: what an outage actually costs you

The measured record answers frequency and duration. The other half of the resilience equation is consequence, and it is site-specific arithmetic no index can supply: what does an interruption of a given depth and duration cost this facility?

The honest way to build it is a consequence table, priced with your operations team, labeled illustrative where it estimates: the cost of a momentary blink (tripped drives, scrapped batch, restart labor), of an hour (idle labor, missed shipments), of a day (spoiled inventory, contractual penalties, recovery time). Multiply against the measured frequencies from Section 02, and resilience stops being a feeling and becomes an expected annual loss you can compare to the annualized cost of mitigating it.

Done this way, the analysis frequently surprises in both directions. Sites with dramatic outage stories sometimes discover their expected annual loss is modest, and the right answer is ride-through hardening for momentary events rather than a resilience plant. Sites with no stories at all sometimes discover a single-shift interruption would breach a customer contract whose penalty dwarfs a decade of energy savings. Neither discovery is available from anecdote.

Section 04Matching mitigation to the measured problem

Each mitigation class answers a different slice of the record, and the record should pick the mitigation, not the other way around:

• Power conditioning and ride-through (flywheels, small storage, robust drive settings) answer MAIFI: frequent momentary events on an otherwise decent circuit.
• Standby generation answers rare-but-long: low SAIFI, high consequence, where hours-to-days of fuel-backed coverage is the need and idle capital the cost.
• Continuous on-site generation with islanding capability answers the combination of poor circuit performance, PSPS exposure, and an economic case that already stands on everyday operation. The resilience is then a priced increment (islanding controls, protection, testing), not the justification.
• The grid alone answers more sites than the resilience industry admits: a strong circuit record and modest consequences make "do nothing beyond good housekeeping" the defensible line, and a neutral study should be willing to write it.

National context supports taking planning seriously without panic: reliability assessments have flagged rising risk in parts of the North American system as load grows and the resource mix turns over, which argues for measuring your exposure now, not for buying whatever is nearest when a storm makes the news.³

An expected-loss table, worked with illustrative numbers

To make Section 03 concrete, here is the arithmetic for a fictional facility, every figure illustrative. Suppose the circuit record shows, on average, two momentary events a year, one sustained two-hour outage a year, and one full-day public-safety shutoff every third year. Suppose operations prices those honestly: a momentary blink trips drives and scraps in-process product for about fifteen thousand dollars; a two-hour outage idles two shifts' worth of throughput for about forty thousand; a full lost day, with spoilage and recovery, runs about a hundred and fifty thousand. The expected annual loss is then two times fifteen, plus forty, plus one third of one hundred fifty: roughly one hundred twenty thousand dollars a year.

Now the mitigations price themselves against that number. Ride-through hardening that eliminates the momentary losses is worth about thirty thousand a year; if its annualized cost is ten, it buys itself three times over, and it is usually the cheapest line on the menu. A standby asset that covers the sustained events but not the momentaries is worth about ninety thousand a year against its own annualized cost of capital, fuel readiness, and the testing regime the standards prescribe; that comparison can land either way, which is precisely why it must be run rather than assumed. And a full islanding microgrid is worth, at most, the whole hundred twenty, a ceiling that quietly disqualifies seven-figure resilience programs at facilities whose measured record is this ordinary. Reverse the inputs, a semiconductor step where a blink costs half a million, a cold-chain warehouse where a lost day breaches contracts, and the same arithmetic justifies spending that would look extravagant anywhere else.

The table takes an afternoon to build and survives every meeting it enters, because it is made of your record and your consequences rather than anyone's fear or anyone's brochure. It is also the exact artifact an underwriter or lender will eventually ask to see, which makes the afternoon doubly paid.

Section 05The five questions before any resilience purchase

1. What is my circuit's interruption history, in writing?Sustained and momentary, several years, from the utility, not from memory.
2. What is my PSPS exposure, from the event record?Actual events touching your circuit and their durations, from documented histories, plus your area's fire-risk designation.
3. What does each outage class cost this facility?Momentary, hour, shift, day, priced with operations, labeled illustrative where estimated.
4. What is the expected annual loss, and what does each mitigation cost per year?The comparison that turns resilience into a line item.
5. Would the mitigation have covered the events in my record?Duration, fuel, and islanding scope checked against the actual history, not the brochure scenario.

A resilience recommendation that arrives without answers to these five is a product looking for a fear. Your record is filed and your consequences are yours to price. Between them, the decision is closer than it looks.