The Gold Standard of Utilities | Relating SAIFI, SAIDI and CAIDI Without Mixing Customer Counts and Durations

Begin with three totals for the same reporting scope. Customer interruptions, CI, add one for each affected customer in each qualifying sustained interruption. Customer interruption minutes, CMI, add the duration experienced by every affected customer. Customers served, N, is the declared denominator population. Then SAIFI = CI/N and SAIDI = CMI/N.

CAIDI = CMI/CI, the average restoration duration across counted customer interruption episodes. If CI is nonzero and every total uses the same ledger, scope and definitions, then SAIDI/SAIFI = (CMI/N)/(CI/N) = CMI/CI = CAIDI. That identity is an audit check, not a guarantee that the underlying ledger was classified correctly.

Contents · Next chapter

SAIFI is commonly expressed as interruptions per customer served for the reporting period. SAIDI is commonly expressed as minutes, or sometimes hours, per customer served. CAIDI is minutes or hours per customer interruption. The word “customer” refers to the reporting convention’s service entity, not automatically to an individual person, meter, household or account.

Write the units beside every value. A figure of 1.4 could mean interruptions per customer, hours per customer or hours per interruption. A table that drops units invites an accidental comparison. If the source reports SAIDI in hours and CAIDI in minutes, convert deliberately before testing their relationship.

Contents · Previous chapter · Next chapter

Assume a utility serves 1,000 customers. Event A interrupts 100 customers for 60 minutes. Event B interrupts 50 customers for 30 minutes. Event C interrupts 200 customers for 15 minutes. Treat these as qualifying sustained interruptions with no overlap for any customer in this first example.

The customer-interruption total is 100 + 50 + 200 = 350. The customer-minute total is 100×60 + 50×30 + 200×15 = 6,000 + 1,500 + 3,000 = 10,500 customer-minutes. Preserve both totals. Adding event durations alone, 60 + 30 + 15, would ignore how many customers experienced each event.

Contents · Previous chapter · Next chapter

For the ledger, SAIFI = 350 customer interruptions / 1,000 customers served = 0.35 interruptions per customer served. This is a population average. It does not mean that 35 percent of customers necessarily had exactly one interruption. Some customers could appear in more than one event while others had none.

The numerator is not simply the number of incidents or fault records. Three events created 350 customer interruption episodes. If one equipment failure affected 500 customers, its event count could be one while its CI contribution was 500. Keep engineering incident counts separate from customer impact counts.

Contents · Previous chapter · Next chapter

SAIDI = 10,500 customer-minutes / 1,000 customers served = 10.5 minutes per customer served. Again, this is an average across the whole denominator, including customers with no qualifying interruption. It does not state that each customer waited 10.5 minutes.

The weighting matters. Event A contributes 6,000 customer-minutes, twice Event C’s 3,000, even though Event C affects twice as many customers. Duration and reach work together. A report that lists event durations without affected-customer counts cannot reconstruct SAIDI.

Contents · Previous chapter · Next chapter

CAIDI = 10,500 customer-minutes / 350 customer interruptions = 30 minutes per customer interruption. The calculation weights each customer interruption episode, not each event equally. An unweighted average of the three event durations would be (60 + 30 + 15)/3 = 35 minutes, which answers a different question.

The difference is useful. The 15-minute event affects 200 customers and therefore contributes many more interruption episodes than the 30-minute event affecting 50. CAIDI’s episode weighting gives the larger event appropriate influence. If the intended question is the average duration of a utility incident, build an event-level measure and label it separately.

Contents · Previous chapter · Next chapter

Divide SAIDI by SAIFI: 10.5 minutes per customer / 0.35 interruptions per customer = 30 minutes per interruption. The customers-served denominator cancels. This matches CMI/CI and confirms that the three summaries are internally consistent for the stipulated ledger.

If published values fail to match exactly, investigate units, rounding and scope before declaring an error. Annual indices may be rounded independently. One figure may exclude major event days while another includes them. A denominator may have changed. The identity is strongest when applied to unrounded values generated from the same record set.

Contents · Previous chapter · Next chapter

In the example, 1,000 customers are served while the ledger contains 350 customer interruption episodes. The number of distinct customers interrupted could be at most 350 and could be lower if people appear in multiple events. Using 350 as the denominator would redefine the metric around affected episodes rather than the served population.

