California’s 2020 rotating outages made the net peak a resource-adequacy accountability test

Summary

• On August 14 and 15, 2020, the California ISO ordered controlled load shed after supply and operating reserves became inadequate during a West-wide heat wave. CPUC-jurisdictional utilities reported about 491,600 customer accounts interrupted on August 14 and 321,000 on August 15; those are interruption counts, not unique people or a statewide harm total.
• The joint CAISO, CPUC, and CEC investigation found no single root cause. It identified extreme weather, resource-adequacy and planning processes that had not kept pace with early-evening needs, and market practices that intensified supply pressure under stressed conditions.
• The decisive accountability gap was temporal. California could show capacity against a monthly gross-peak obligation, yet some of that capacity was unavailable, derated, not scheduled, dependent on constrained imports, or unable to perform at the hours when solar output was falling and net demand was highest.
• Repair must be demonstrated with an hourly ledger: firm resources at the critical net peak, import assumptions tested against correlated regional heat, battery state of charge, generator outage and derate evidence, demand-response availability and metered delivery, day-ahead schedule error, controlled-outage execution, and the resilience of essential services and vulnerable customers.

Two evenings turned a planning abstraction into customer interruption

California’s electricity system entered mid-August 2020 under an unusually broad heat event. High temperatures were not confined to one coastal load pocket or one balancing authority. Demand was elevated across the western United States, reducing the spare generation that neighboring systems could send into California. Wildfires also threatened transmission corridors. Inside the California ISO balancing authority area, generation outages and temperature-related derates reduced available supply while demand remained high into the evening.

The authoritative starting point is the joint agencies’ Final Root Cause Analysis. It records a Stage 3 Emergency at 6:38 p.m. on August 14 after the system could no longer cure a contingency-reserve deficiency with available generation. CAISO ordered two phases of controlled load shed, each nominally 500 megawatts. Distribution utilities executed the interruptions. Restoration began as conditions improved, although actual customer durations varied by utility and circuit.

The second day followed a different path to the same emergency boundary. Clouds caused solar production to decline sharply in the afternoon. Wind production later fell, and a generator ramped down after an erroneous dispatch from its scheduling coordinator. CAISO declared Stage 3 at 6:28 p.m. and ordered about 500 megawatts of controlled load shed. Wind production then recovered by more than 500 megawatts, allowing CAISO to cancel Stage 3 and order restoration roughly twenty minutes after the load-shed instruction.

These were not uncontrolled collapses. Controlled interruption protected system frequency and reduced the risk that a further generator or transmission loss would destabilize more of the western grid. That operational necessity matters. It does not erase earlier accountability. The fact that operators chose the least-bad emergency action says little by itself about whether procurement, forecasts, schedules, resource availability, and contingency design had supplied a reasonable margin before the emergency.

The two days also show why “the blackout” is an imprecise label. August 14 and August 15 had different weather, renewable profiles, generator events, demand-response performance, and restoration timelines. A useful review asks what was controllable at each interval. It does not compress every shortfall into a single technology or repeat a statewide peak number as though it described the hour at which reserves were exhausted.

Customer counts measure interruption, not total harm

The joint report’s utility tables identify approximately 491,600 CPUC-jurisdictional customer accounts affected on August 14 and 321,000 on August 15. The August 14 durations ranged from about fifteen to 150 minutes across the three large investor-owned utilities; the August 15 range was about eight to 90 minutes. Actual utility response exceeded the nominal CAISO instructions in aggregate on both days because load shed occurs through available distribution blocks, not through a perfectly divisible statewide switch.

Those figures must be kept in their proper units. They are customer-account interruptions reported by utilities, not a count of unique individuals. The same account may have been affected on both days. A residential meter can represent one person or a household; a commercial meter can represent a store, workshop, clinic, office, communications cabinet, or multi-tenant site. The public record reviewed for this article does not establish a complete outage-attributed death count, a statewide economic-loss figure, or a customer-by-customer harm registry.

The CPUC’s final joint-agency release correctly frames the event as a shared-responsibility problem. That framing should not become shared vagueness. Customers experienced the physical consequence, but they did not set planning-reserve rules, qualify capacity, approve scheduled outages, procure imports, build market software, or decide which feeders would be disconnected.

