Evidence checked October 31, 2026. Numerical examples are hypothetical engineering illustrations, not descriptions of actual facilities. Proposed evaluation practices are editorial analysis rather than operating instructions or regulatory requirements.
1. The growing fleet brings a harder question
The US Energy Information Administration reported on August 7, 2026, that utility-scale battery nameplate capacity had reached nearly 52 gigawatts after additions during the first half of the year. Its analysis distinguished operational capacity from projects operators expected to build later. That expansion makes a familiar question more important: how much of the advertised power can a battery actually deliver when the system is under stress? EIA capacity analysis.
A large battery can be genuinely valuable without behaving like a continuously fueled generator. Its usefulness depends on the task. Supplying a sharp evening peak is different from supporting an isolated building for several days. Responding rapidly to a frequency disturbance is different from replacing a long interruption in energy supply. A project can perform one task exceptionally well while being poorly matched to another.
The word emergency often conceals these differences. A transmission operator may mean a regional shortage of available generation. A town may mean that a storm has severed its local distribution line. A hospital may mean loss of its normal electrical supply. A battery connected to an intact grid might help with the first situation but have no authorized or engineered way to energize the local circuit in the second. Identical headlines about backup power can therefore describe very different capabilities.
The right inquiry begins with the promised service. How much power, at which electrical boundary, for how long, starting under what conditions? Does the promise require the battery to be full when the event begins? Must it retain energy for a second event? Is charging available during the interruption? Which equipment has to remain operational for the battery to reach the customer?
These questions are more useful than treating batteries as either a universal solution or an unreliable distraction. Storage can move energy into periods when it is more valuable, provide fast response and support specific reliability objectives. Its constraints are measurable. Good planning makes those constraints visible and designs around them; weak planning hides them inside an attractive capacity total.
This article follows the technical chain from power and energy ratings to dispatch, degradation, system accreditation and actual emergency performance. The emphasis is on what evidence would establish the claim. No single percentage can replace that chain, because different failures can produce the same final symptom: the promised electricity is unavailable when someone needs it.
2. Megawatts describe a rate; megawatt-hours describe an inventory
A battery's power rating describes the rate at which it can deliver electricity. Its energy rating describes an amount. Dividing usable energy by rated power gives a simplified duration at that power. The grid-storage FAQ produced through the Greening the Grid program uses these distinctions and explains why state of charge affects the services available at a particular moment. Grid-scale battery FAQ.
Consider a hypothetical system rated at 100 megawatts with 400 megawatt-hours of usable energy measured at the same output boundary. Under idealized constant-power assumptions, it can deliver 100 megawatts for four hours. It could instead deliver 50 megawatts for eight hours if no other restriction or loss changes the calculation. Those alternatives are not two independent services available simultaneously. They spend the same stored-energy inventory at different rates.
Now suppose an emergency requires 120 megawatts for one hour. The system has enough energy in the abstract, but its 100-megawatt power limit prevents it from meeting that requirement alone. Conversely, a demand for 100 megawatts over six hours respects the power limit but requires 600 megawatt-hours, exceeding the assumed inventory. Power sufficiency and energy sufficiency are separate tests.
A duration label can become misleading when the underlying boundaries differ. If the energy figure describes cells on the direct-current side while the power figure describes the alternating-current point of interconnection, the simple ratio may omit conversion losses or other constraints. A reviewer should ask whether both ratings describe the same installed system and the same measurement conditions.
The phrase four-hour battery should therefore be treated as shorthand with a definition attached. Does it mean beginning-of-life usable energy divided by an inverter rating? A guaranteed net delivery at the grid meter? A market classification? A design target after future additions? The answers can differ without anyone having changed the physical equipment. The label alone does not resolve them.
Nor should duration be mistaken for a fixed discharge schedule. An operator may vary output every interval, preserving energy when demand falls and increasing it when demand rises. The relevant calculation is then accumulated energy across the dispatch sequence, subject to power limits and losses. A battery can serve a long event at partial output while still being unable to serve a shorter event with a higher peak.
This elementary distinction is the foundation of serious scrutiny. Before asking whether a project is large enough, decide which dimension of large matters: instantaneous support, total energy, or both. A public presentation should show both quantities rather than using the larger-looking number without its unit.