That alternative may answer a useful question, such as impact among affected customers, but it is not the same index. Store the served population, distinct affected customers and interruption episodes in separate fields. Their labels should survive into charts and exports.

Contents · Previous chapter · Next chapter

Suppose one customer is interrupted in Event A and again in Event C. That customer contributes two to CI. If the durations are 60 and 15 minutes with no overlap, the customer contributes 75 to CMI. A distinct-customer count would still be one.

This distinction explains why SAIFI can exceed one. It is an average number of qualifying interruptions per customer served, not a proportion bounded by 100 percent. Converting SAIFI of 1.2 into “120 percent of customers had an outage” is invalid without the distribution of repeat interruptions.

Contents · Previous chapter · Next chapter

For every event-customer record, multiply the qualifying duration by one affected customer, then sum. In aggregated form, multiply event duration by affected-customer count only when every customer in that row shares the same start and restoration time. Staggered restoration needs more detailed rows.

If 100 customers are restored after 20 minutes and another 100 after 50 minutes, one 50-minute duration for all 200 overstates CMI. The correct contribution is 100×20 + 100×50 = 7,000 customer-minutes. Restoration steps belong in the ledger.

Contents · Previous chapter · Next chapter

Two event tickets can overlap for the same customer. Adding both durations can double-count physical interruption time, depending on the reporting convention and event construction. First build an interval timeline by customer or by a group known to share status, merge overlapping physical intervals where required, and only then aggregate.

The existing overlapping interruptions guide develops this interval problem. The present guide begins after the qualifying episodes and customer-minutes have been established. Linking the two prevents an index identity from hiding a duplicated ledger.

Contents · Previous chapter · Next chapter

Reliability reporting often distinguishes momentary interruptions from sustained interruptions using a declared duration rule. Do not infer the boundary from ordinary language. A brief interruption may count in one metric family and not another, and different standards or jurisdictions may apply different conventions.

Store the raw times and the classification rule version. If the boundary changes, rerun classification rather than editing a summary by hand. The formulas remain simple; the difficult evidence often lies in deciding which records qualify.

Contents · Previous chapter · Next chapter

Some reliability reports show values including major event days and values excluding them under an adopted method. Exclusion changes CI and CMI, and sometimes the period’s interpretation, but it does not change the basic formulas. A label such as “excluding MEDs” is therefore part of the metric identity.

Never compare one utility’s all-events SAIDI with another utility’s exclusion-based SAIDI as though scope were identical. Check the governing reporting notes. When presenting both, calculate both from traceable record sets and explain what decision each view supports.

Contents · Previous chapter · Next chapter

Maintenance outages and unexpected faults can be grouped or separated depending on the reporting framework. A customer experiences both as lost service, but operational accountability may ask different questions. The calculation should follow the declared category rather than silently removing an inconvenient event.

Keep a classification column and publish the inclusion rule. If a summary excludes planned work, say so near the number. A later analyst can then construct an all-interruptions view without reconstructing lost records.

Contents · Previous chapter · Next chapter

Modify Event A. Fifty customers are restored after 30 minutes and fifty after 90 minutes. Event A still contributes 100 customer interruption episodes, but its CMI becomes 50×30 + 50×90 = 6,000, the same as the original 100×60 example. SAIFI, SAIDI and CAIDI therefore remain unchanged.

The distribution is very different. Half the affected customers experienced a much longer interruption. This shows a limit of aggregate indices: equal totals can hide unequal experiences. Add percentile, maximum or customer-level distribution evidence when the decision concerns extremes or fairness.

Contents · Previous chapter · Next chapter

Now suppose the same 100 customers experience two separate 15-minute interruptions. Their contribution is CI = 200 and CMI = 3,000. A single 30-minute interruption to those 100 customers would give CI = 100 and the same CMI = 3,000.

SAIDI is unchanged if the rest of the ledger and denominator stay fixed, but SAIFI is higher and CAIDI lower for the repeat case. The indices distinguish frequency from average duration, yet neither fully describes disruption patterns. A process sensitive to restarts may care greatly about two interruptions even when total lost minutes match.

Contents · Previous chapter · Next chapter

Suppose the same 350 CI and 10,500 CMI are divided by 1,100 customers served after an annexation or account change. SAIFI becomes about 0.318 and SAIDI about 9.545 minutes. CAIDI remains 30 minutes because it does not use N.