Short outages can create long secondary effects. A small business may lose refrigerated inventory or interrupt a production run. A traffic signal can fail. A resident using electrically powered medical equipment may need a backup plan. A communications or water facility may transfer to standby power and consume limited fuel. A school, cooling center, care home, or local government office can remain impaired after grid supply returns if its equipment does not restart cleanly.

That is why accountability needs two ledgers. The grid ledger records megawatts, reserves, frequency, bids, schedules, outages, and restoration. The continuity ledger records which essential functions lost service, whether backup power started, how long it carried load, which customers lacked safe alternatives, and whether repeated interruptions fell disproportionately on the same communities. The first prevents system collapse. The second shows whether the method used to protect the system transferred avoidable harm to people who had the least ability to absorb it.

Gross peak and net peak were different control surfaces

The most important clock on August 14 did not stop at the gross demand peak. System demand peaked at 4:56 p.m. The net-demand peak—the demand left after variable generation, especially solar, is subtracted—arrived at 6:51 p.m. By then air-conditioning load remained high while solar output was falling rapidly. The grid needed other resources to replace the declining solar contribution and maintain contingency reserves.

Traditional monthly resource-adequacy obligations were substantially oriented toward the forecast monthly gross peak plus a planning reserve margin. That method can identify a large portfolio on paper. It does not automatically demonstrate that the portfolio can deliver during every critical hour. A solar resource may contribute meaningfully at the gross peak and much less two hours later. A battery may have nameplate capacity yet lack sufficient charge or duration. A gas unit may be listed at normal capability but derate in extreme ambient heat.

A demand-response program may receive a planning credit larger than the load it can meter and sustain at the relevant time.

The Final Root Cause Analysis did not say that solar caused the outages. It said planning targets had not kept pace with the changing shape of critical need. That distinction is more than political etiquette. It identifies the control failure. If a system adds daytime renewable output without aligning procurement, storage, flexible generation, imports, and demand response to the evening ramp, the problem is not the existence of solar. The problem is a planning and performance framework that counts unlike resources as though their capacity were interchangeable across time.

The agencies’ recommendation to apply resource-adequacy targets to both gross peak and the critical net-peak hour was therefore a direct response. It converted an observed timing gap into a procurement test. But adding another compliance hour is only a first step. A five-hour heat-wave ramp, a cloudy afternoon, a transmission constraint, or a battery with an unexpected state of charge can still defeat a single-hour showing.

The better accountability question is a duration curve: for each stressed hour, what physical supply can be delivered after correlated derates, realistic imports, storage energy limits, planned outages, and demand-response availability? A certificate should be the beginning of that evidence, not the end. The net peak made visible what customers had already paid the system to know: adequacy is not merely how many megawatts exist, but whether the right megawatts can arrive at the right place and time.

The reserve margin was a probability assumption, not a vault of power

Planning reserve margins are often described as though operators keep a fixed percentage of extra electricity in storage. In practice, a reserve margin is a planning construct applied to a forecast. It helps procure capacity for uncertainty in load and resource availability. The margin does not guarantee that each counted resource will bid, start, remain online, receive fuel, avoid heat-related derating, or deliver through an unconstrained path.

The August event exceeded conditions embedded in the prevailing planning framework. The heat was widespread and persistent. California’s load was high at the same time neighboring regions needed their own resources. Some resources included in monthly adequacy showings were on planned or forced outage. Others produced less than their qualifying amount in the critical evening window. Demand response did not appear in real time at the full quantity implied by its planning credit.

This does not make reserve margins useless. It shows that their assumptions need audit. The California Energy Commission forecasts demand and allocates obligations through processes that combine historical weather, economic conditions, energy efficiency, behind-the-meter generation, and other inputs. CPUC-jurisdictional load-serving entities procure qualifying capacity against their obligations. CAISO receives showings and operates markets and the grid. A failure can emerge from the interface even when each actor completes a formal step.

The governor’s August 17 investigation letter was pointed because the emergency contradicted public expectations about the available portfolio. The letter demanded an account of forecasting and planning, procurement, market practices, and operational decisions. Its criticism is a request for investigation, not a technical finding by itself. The later joint report provides the evidence and boundaries.

A modern reserve-margin audit should disclose the sensitivity of its result to a West-wide heat event, drought-reduced hydro, wildfire-related transfer limits, coincident generator derates, storage duration, and forecast error. It should show how many hours fall below the reliability target, not only the expected annual peak. It should also publish the corrective action that follows when the probability tail becomes more severe than the historic record used to estimate it.