3. A full battery is an initial condition, not a permanent characteristic
The same installed system can enter two emergencies with different amounts of energy available. A battery that has just completed a profitable discharge may have less energy remaining than one held in reserve. Weather forecasts, price forecasts and operational schedules influence that starting point. A claim about emergency duration is incomplete until it specifies the state at which the event begins.
The System Advisor Model's battery-dispatch documentation makes minimum and maximum state-of-charge limits explicit. It also notes that these percentages refer to available capacity, which changes as batteries degrade or are replaced. The documentation is a model description, not a guarantee about any operating plant, but its distinction is useful: a percentage must be attached to an amount that can itself change. SAM battery-dispatch documentation.
Imagine that the hypothetical 400-megawatt-hour system begins with 60 percent of that usable inventory available. In the simplified example, it has 240 megawatt-hours rather than 400. At 100 megawatts it can supply 2.4 hours before exhausting that stated inventory. If an additional reserve must remain untouched, the deliverable amount is smaller still. A four-hour nameplate does not make the starting condition disappear.
The state-of-charge estimate also deserves technical attention. It is an estimate maintained by the battery-management system, not a transparent tank gauge showing liquid through glass. The public need not inspect proprietary estimation algorithms to ask whether the estimate is calibrated, how uncertainty is handled and whether actual delivered energy agrees with the expected inventory during controlled tests.
A reserve policy creates an economic trade-off. Holding more energy for an uncertain emergency can reduce opportunities to earn revenue or provide other services before the event. Spending that energy may be commercially rational under one contract but inconsistent with a separate backup promise. The conflict should be settled in the operating agreement rather than discovered during an outage.
The appropriate reserve can depend on the objective. A system intended to bridge a brief interruption has a different requirement from one intended to carry essential loads through a long storm. A forecast-sensitive policy might raise the reserve before anticipated stress, but that policy introduces dependence on forecast accuracy and the ability to recharge in time. A fixed reserve is simpler to explain but may sacrifice other value.
Evaluation should therefore examine a distribution of starting states, not only a demonstration that begins fully charged. A test with ideal initial conditions establishes a useful upper bound. It does not establish how often the same performance will be available under ordinary dispatch. The distinction is especially important when an emergency can begin without warning.
4. The meter boundary determines the usable promise
Energy stored in cells is not automatically equal to energy delivered to a customer. Conversion equipment, cables, transformers and auxiliary systems sit between the electrochemical inventory and the useful service. A project's internal reporting can be correct while differing from the net energy measured at its grid connection. The claim should identify which side of that boundary it describes.
Round-trip efficiency is commonly discussed as a ratio relating energy delivered to energy used for charging over a defined cycle. For a public purchaser, the important question is often the net alternating-current result at the interconnection, including the relevant system losses. The grid-storage FAQ explicitly distinguishes possible measurement boundaries and notes the importance of the AC-to-AC view for utilities. Grid-scale battery FAQ.
Suppose a hypothetical test draws 500 megawatt-hours from the grid and later returns 425 megawatt-hours at the same meter boundary. The measured round-trip energy ratio is 85 percent. That does not mean the battery can necessarily discharge at 85 percent of its rated power, nor does it tell us its duration without the corresponding power and inventory information. Efficiency and capacity are related operationally but are not interchangeable indicators.
The test also needs comparable starting and ending conditions. If the battery finishes with more stored energy than it began with, part of the charging energy remains in inventory. If it finishes with less, some delivered energy came from the initial inventory. Without correcting or controlling for that difference, the apparent round-trip result can be distorted.
Auxiliary consumption can become significant to the interpretation of long waiting periods. A test that immediately charges and discharges may answer a different question from a system held ready for days while control and thermal-management equipment operates. The reviewer should state the standby interval and which loads are included, rather than assuming that one efficiency number describes every use pattern.
Emergency planning needs the net service at the relevant load. If the promise concerns a particular facility, losses and limitations between the battery and that facility matter. If the promise concerns grid injection, the point-of-interconnection meter may be the appropriate boundary. The answer depends on the contract, not on which measurement produces the most favorable percentage.
A useful acceptance test therefore records the meter location, calibration, ambient conditions, operating limits, starting inventory and ending inventory. These details are not bureaucratic decoration. They establish what was actually demonstrated and allow a later test to be compared with the first without silently changing the question.