This is why year-to-year improvements in SAIFI or SAIDI can partly reflect denominator change. Preserve the population definition and effective date. A stable CAIDI does not prove restoration practice was unchanged; it only shows the ratio of the reported CMI and CI stayed the same.

Contents · Previous chapter · Next chapter

If the source supplies CI, CMI and N, calculate from those totals. Avoid reconstructing CAIDI from SAIDI and SAIFI that have each been rounded to one decimal place. For example, small denominators or low event counts can make rounding error material.

Publish a reasonable number of digits and keep the unrounded computation internally. A footnote can state the rounding rule. False precision is unhelpful, but so is rounding so aggressively that the identity appears broken.

Contents · Previous chapter · Next chapter

An interruption still open at the reporting cut-off has no final duration. Assigning zero understates CMI; assigning the current elapsed time changes the meaning from completed duration to duration observed so far. Mark the record as open and declare the treatment.

For operational dashboards, current customer-minutes out may be useful. For a completed annual index, the reporting rule may allocate or update the event differently. Do not improvise across periods. Preserve the raw start, restoration status, last observation and later correction history.

Contents · Previous chapter · Next chapter

Start and restoration timestamps can come from devices, operators or customer systems with different clocks. A five-minute offset matters for short events and restoration sequencing. Record time zones, synchronisation status and resolution.

When exact restoration time is inferred from polling intervals, describe the rule. A midpoint estimate and an end-of-interval assignment produce different CMI. The summary should not appear more exact than the telemetry supports.

Contents · Previous chapter · Next chapter

Customers can transfer between circuits, merge accounts or change identifiers. A reliability ledger needs a stable way to count the service entity intended by the reporting convention. Duplicate identifiers inflate affected counts; missing mappings lose impact.

Run integrity checks before calculating indices: unique current denominator records, valid event-customer joins, no impossible restoration-before-start times and traceable identifier changes. These checks are mundane but often more consequential than the final division.

Contents · Previous chapter · Next chapter

Compare the customer count derived from the interruption management system with telemetry, call records or restoration logs where appropriate. Differences do not automatically identify which system is right; they reveal a reconciliation task.

Document accepted tolerances and unresolved gaps. If an estimate replaces missing customer detail, label the method and keep it separate from direct observations. An index can be calculated from estimated inputs, but the evidence grade should remain visible.

Contents · Previous chapter · Next chapter

Do not average area-level SAIDI values equally unless equal weighting is the intended question. To aggregate areas under compatible definitions, add their CMI and divide by the sum of their customers served. This is a customer-weighted mean.

The same principle applies to SAIFI using CI. CAIDI for the combined scope is total CMI divided by total CI, not the equal average of area CAIDI values. Reconstruct from numerators and denominators whenever possible.

Contents · Previous chapter · Next chapter

SAIDI of 10.5 minutes could arise because everyone loses service briefly or because a small group experiences a long outage. SAIFI of 0.35 could arise from widely distributed single events or repeated events in one area. CAIDI of 30 minutes similarly hides the spread of restoration durations.

Add distributional measures aligned with the question: distinct customers affected, repeat-interruption counts, customer-level total duration, restoration percentiles and geographic concentration. Keep them as complements rather than relabelling them as SAIDI, SAIFI or CAIDI.

Contents · Previous chapter · Next chapter

Network design, density, vegetation, weather and reporting systems differ across territories. The indices describe outcomes under those conditions; they do not by themselves isolate management quality or causal performance. A ranking without context can overclaim.

Use like-for-like scopes and explain major structural differences. For internal learning, compare an area’s current ledger with its own prior periods under stable definitions, then investigate the event categories driving change. Causal claims need more evidence than an index table.

Contents · Previous chapter · Next chapter

For each interruption segment, keep event ID, customer group, number affected, start, restoration, qualifying minutes, CI contribution and CMI contribution. Sum CI and CMI at the bottom. Store the customers-served denominator separately with its effective period and definition.

The table should allow a reviewer to reproduce every numerator. Avoid embedding formulas only in hidden spreadsheet cells. A calculation note can state inclusion rules, time treatment, exclusions, units and rounding. This small package turns a headline metric into inspectable evidence.