The public should be able to distinguish three claims: the portfolio met the procurement rule; the portfolio was offered and available to the market; and the portfolio physically served load through the critical period. August 2020 showed that the first claim did not establish the third.

Regional imports were correlated, not independent insurance

California has long benefited from electricity imports. Geographic diversity can improve reliability because weather, hydrology, demand, and generator availability do not always move together. The August 2020 heat wave weakened that diversity. Neighboring balancing authorities faced high demand at the same time, limiting surplus energy. An import that is economically available on an ordinary summer day may not be firm during a West-wide emergency.

The CAISO, CPUC, and CEC response to the governor during the heat wave stated that regional conditions constrained energy supporting imports during late-afternoon and evening hours. It also rejected the claim that California’s clean-energy commitment, by itself, caused the rotating outages. That contemporaneous boundary is important: constrained imports and changing resource profiles were part of the event, but neither supports a single-cause narrative.

Import accountability begins with product definition. Is the import backed by a named physical resource or only by a seller’s portfolio? Is transmission reserved? Can the exporting balancing authority recall the energy during its own emergency? Does the capacity qualification reflect simultaneous regional stress? Is delivery tested at the net peak? A planning model that treats every import as statistically independent of California heat can overstate diversity.

Market schedules add another distinction. A day-ahead import schedule is an accepted commercial position. Real-time delivery depends on the exporting resource, transmission, intertie limits, and the external system’s conditions. Conversely, real-time Energy Imbalance Market transfers can provide valuable voluntary energy even when a forward schedule was not available. The joint report found that these transfers materially reduced the shortfall but could not restore the contingency reserve before the net peak.

The remedy is not autarky. Building every possible contingency inside one balancing area would be costly and can waste the diversity benefits of a western market. The remedy is to price and disclose firmness. Procurement should separate firm, resource-specific, deliverable imports from non-firm market opportunity. Stress tests should assume that uncontracted imports can fall sharply during a correlated event. Operators should retain the ability to use voluntary regional transfers without counting them twice as guaranteed planning supply.

When import assumptions fail, customers should be able to see whether the gap came from unavailable external energy, transmission, schedule priority, or an internal procurement choice. “Imports were lower” is an observation. Accountability requires the contract and control path behind it.

Capacity showings did not equal real-time production

California’s resource-adequacy program created obligations and showings intended to ensure that capacity was available to CAISO. The event exposed several reasons why shown capacity can diverge from real-time production. Some capacity was on outage. Some was derated by heat. Some had bidding or must-offer limits. Variable resources produced according to weather rather than their monthly qualifying value. Demand-response availability and delivery differed from planning credits.

The CPUC Safety and Enforcement Division’s generation-outage investigation examined forced outages of two hours or longer and greater than 50 megawatts during the heat period. Its redacted public report described more than 2,700 megawatts of derates across resource types on August 14 and 15 under its methodology, including substantial natural-gas outages. It also preserved an important limit: the existence of outages did not mean one generator or fuel caused the rotating outages.

Extreme heat reduces the capability of some thermal generators. Air is less dense, cooling systems work under more difficult conditions, and components may reach operating limits. A nameplate or normal-temperature rating can therefore overstate hot-weather capability. Similar performance realism applies across the fleet: hydro depends on water conditions, solar on irradiance, wind on wind speed, storage on power and energy state, and demand response on customers that are available and able to curtail.

The California ISO Department of Market Monitoring’s third-quarter 2020 report adds market evidence. It found that high-load-hour resource-adequacy bidding and availability gaps were not confined to one class. Gas units represented a material share of unavailable RA capacity, with ambient derates accounting for about half of derated gas RA during the heat-wave period. Solar and wind also represented a share of RA capacity not available during high-load evening hours because production was predictably below output during the gross-peak hours used in qualification.

A credible reform needs availability data by hour, cause, and resource obligation. It should distinguish planned outage, forced outage, ambient derate, fuel or water limitation, transmission constraint, bid omission, dispatch instruction, and actual output. It should also show replacement responsibility: when an RA resource takes a planned outage, who must procure substitute capacity, by when, and for which hours?