5. Dispatch spends a scarce inventory across time
A storage operator faces a sequence rather than a single moment. Discharging now can reduce the ability to discharge later. Charging now can increase demand on a stressed system while creating capacity for a later period. The best decision depends on the expected pattern of need, the available charging opportunity and the consequences of being wrong.
California ISO's market-monitoring notice for its 2025 battery report describes batteries as important suppliers during late-afternoon and early-evening high net loads and as significant loads during peak solar hours. That is an observed system role, not evidence that every battery follows an identical schedule or can cover every emergency. California ISO battery-report notice.
Consider a hypothetical evening with two shortage periods. The first requires 100 megawatts for two hours, and the second requires the same power for three hours. A fully charged 400-megawatt-hour battery cannot satisfy both in full without recharging because the combined requirement is 500 megawatt-hours. Dispatching aggressively in the first period may leave a shortfall in the second. Saving too much for the second may leave avoidable unmet demand in the first.
This is not a defect unique to one control algorithm. It is a constraint imposed by the inventory. Forecasting and optimization can improve allocation, but they cannot create energy. A reliability analysis should show how the dispatch rule handles uncertainty rather than assuming perfect knowledge of the entire future event.
Perfect-foresight simulations can still be useful. They establish what the equipment could achieve if the timing of demand and charging opportunities were known. Their results should be labelled accordingly. Comparing them with a realistic forecast-based strategy helps separate the value of additional hardware from the value of better information and control.
A reserve commitment further complicates dispatch. The system may have to retain headroom for an ancillary service while also scheduling energy delivery. Whether the same physical capacity can support both commitments depends on the service definitions and timing. Public descriptions of stacked benefits should not add up maximum outputs from incompatible operating states.
A good operational study makes the constraints visible in each interval: available energy, maximum charge and discharge rates, required reserves, grid limits and forecast inputs. The resulting schedule can then be tested against alternative event shapes. The question is not whether an optimizer can find an attractive schedule for one selected day, but whether the policy remains useful when the day develops differently from the forecast.
6. Recharging can become the limiting event
A battery that performs well during one evening can face a different challenge the next day. Replenishing the inventory requires an available energy source and a path through the electrical network. The nameplate discharge rating does not establish the charging rate, the permitted charging hours or the availability of generation during a prolonged disturbance.
In a simplified hypothetical case, a battery must replace 400 megawatt-hours of deliverable inventory. If the charging pathway supplies only 50 megawatts, even a lossless calculation requires eight hours. Real losses and operating restrictions can lengthen the needed window. If only four hours are available before the next event, the battery cannot be assumed to return to its initial condition.
A system paired with solar generation has an additional time pattern to consider. Solar availability may help refill storage during daylight, but the combined plant's interconnection and control arrangements determine what power can move where. A claim that the battery is paired with renewables does not by itself establish charging during every weather condition or outage configuration.
The recharge question becomes particularly important in multi-day stress studies. Resetting the battery to full at the beginning of every simulated day can artificially remove the very constraint the study should examine. Chronological modeling preserves the connection between one day's dispatch and the next day's starting inventory. The model should explain any reset or assumed external energy supply.
Charging can also compete with other loads. During a regional shortage, drawing electricity to prepare for a later shortage may worsen the immediate balance unless coordinated with the system's needs. An operating policy must decide when recharging is allowed and how it responds to emergency instructions. This is a system interaction, not simply a feature of battery chemistry.
The evaluator should ask what happens after successful performance. How quickly does the system recover its readiness? What source provides the energy? What conditions would prevent recovery? A battery with excellent first-event performance and slow replenishment can still be valuable, but its service must be described honestly for sequences of events.
This changes the meaning of resilience. Resilience involves maintaining or restoring useful service through a disturbance, not merely surviving the first hour. For storage, the replenishment plan is part of that capability. The planning record should follow energy through the entire disturbance and recovery period instead of ending the analysis at the moment the first discharge succeeds.
7. Degradation changes the denominator over the project life
The energy available from a battery system can change with age and use. A design rating at commissioning is therefore not a complete statement of performance several years later. Project economics often anticipate maintenance, replacement or additional modules, but those future actions should be distinguished from capacity already installed and tested.