Contents · Previous chapter · Next chapter

CMI cannot be negative. CI and N should be nonnegative integers under ordinary customer-count conventions. SAIDI/SAIFI should equal CAIDI within declared rounding when SAIFI is positive and scope matches. The sum of area numerators should match the system numerator when categories are mutually exclusive and exhaustive.

Flag a zero CI separately because CAIDI is then undefined rather than zero. A zero-duration interruption may signal a classification or timestamp issue. Automated checks should route anomalies to review instead of silently coercing them into acceptable-looking values.

Contents · Previous chapter · Next chapter

CAIDI is a customer-interruption-weighted average duration. It is not the median event duration, the longest restoration time or the duration every customer experienced. Large events exert more influence because they create more customer interruption episodes.

Use the precise phrase “average duration per customer interruption within the declared scope.” If the audience needs event restoration performance, publish event-level distributions too. One measure cannot serve every operational question.

Contents · Previous chapter · Next chapter

SAIFI can exceed one because one customer can contribute multiple interruption episodes. A percentage of distinct customers affected is bounded by 100 percent and uses a different numerator. The two may move differently when repeat interruptions concentrate in one group.

Report both when useful, but never convert between them without the customer-event incidence data. A count of 350 episodes among 1,000 customers does not reveal whether 350 distinct customers had one event or 175 had two.

Contents · Previous chapter · Next chapter

Adding the clock duration of events ignores reach. SAIDI starts from customer-minutes, not event-minutes. A 10-minute interruption to 10,000 customers contributes more CMI than a 60-minute interruption to 100 customers.

Conversely, multiplying the longest event duration by all served customers would overstate impact. Construct the customer exposure first. The average follows from the ledger; it should not be estimated from a handful of visible events when detailed impact records exist.

Contents · Previous chapter · Next chapter

The broad accounting idea transfers: distinguish service incidents, affected service points, duration exposure and denominator population. But the acronyms and qualification rules may not transfer unchanged. Water pressure loss, telecom access and district cooling have different service boundaries and restoration definitions.

Start with the service’s own standard and data model. Use the electricity example to ask better questions, not to impose an unrelated metric. A learner demonstrates transfer by rebuilding the ledger and units for the new service while preserving the distinction between frequency, duration and affected population.

Contents · Previous chapter · Next chapter

Contents · Previous chapter · Next chapter

Consider a second fictional system serving 2,000 customers. Event D interrupts 300 customers for 20 minutes, Event E interrupts 100 for 90 minutes, and Event F interrupts the same 100 customers again for 10 minutes on another day. CI is 300 + 100 + 100 = 500. CMI is 300×20 + 100×90 + 100×10 = 16,000 customer-minutes. SAIFI is 0.25, SAIDI is 8 minutes and CAIDI is 32 minutes. The identity 8/0.25 = 32 checks.

Now combine E and F into one uninterrupted 100-minute episode for those 100 customers. CI falls from 500 to 400 while CMI remains 16,000. SAIDI stays 8 minutes, SAIFI falls to 0.20 and CAIDI rises to 40 minutes. The service pattern changed from two interruptions to one, even though total customer-minutes did not. This shows why frequency, total duration and average episode duration must remain separate.

Next suppose the reporting convention excludes Event E as a major-event-day record but includes D and F. Recalculate from the filtered ledger instead of subtracting a rounded published index. State the exclusion in the title and keep the all-events view. The exercise tests scope control as well as arithmetic: a learner must preserve the qualifying records, denominator and units, then explain which customer experiences the averages still hide.

Ask which events qualify, which customers form N, whether major events and planned work are included, how overlaps and staged restoration are handled, what units are used and whether values are rounded. Request CI and CMI when the decision needs verification.

Then test CAIDI = SAIDI/SAIFI, inspect year-to-year definition changes and review distributional evidence. A failed identity is a prompt to reconcile, not immediate proof of misconduct. A successful identity confirms internal arithmetic but not the accuracy of the underlying events.

Contents · Previous chapter · Next chapter

Contents · Previous chapter · Next chapter

Preserve the qualifying event ledger, customer-segment restoration records, CI and CMI totals, customers-served denominator, period, units, inclusion rules, major-event treatment, planned-event treatment and calculation version. Keep the unrounded values behind every public index. A summary table without these fields may communicate a result but cannot support reconciliation.

Add the distinct customers affected and at least one distributional view when the audience could mistake an average for every customer’s experience. Record revisions and the reason for each change. This package lets another reviewer reproduce the three indices, test CAIDI = SAIDI/SAIFI, inspect scope and identify which impact questions still require finer data.