The accountability unit is not a public list of “bad plants.” It is a closed loop from capacity credit to actual performance and corrective action. A plant can have a legitimate safety outage and still create a portfolio gap that another actor must cover.

Day-ahead schedules concealed part of the operational need

The day-ahead market exists to commit and schedule resources before real time. Accurate load scheduling helps the market procure enough energy and lets CAISO commit additional units through residual unit commitment when day-ahead supply is limited public evidence. During August 2020, load-serving entities collectively under-scheduled demand in the day-ahead market on the critical days. Convergence bids and other market mechanics also affected the apparent balance.

Under-scheduling is not automatically misconduct. Forecasts change. Load-serving entities can use real-time markets, and incentives may differ across market designs. The accountability issue is systemic: in scarce conditions, aggregate under-scheduling can make the day-ahead market appear better supplied than the physical system will be. If the operator’s residual commitment process does not adjust enough, fewer resources may be positioned before the evening ramp.

Federal regulators examined the same interface. The FERC staff preliminary-observations proceeding reviewed demand, outages, resource adequacy, scheduling, and market performance. Its associated technical slides were staff observations, not a final enforcement judgment. They are useful because they show the event’s causes crossing state planning and federally regulated wholesale-market boundaries.

The repair is not to require perfect forecasts. It is to create incentives and operator tools that make aggregate schedule error visible early. CAISO should know the distribution of error by load-serving-entity class, hour, and weather condition. Market entities should face a credible consequence when persistent under-scheduling transfers commitment risk to real time. At the same time, rules must avoid punishing reasonable forecast error so harshly that entities over-schedule and create different inefficiencies.

Residual unit commitment should also be stress-aware. A model that waits for a large forecast error before committing additional generation may be too slow when start times, ramp rates, imports, and solar decline interact. Operators need authority and transparent criteria to position resources for a credible net-peak risk without converting every warm day into expensive over-procurement.

Accountability here is evidence of calibration: how much under-scheduling occurred, what additional commitment it suppressed, what rule changed, and whether later stressed days showed smaller errors and earlier resource positioning.

Exports were a priority and design question, not a simple culprit

Exports during a California emergency became a politically charged feature of the event. The intuitive claim is that electricity should not leave a system while local customers are being interrupted. The market and legal boundary is more complicated. Some exports may be supported by resources contracted for external load. Some are wheel-through transactions using the transmission system. Some schedules have priorities defined by tariff and reliability rules. Canceling them can transfer the emergency to neighboring systems, undermine contracts, or violate equal-treatment obligations.

The Final Root Cause Analysis found that day-ahead market practices involving exports and load scheduling aggravated stressed conditions. It did not establish that all exports were discretionary California energy or that canceling one category would by itself have prevented both outages. A responsible account must preserve that distinction.

CAISO’s later market-enhancements initiative addressed load, export, and wheeling priorities alongside scarcity pricing, demand response, storage, imports, Energy Imbalance Market sufficiency, and planned-outage substitution. The breadth of the initiative is evidence that no single export switch could close the gap.

A durable priority framework should answer four questions before a crisis. First, what physical resource supports each export? Second, is the transaction serving external load that has equivalent reliability standing? Third, how will priorities change as CAISO moves from normal conditions through alerts and emergency stages? Fourth, what information will be available to operators soon enough to act without improvising legal interpretations in real time?

The public ledger should show gross imports, gross exports, net imports, export categories, and curtailments. Netting can hide simultaneous flows with different firmness. A balancing area can be a net importer while still honoring exports backed by dedicated resources. Conversely, a schedule labeled as an export may depend on internal supply that is not physically incremental.

The accountability standard is advance clarity. If market entities know the priority rules, they can procure and schedule accordingly. If customers know which export categories remained and why, debate can focus on the actual tradeoff. Emergency discretion will still be needed, but it should operate inside a tested hierarchy rather than a post-event argument.

Demand response existed, but credited capacity and metered delivery diverged

Demand response can be an ideal net-peak resource because it reduces load at the moment the grid is strained. It avoids some transmission and distribution losses and can be faster than starting generation. Yet a planning credit is only useful if eligible customers are available, receive the dispatch, respond, sustain the reduction, and can be measured against a credible baseline.