A NYSERDA energy-storage performance evaluation discusses degradation in relation to characteristics including chemistry, operating temperature, cycling and state-of-charge range. Its analysis used a model to estimate future degradation for studied systems. Those estimates should not be mistaken for direct observations of every future year. The report is useful because it connects operating choices to the condition being forecast. NYSERDA performance evaluation.
Suppose the hypothetical system's tested usable energy falls from 400 to 360 megawatt-hours while its permitted discharge power remains 100 megawatts. Its simplified full-power duration becomes 3.6 hours. Calling it four-hour storage may still describe its original design class, but a current performance promise needs to explain whether augmentation or another measure restores the required delivery.
Capacity fade is not the only relevant change. Power capability and efficiency can also depend on condition and operating circumstances. An acceptance regime should therefore test the service that matters rather than relying on a single remaining-capacity percentage. A project contracted to deliver a particular power profile should demonstrate that profile within the agreed limits.
A warranty can allocate financial responsibility, but it is not a substitute for measured performance. The warranty may define permissible operating conditions, throughput limits, exclusions and remedies. A remedy that arrives months later may matter economically while doing nothing to supply electricity during a particular emergency. Reliability planning should separate the existence of a warranty from the availability of replacement capability when needed.
Augmentation plans deserve the same distinction between commitment and accomplishment. A financial model can assume additional modules in a future year. The operational record should later show whether they were procured, installed, commissioned and integrated into the control system. Until then, the projected capacity belongs in a planning scenario, not an observed inventory.
A practical reporting approach publishes both current tested capability and the expected future trajectory with its assumptions. That allows decision-makers to see whether the project remains on track and whether an intervention is needed. It also prevents a successful commissioning test from becoming a permanent proxy for a system whose physical condition continues to evolve.
8. Temperature, equipment and controls shape available power
A battery project is an electrical and thermal system, not a collection of cells alone. Its performance depends on conversion equipment, protection, communications and support systems as well as the storage medium. A failure in one part can reduce available output even when substantial energy remains inside the cells.
The useful distinction is between stored energy and deliverable energy under the current operating constraints. If an inverter is unavailable, the remaining cells cannot necessarily route all their energy through the surviving equipment at the original rate. If a protection setting limits operation, the controller should obey it. Treating those limits as irrelevant because the nameplate remains unchanged would misdescribe the available service.
A hypothetical plant with ten equal 10-megawatt conversion blocks illustrates the point. If two blocks are unavailable and there is no cross-routing that restores their output, the plant's accessible power may fall to 80 megawatts. The energy consequence depends on the electrical arrangement and which battery sections remain connected. The evaluator should not infer it from the number of failed blocks alone.
Temperature can affect both immediate operation and long-term condition. The NYSERDA evaluation includes operating temperature among the characteristics relevant to degradation. For a specific project, however, the appropriate performance limits must come from its tested equipment and operating documentation rather than a generic assumption about all batteries. NYSERDA evaluation.
Emergency conditions can create correlated challenges. The same heat event that increases demand may stress thermal-management equipment. A storm can affect communications and the external network simultaneously. These possibilities do not establish that a particular plant will fail, but they show why an analysis based on independent, average equipment availability may miss important event-specific behavior.
Testing should therefore include the relevant operating envelope, with safe procedures and qualified personnel. A public report can state the ambient range and constraints under which a capability was demonstrated without publishing sensitive control details. It should distinguish a laboratory component test from a full-system field test at the interconnection.
The operational record also needs reasons for curtailed output. A reduction caused by a grid instruction is different from one caused by depleted inventory or an equipment fault. All may appear as low generation in a simple chart. Diagnosing them correctly is necessary before drawing conclusions about technology reliability or deciding which investment would improve performance.
9. Capacity accreditation asks a system-level question
A battery's contribution to resource adequacy depends on when the system needs support and what other resources are available. Effective load carrying capability, or ELCC, is one approach to evaluating that contribution. It is not a physical replacement for the nameplate rating and does not describe a guaranteed output in every possible emergency.
PJM's April 20, 2026, fact sheet explains that its ELCC approach considers resource contributions during high-risk periods and interactions among resources. It describes a marginal approach and notes that risk hours can change as the resource mix changes. These are features of PJM's described framework, not a universal rule for every electricity market. PJM ELCC fact sheet.