For teaching, include one counterexample in which equal SAIDI values arise from different frequencies and restoration durations. Require the reader to describe both ledgers rather than rank them from one average. This demonstrates that the record supports calculation and exposes the information the indices compress. It also prevents a neat identity from being mistaken for a complete picture of reliability.

Keep the event rows visible throughout that comparison and review.

Can SAIFI be zero? Yes when no qualifying customer interruptions occur. Then SAIDI is also zero for that ledger, while CAIDI is undefined because CI is zero. Can CAIDI fall while SAIDI rises? Yes if frequency rises enough or the mix of interruption episodes changes. Does excluding a major event always improve all indices? It reduces included CI and CMI, but the resulting ratios depend on the removed event mix.

Should distinct customers interrupted replace CI? No; they answer different questions. Can I average monthly SAIDI values? Only with compatible numerators and denominators; reconstruct annual CMI and N under the reporting convention. Does one low annual value prove resilience? No. The index is an outcome summary, not a complete causal or risk assessment.

Contents · Previous chapter · Next chapter

CAIDI concentrates on the duration of counted interruption episodes, while SAIFI captures how often they occur across the served population. A programme can improve restoration and reduce CAIDI without reducing event frequency. Another programme can prevent interruptions and lower SAIFI while the remaining rare events are long, raising CAIDI.

Evaluate the intervention against the mechanism it targets. Vegetation management, automation, spare equipment and crew dispatch do not act on the indices in identical ways. A movement in one aggregate does not prove which intervention worked. Link event categories, operational changes and comparison periods before making causal claims.

Contents · Previous chapter · Next chapter

Annual average interruption indices describe observed service continuity over a period. Resilience questions often concern the ability to prepare for, withstand and recover from severe events. A system with favourable routine averages may still have vulnerable critical dependencies, while a severe year may dominate a long-run average.

Use scenario, asset and recovery evidence alongside historical indices. Do not rename one annual SAIDI result as a complete resilience score. The index can contribute evidence about consequences and restoration, but resilience claims require a wider system model.

Contents · Previous chapter · Next chapter

A trailing twelve-month dashboard and a calendar-year report can cover different events even when viewed on the same day. Their denominators may also reflect average customers, year-end customers or another convention. Label period start, period end and denominator rule.

When a major event leaves the rolling window, the metric can improve abruptly without a new operational change. A timeline of included events explains the step. Never join rolling and fixed-period values into one trend without marking the definition break.

Contents · Previous chapter · Next chapter

Utilities may revise affected-customer counts or restoration times after field reconciliation. Store the original observation, correction reason, approval and effective reporting version. Recomputing all indices from the corrected ledger is safer than patching a headline value.

If a public value changes, publish the revision date and scope. Version history prevents analysts from treating two extracts as independent observations. It also makes the final numerator defensible during audit.

Contents · Previous chapter · Next chapter

One service point can supply a household, business, institution or shared facility with very different numbers of people. Reliability indices based on customers served do not directly measure person-hours without additional demographic and occupancy evidence.

Avoid translating customer-minutes into human-minutes by assumption. For emergency planning, critical-load and population impacts may need separate datasets. Keep the standard customer index intact while adding the impact lens required for the decision.

Contents · Previous chapter · Next chapter

In a system serving few customers, one event can move annual SAIFI and SAIDI sharply. The arithmetic is correct, but year-to-year comparison has high event sensitivity. Show the underlying counts and perhaps a multi-year view rather than presenting a single percentage change as a stable trend.

Do not hide volatility by pooling incompatible systems. Explain scale, event mix and uncertainty. A small denominator is context for interpretation, not a reason to discard the record.

Contents · Previous chapter · Next chapter

If the served population changes during the year, the reporting convention may use an average, monthly values or another defined denominator. Using only the year-end count can bias the result. Consult the source’s method rather than inventing a denominator.

For an internal exploratory measure, one can sum monthly numerators and use compatible exposure weighting, but it should not be labelled as a standard annual index unless it follows the adopted definition. Document the temporal weighting explicitly.

Contents · Previous chapter · Next chapter

Event records may contain cause labels such as weather, equipment, vegetation or unknown. These are useful for decomposition, yet they can change after investigation and may mix immediate trigger with underlying condition. Keep code version and review status.