CAISO dispatched large amounts of reliability demand response on both days. The joint report records substantial metered response from the investor-owned utilities’ emergency programs. It also shows much weaker performance from some proxy demand resources relative to credited or shown RA quantities, especially on August 15. Program rules constrained how early and how often certain resources could be called, forcing operators to preserve them for the period of greatest need.

The Department of Market Monitoring’s posting for its demand-response review states that demand response met about four percent of system RA requirements in August and September and that a large share was not available for dispatch across peak net-load hours. The underlying performance report examines bidding, dispatch, baselines, compensation, and incentives in greater detail.

Several quantities must not be confused. Credited capacity may include gross-ups for avoided losses and planning reserve margin. A bid is an offer to reduce. An award is a market instruction. Metered load drop is an estimate relative to a baseline, not a directly observed generator output. The baseline itself can be distorted by unusual weather, prior conservation, customer behavior, or adjustment caps. A resource can appear to underperform because the baseline is wrong, or appear to perform because ordinary demand happened to fall.

Repair should be tested before a Stage 3 emergency. Programs need representative hot-evening dispatches, customer and device telemetry, exception records, and sustained-performance tests. The operator should see availability by interval rather than a monthly block. Providers should disclose the population behind the aggregation, while protecting customer privacy, so planners can identify correlated non-performance.

Demand response is not imaginary generation, and it should not be treated as such. It is a service with a different evidence chain. The right response to 2020 was not to discard it, but to align qualification, availability, dispatch rules, measurement, and payment with the exact hours in which the system depends on it.

Accountability follows the control map, not the loudest institution

CAISO is the visible grid operator. It runs wholesale markets, dispatches resources, maintains system balance, declares emergency stages, and instructs distribution utilities to shed load. It does not own most generation, determine every state procurement requirement, forecast every load-serving entity’s retail demand, or select each distribution feeder that is interrupted.

The CPUC sets resource-adequacy requirements for its jurisdictional load-serving entities, authorizes procurement by investor-owned utilities, regulates retail service, and oversees covered generating facilities. The CEC develops demand forecasts and reliability analyses and has its own siting and policy responsibilities. Publicly owned utilities and other local regulatory authorities have separate procurement paths. Load-serving entities forecast and schedule their demand and procure qualifying capacity. Generator owners control maintenance and operation. Scheduling coordinators submit bids and instructions.

Demand-response providers control their enrolled portfolios and telemetry. Distribution utilities execute controlled outages on physical circuits.

That map prevents two opposite errors. The first is to blame CAISO for every unavailable megawatt because it issued the interruption order. The second is to treat the operator as a passive messenger because it did not own the assets. CAISO controlled market design proposals, operational forecasting, resource commitment, emergency communication, and the timing and magnitude of load-shed instructions. Those are material controls.

The California Energy Commission’s review of incremental resources for summer 2021 shows how repair crossed institutional lines. Temporary changes at existing plants, new supply, storage, and other measures required regulatory action, owner performance, market integration, and environmental boundaries. A megawatt authorized by one body still had to be built or enabled, qualified, scheduled, and delivered.

Shared responsibility should therefore produce named handoffs. Who owns the demand forecast version? Who validates the critical net-peak hour? Who confirms replacement capacity for an outage? Who determines import firmness? Who observes battery state of charge? Who can compel an RA resource to bid or explain a failure? Who verifies demand-response baselines? Who informs essential services before a rotating block is opened?

If each answer ends with “another agency,” the system has recreated the 2020 gap. A control map is valuable only when every handoff has an accountable owner, a deadline, and evidence that the next owner received usable information.

Rotating outages were not Public Safety Power Shutoffs

California customers were already familiar with Public Safety Power Shutoffs, in which a utility de-energizes lines to reduce wildfire ignition risk. The August 2020 interruptions were different. CAISO ordered controlled load shed because bulk-system supply and reserves were inadequate. Utilities then used rotating-outage blocks on their distribution systems. The purpose, trigger, expected duration, and restoration logic differed from a PSPS.

The CPUC’s rotating-outage guidance makes that distinction explicit and tells customers to prepare for temporary service interruption during grid shortages. Conflating the two mechanisms damages accountability. A PSPS review asks whether fire risk justified de-energization and whether utilities managed warning, access, and restoration. A rotating-outage review asks why sufficient system resources and reserves were unavailable and how load-shed blocks were chosen.