The intuitive issue is straightforward. If a system's most difficult hours occur in a short evening window, a limited-duration battery may cover a substantial share of the need. If the difficult period becomes much longer, the same battery may exhaust its energy before the event ends. Its physical hardware can remain unchanged while its incremental reliability contribution changes.
The contribution can also depend on how many similar batteries already exist. Additional storage may initially cover the sharpest shortage periods. As those periods are addressed, the remaining risk can shift toward longer events that are harder for the same duration to cover. This is an analytical explanation of possible saturation, not a claim that a particular market has reached a specific threshold.
Accreditation therefore belongs to a modeled system and a stated period. A number published for one delivery year should not automatically be carried into a later year with different demand, generation and transmission assumptions. Nor should a class-level value be presented as proof that every individual asset will perform identically.
A public procurement team should ask how its reliability objective relates to the accreditation framework. Buying market-accredited capacity may serve a regional adequacy requirement. It may not provide islanded power to a particular building. Conversely, a local backup system can be valuable to its customer even if it does not participate in the regional capacity market.
The most useful disclosure shows the physical rating, the current applicable accredited quantity and the service being purchased as separate fields. That prevents a planning metric from being mistaken for an engineering guarantee and prevents the engineering nameplate from being mistaken for a system-level reliability contribution. Both numbers can be valid, but they answer different questions.
10. Grid support and local backup need different equipment
A grid-connected battery can inject power while the grid is functioning without being able to energize an isolated local network. Local backup requires an electrical design that safely separates the intended island, controls voltage and frequency as needed, coordinates protection and supplies the selected loads. The presence of a battery nearby does not establish that arrangement.
The System Advisor Model's grid-outage documentation separately models critical loads and outage periods for behind-the-meter storage. It distinguishes checking specified outages from estimating autonomy for hypothetical outages beginning at different times. That separation is a useful planning concept even though the software's assumptions are not a certification of any real installation. SAM grid-outage documentation.
For a hypothetical community center, the essential load might include communications, refrigeration and a limited conditioned space rather than the entire normal building load. Defining those circuits changes the energy requirement. A statement that the battery can power the building is ambiguous unless it says whether all loads or only a defined critical subset are included.
Power peaks also matter. A low average load can conceal short periods when equipment starts or several devices operate together. The system must meet the relevant instantaneous demand or manage it through an approved load-control strategy. Calculating hours from average consumption alone can produce an attractive autonomy estimate while overlooking the power constraint that interrupts service first.
The transition into islanded operation is another separate performance question. A facility may tolerate a brief interruption or may require a different continuity standard. The battery's energy capacity does not establish transfer time. The load's needs, switching equipment and control design determine whether the transition is acceptable.
Reconnection must also be planned. Restoring normal grid service is not simply a matter of continuing the same isolated operation. Qualified engineers and operators need appropriate coordination and protection. Public communication should avoid implying that residents can improvise a connection or that a large battery automatically creates a safe microgrid.
A meaningful backup claim therefore describes the supported loads, operating mode, transition behavior, autonomy assumptions and tested configuration. It can then be checked against the actual facility. This is a different evidence package from a grid-market performance report, and a project promising both services must demonstrate both rather than borrowing confidence from one to cover the other.
11. Test an event sequence, not a promotional snapshot
A photograph of a battery enclosure proves very little about operational capability. A single chart showing a successful discharge proves more, but its meaning still depends on the test conditions. The strongest evaluation connects the requested service, the initial state, the control instructions and the measured response over the relevant period.
For a hypothetical acceptance program, the first test might demonstrate maximum net discharge power under specified conditions. A second might demonstrate usable energy at an agreed profile. A third might examine recovery through charging. A fourth might evaluate a sequence resembling the intended reliability service. These are proposed test categories, not a universal standard or a direction to operate equipment outside approved limits.
A sequence can reveal constraints hidden by isolated tests. The battery might meet a power test and an energy test separately but fail a combined dispatch that follows earlier service. It might successfully support one outage but begin the next with insufficient inventory. Testing the chronology exposes the interaction between otherwise satisfactory individual measurements.