When comparing cause-specific indices, ensure events are assigned consistently and categories are mutually interpretable. “Unknown” should remain visible rather than redistributed to make totals look complete. Causal learning begins with honest classification limits.

Contents · Previous chapter · Next chapter

A bottom-up calculation sums every customer segment. A top-down operational dashboard may provide total customers out over time. Integrating that curve can approximate customer-minutes. Agreement between independent routes increases confidence; disagreement reveals timing, coverage or classification issues.

Define sampling and interpolation when integrating a curve. A five-minute polling interval cannot locate every restoration exactly. Use the comparison as a diagnostic, not a forced match.

Contents · Previous chapter · Next chapter

Give learners two ledgers with the same SAIDI but different SAIFI and CAIDI. Ask them to reconstruct CI, CMI and N, describe the customer experience each could represent, and choose one additional distributional measure. Then change N while holding numerators fixed and explain which indices move.

The assessment tests more than formula recall. A successful answer preserves units, scope and limits, recognises non-unique ledgers and refuses to infer a customer distribution from averages alone. That is the transfer standard for responsible use.

Contents · Previous chapter · Next chapter

Planning models may forecast customer interruptions and duration under assumed failure and restoration scenarios. Realised SAIFI, SAIDI and CAIDI are calculated from observed qualifying records. A forecast is not an observed index and should not be inserted into the historical series.

Use forecasts to compare scenarios and resource plans; use realised values to evaluate what occurred. When reviewing accuracy, preserve the assumptions and compare like scopes. A miss may arise from event frequency, event severity, customer exposure or restoration time, so decompose it before revising the model.

Contents · Previous chapter · Next chapter

Standard customer weighting gives one service point the same count contribution as another. Hospitals, emergency services and vulnerable users may require separate continuity measures and response priorities. Their importance should not be hidden inside a system average.

Keep the standard index for comparability and publish a protected, appropriately aggregated critical-service view where authorised. Do not secretly assign extra weight inside SAIDI and retain the same label. A changed weighting creates a different measure.

Contents · Previous chapter · Next chapter

Record the time to restore 50, 80, 90 and 100 percent of affected customers for large events, using a declared starting population. Two events can share CAIDI while one restores most customers quickly and leaves a small tail, and the other restores everyone gradually.

Milestones complement customer-minute totals by exposing the recovery curve. They still require accurate affected-customer and restoration records. Changing the starting count midway invalidates comparisons unless revisions are reconciled.

Contents · Previous chapter · Next chapter

Public reporting need not expose household-level outage histories. Aggregate numerators, denominators, inclusion rules, units and revision dates can support verification while customer records remain protected. Geographic detail should match privacy and security policy.

Do not use confidentiality as a reason to omit every method note. Method transparency and record confidentiality can coexist. The audience needs enough information to know what the index means and whether comparisons are valid.

Contents · Previous chapter · Next chapter

From the reported indices and denominator, one can reconstruct approximate totals: CI ≈ SAIFI×N and CMI ≈ SAIDI×N. From CAIDI and CI, CMI ≈ CAIDI×CI. Agreement among these routes within rounding is a strong coherence check.

The reconstructed totals may not be integers because published indices are rounded. Do not treat fractional reconstructed customers as raw records. Use them to identify large inconsistencies and then return to the unrounded source ledger.

Contents · Previous chapter · Next chapter

This guide uses a small fictional ledger with clean timestamps and classifications. Real reporting may follow IEEE, regulator, jurisdiction or utility-specific rules, and public datasets may revise records. Definitions, exclusions and units must come from the actual reporting source.

Use the arithmetic to audit coherence, not to bypass domain standards. Keep customer privacy and critical-infrastructure security in mind when publishing granular records. Share enough method for verification without exposing protected operational detail.

Contents · Previous chapter · Next chapter

Continue with the Gold Standard of Utilities for the subject map, the Gold Standard of Education for the transfer framework and the How X Works hub for connected mechanisms. The overlapping interruptions guide should be read first when the ledger contains overlapping customer intervals.

For United States reporting context, consult the US Energy Information Administration’s reliability definitions table, Form EIA-861 data page and Electric Power Annual technical notes. EIA’s discussion of major event days and recent reliability results shows why inclusion scope belongs beside every comparison.

Contents · Previous chapter · Return to Utilities