Distribution execution deserves its own audit. CAISO instructions are in megawatts, but utilities interrupt circuits. A circuit may contain homes, businesses, traffic controls, communications equipment, medical users, water facilities, and distributed generation. Some essential facilities may be exempt or placed on protected circuits, but physical network topology can make perfect targeting impossible. Actual megawatts differ from requested amounts because block load changes with time and weather.

The public needs to know whether utilities rotated interruptions as intended, whether any block stayed off longer than planned, which critical facilities lost service, and how medical-baseline or access-and-functional-needs customers were supported. It also needs an equity analysis. Repeatedly using the same circuits can concentrate risk in communities with fewer resources, while broad exemptions can leave too little interruptible load and force deeper outages elsewhere.

These questions do not imply that utility operators acted improperly in 2020. The joint report generally shows utilities executing emergency instructions under rapidly changing conditions. The point is that controlled load shed is itself a resilience system. It requires current feeder maps, tested communications, restoration discipline, and customer continuity measures before CAISO calls for it.

Emergency improvisation showed both capability and planning debt

After the first two outages, state and grid officials expected even larger shortfalls from August 17 through 19. The governor issued emergency actions, agencies coordinated with utilities and large customers, additional supply and demand reductions were sought, and public conservation campaigns intensified. Further rotating outages were avoided.

The governor’s executive-order announcement records temporary measures intended to free additional capacity, including use of backup resources that would not ordinarily operate in the same way. These steps demonstrate useful emergency power. They also expose planning debt: if reliable service depends on extraordinary waivers, urgent calls, and last-minute conservation, the ordinary framework did not fully cover the stress case.

Emergency measures carry costs and boundaries. Backup generators may have higher emissions. Industrial or water loads may defer operations. Consumers who pre-cool homes or reduce use may face discomfort or health constraints. Utilities and agencies can incur unplanned expense. None of these costs means the action was wrong; they should be recorded so that the system does not count emergency improvisation as free capacity.

Communication was a resource. Flex Alerts and public appeals reduced demand and bought operating margin. Yet voluntary conservation is difficult to forecast and verify. Households have different ability to respond. A renter in an inefficient building during a heat wave may have far less flexibility than a large facility with automated controls and backup power. The public appeal should therefore sit behind contracted, measurable resources, not replace them.

The closeout file for August 17–19 should ask what each intervention delivered by interval, what legal authority enabled it, what it cost, and whether it was later converted into a durable program. A temporary generator that produced 50 megawatts is not the same as a press release promising 50 megawatts. A conservation estimate should show the counterfactual method and uncertainty.

Emergency success is evidence of coordination capacity. It is not proof that the original plan was adequate. A mature system learns from the improvisation and decides which capabilities belong in the standing portfolio.

Procurement repair had to target the evening, not merely add nameplate

The CPUC opened emergency reliability rulemaking R.20-11-003 after the event. Its summer-reliability record shows decisions and proposals intended to add resources for 2021 through 2023, including supply, storage, and demand-side measures. The key design choice was availability during the peak and net-peak window.

The Commission directed the three large investor-owned utilities to seek additional capacity, and its emergency-procurement advice-letter page preserves the implementation trail. That is better accountability than a single procurement total because it allows reviewers to trace authorization into proposed contracts and approvals.

Still, procurement quantity can conceal quality. A resource should be evaluated for start time, ramp rate, minimum run, duration, fuel or energy limit, location, transmission deliverability, outage history, ambient derate, and ability to perform across consecutive days. A four-hour battery can be exceptionally valuable during an evening ramp, but its contribution depends on charge opportunity and state-of-charge management. A demand-response contract depends on enrolled load and dispatch rights. A thermal unit depends on maintenance, fuel, cooling, and emissions permissions.

The CEC’s 2021 and 2022 reliability analysis adopted more severe assumptions, including the possibility that uncontracted imports would not be available during West-wide heat or wildfire constraints. It estimated potential shortfalls under extreme conditions and included drought effects on hydro. Such analysis is not a prediction that a blackout will occur. It is a planning stress test.

Procurement accountability needs a reconciliation table: authorized megawatts, contracted megawatts, online date, qualifying capacity, critical-hour capability, actual availability, and customer cost. Delayed projects should remain visible rather than being silently replaced in a headline total. Temporary resources should be labeled temporary. Costs should be compared with the reliability contribution they actually supplied.

The goal is not to buy every proposed megawatt. It is to make the portfolio’s hour-by-hour risk legible and close the highest-consequence gaps with resources that can be verified before customers are asked to rely on them.

Storage was part of repair, not a retrospective absolution

Battery storage grew rapidly after 2020 and is well suited to shift daytime energy into the evening. Storage can also provide fast reserves and help manage steep ramps. It does not eliminate the need for forecasting, transmission, flexible demand, firm imports, or other generation. Nor should later storage success be used to claim that one technology would certainly have prevented every 2020 event under every dispatch strategy.

Storage accountability begins before discharge. What state of charge is required before the net peak? Who can charge the battery when day-ahead prices, real-time prices, and reliability needs differ? How much duration remains after reserves are supplied? Does a resource have a charging constraint or local transmission limit? Can it perform on consecutive hot days? How are degradation and forced outages reflected in qualifying capacity?

The joint report used illustrative battery behavior to show that charging and discharging decisions can either help or worsen a tight interval. The market enhancements that followed addressed storage operation during stressed conditions. That is the correct lens: batteries are controllable resources, and the system should specify the evidence needed to count them.

Later performance offers a useful but bounded test. A CAISO filing on the September 2022 heat wave reports that the system served a record peak near 52,061 megawatts without ordered rotating outages. It attributes the outcome to several factors, including additional resource-adequacy procurement, new resources, more than 3,500 megawatts of lithium-ion battery storage, conservation, and enhanced coordination. That is evidence of improved capability, not a controlled experiment proving that any one reform caused the result.

The 2022 event also warns against declaring the problem solved. A different heat pattern, wildfire, drought, generator outage set, or import condition can create another combination. Lithium-ion storage has duration and safety limits. Conservation may not repeat at the same level. New load from electrification and data centers changes the forecast.

Verifiable repair therefore uses repeated performance, not one victory. Each stressed season should report storage availability, charge sufficiency before critical hours, dispatch compliance, reserve provision, forced outage, and energy remaining after the peak. The same standard should apply to every resource class in terms appropriate to its physical behavior.

Market reform must show behavior changed under stress

CAISO’s market-reform work addressed load and export priorities, import incentives, scarcity pricing, demand-response dispatch, storage, planned outages, and Energy Imbalance Market sufficiency. FERC approved relevant tariff changes. Formal approval is an important governance milestone. It is not proof that entities and operators behaved differently when conditions tightened.

The evidence should be comparative. Did day-ahead under-scheduling fall on later high-risk days? Were more RA resources bid through the net peak? Did residual unit commitment position adequate capacity earlier? Were planned-outage substitutions completed? Did import schedules become firmer? Were exports categorized and curtailed according to the new priority rules? Did demand-response availability align more closely with credited capacity? Did storage enter the evening with a sufficient charge?

Market rules also create distributional effects. Paying more for scarcity can attract imports and reward available resources, but it raises costs and may not produce physical supply when the entire West is short. Penalties can improve performance, but poorly designed penalties can drive resources out of the RA program or encourage conservative declarations. Procurement mandates can increase reliability while transferring costs to customers for years.

This is why a repair ledger must include cost as well as reliability. Customers should see the incremental cost of emergency procurement, market uplift, new reserves, and storage alongside the risk reduction. Cheap but unavailable capacity is not cheap. Expensive capacity that rarely runs may still be justified if it covers a high-consequence tail, but the value should be demonstrated.

Governance also changed. The event strengthened the argument for tighter coordination among planning, procurement, and operation. Yet coordination can blur accountability if decisions become collective without a recorded owner. Every joint process should preserve which agency approved the forecast, which entity procured the resource, which rule governed performance, and which operator accepted the remaining risk.

The test is observable behavior at the hour of need. A revised tariff, a signed contract, or a new committee is an input. The outcome is adequate reserves without uncontrolled or controlled customer interruption under the stress case the reform claimed to address.

Unknowns should remain visible

The public record is detailed but not complete. It does not provide a single, final causal allocation among every load-serving entity, generator, scheduler, demand-response provider, and agency. Redactions in the generation investigation limit plant-specific detail. Commercial contracts and resource-specific bidding information are not all public. Customer interruption tables do not reveal every circuit or every secondary consequence.

The record also does not support several common claims. It does not establish that renewable generation alone caused the outages. It does not establish that all exports were improper. It does not show that one gas-plant outage caused either emergency. It does not prove that every community-choice aggregator or other load-serving entity under-scheduled unlawfully. It does not provide a verified outage-attributed mortality or statewide loss figure.

Some findings remain supported inference rather than adjudication. A more granular hourly RA framework would likely have exposed evening risk earlier, but the precise portfolio that would have prevented each interruption depends on bids, outages, imports, dispatch, and customer response. More storage would likely have helped, but its effect depends on state of charge and dispatch. More firm import procurement could have helped, but regional sellers and transmission needed to be capable of delivery.

Transparency should not require publication of information that creates security, privacy, or market-manipulation risk. Plant details, critical-facility feeder maps, and individual customer data need protection. Aggregated evidence can still show performance, exceptions, corrective actions, and the distribution of customer burden.

Uncertainty is part of accountability when it is bounded. The agencies did not need to prove a single root cause because the system failed through interacting controls. They did need to separate observed facts from recommendations and to preserve the places where evidence was unavailable. The final report generally does that. Future reporting should continue the discipline rather than converting later success into a claim that every causal gap has closed.

A verifiable net-peak ledger

California’s repair can be judged with a compact set of measures.

First, planning: publish the loss-of-load risk by hour under correlated heat, low hydro, wildfire constraints, realistic thermal derates, limited uncontracted imports, and storage duration. Show forecast error and the sensitivity of the result to climate-informed weather.

Second, procurement: reconcile obligations, shown capacity, firm imports, replacement capacity, online dates, qualifying values, and critical-hour deliverability. Do not count an authorization as an operating resource.

Third, markets: report day-ahead schedule error, residual commitment, RA bidding, exports by priority category, imports by firmness, Energy Imbalance Market transfers, scarcity intervals, and deviations from dispatch.

Fourth, physical performance: record planned and forced outages, ambient derates, start failures, ramp limits, storage state of charge, demand-response availability and metered delivery, transmission constraints, and contingency reserves.

Fifth, customer continuity: disclose requested and actual load shed, customer-account interruptions, duration distribution, restoration error, essential-service impacts, medical and access support, repeat-circuit exposure, and after-action fixes. Preserve the distinction between rotating outages and wildfire-related shutoffs.

Sixth, governance: name the owner, deadline, evidence source, exception, and closure test for every corrective action. A recommendation is open until a test demonstrates performance. An agency should not close an item merely because a rule was adopted if implementation data are still missing.

These measures turn “resource adequacy” from a monthly compliance noun into an operating claim. They also create a fairer allocation of responsibility. A generator is judged on availability and truthful outage reporting. A load-serving entity is judged on procurement and scheduling. A demand-response provider is judged on measured delivery. A utility is judged on controlled-outage execution. CAISO is judged on markets, forecasts, commitment, reserves, and emergency operation. CPUC and CEC are judged on requirements, approvals, forecasts, and oversight.

The point is not to guarantee that California will never shed load. No electricity system can economically eliminate every conceivable emergency. The point is to show that known failure combinations have been reduced, residual risks are visible, and emergency burdens are managed deliberately.

Accountability lives in the hours after the headline peak

The August 2020 outages are easy to remember as a clash among heat, renewables, imports, gas plants, and market rules. That shorthand misses the deeper lesson. The system’s formal capacity picture looked stronger than the deliverable portfolio at the exact evening intervals when reserves mattered. Multiple actors had completed pieces of a process, yet their combined evidence did not prevent controlled interruption.

CAISO was right to shed load rather than gamble with wider instability once reserves were exhausted. The agencies were right to reject a single-cause explanation. The post-event procurement, market, forecasting, demand-response, storage, and coordination reforms addressed real gaps. Later performance under record demand supplies encouraging evidence.

But accountability is not reassurance by accumulation. More megawatts, more pages of rules, and more committees do not by themselves prove resilience. The proof is an hour-by-hour chain from forecast to contract, bid, physical delivery, reserve, customer service, and corrective action. It must survive the correlated conditions that make ordinary assumptions fail.

California’s net peak is therefore both a technical curve and a governance boundary. As the sun falls, the system reveals which capacity is real, which imports are firm, which customers can respond, which assets can sustain output, and which institutions can coordinate without relying on improvisation. That is the moment at which resource adequacy becomes accountable.