The test record should preserve unsuccessful attempts as well as successful ones. An interruption caused by a communications problem may be repairable, but omitting it from the record makes the final result appear more repeatable than the evidence supports. A retest can demonstrate a correction while retaining a clear account of what changed.
Measurement resolution should fit the service. A monthly energy total cannot establish a fast response. A second-by-second trace may establish response timing but still fail to show sustained availability across several hours. The reporting interval should be chosen because it resolves the question, not because the available dashboard happens to use it.
Data quality also matters. Missing intervals, clock misalignment and different sign conventions can complicate comparisons between dispatch instructions and metered output. A reviewer should reconcile these issues before attributing a deviation to the battery. Otherwise an apparent performance failure may be a data-processing error, or a real failure may be concealed by an averaging choice.
The final report should state which claims the tests support and which remain untested. A successful controlled demonstration is valuable evidence. It should not be inflated into a statement that every future emergency will be covered. The purpose of testing is to establish a dependable operating envelope and identify the conditions beyond it.
12. Read performance with the counterfactual kept modest
After an emergency, observers often ask how many outages a battery prevented. That question requires more than its recorded output. The system might have used another resource, reduced exports, changed demand or experienced a different network constraint without the battery. Estimating the avoided consequence requires a counterfactual model and explicit assumptions.
Measured delivery is the firmer starting point. How much power did the system supply, when, for how long and at what boundary? Did it follow the instruction it received? What was its initial inventory, and what remained? Those facts can establish performance without claiming an exact number of customers who would otherwise have lost service.
A hypothetical battery may deliver 200 megawatt-hours during a difficult evening. Dividing that figure by a typical household's daily consumption does not establish the number of homes whose outage was prevented. The actual loads, timing and network conditions differ. Such a comparison may illustrate scale if clearly labelled, but it should not masquerade as a reliability outcome.
Likewise, a low-output period does not automatically establish failure. The battery might have been held in reserve, constrained by the network or not dispatched because another resource was cheaper or more suitable. The operational record should distinguish availability from utilization. A system can be available and idle, or generating while still unable to meet its full commitment.
A fair evaluation compares actual behavior with the obligation and conditions in force. If the contract required readiness, check readiness. If it required a dispatch response, compare the instruction with delivery. If it promised local backup, examine the supported loads and outage configuration. This avoids judging a system against a service it was never designed or contracted to provide.
The wider system benefit can then be modeled with appropriate caution. Analysts should identify alternative dispatch assumptions and test sensitivity to uncertain inputs. A range of plausible avoided outcomes may be more honest than a precise single figure. The uncertainty belongs in the conclusion, not only in a technical appendix nobody reads.
This approach allows genuine success to stand on solid evidence. A battery that delivered the requested service during a stressful event has demonstrated something important. There is no need to attach an unsupported dramatic counterfactual to make that achievement meaningful.
13. A specification that survives the emergency
A useful storage specification begins with a service profile rather than a slogan. It defines the power required across time, the output boundary, the expected event sequence and the allowed starting conditions. It also states which uncertainties the project must tolerate and which assumptions remain outside its responsibility.
For a hypothetical public purchaser, the specification might require a defined net output for a stated duration at a particular stage of project life, with an explicit reserve policy and a recharge plan. The exact numbers would follow an engineering study. The important editorial point is that the requirement should connect directly to a measurable test and an operating obligation.
The agreement should allocate control authority. Who can discharge the battery for ordinary market activity? Who changes the reserve before forecast stress? Who receives emergency instructions? If several organizations share the system's value, their priorities need to be reconciled. A promise of backup can become fragile when another party has authority to spend the same energy without considering that promise.
Performance guarantees should also address aging and maintenance. The purchaser needs to know whether a stated capability is maintained through the contract or only demonstrated at commissioning. If augmentation is required, the responsibility, timing and verification should be clear. If a component is unavailable, the reporting should identify the resulting service limitation promptly.
The IEA's discussion of battery deployment describes projects increasingly combining several services and revenue streams. That trend makes clear service definitions more important, not less. A flexible asset can create several forms of value, but the physical inventory and equipment limits still have to support the commitments made at the same time. IEA battery-system commentary.
Public reporting need not expose every commercial detail. It can state tested net power, tested usable energy, current readiness policy, relevant limitations and the status of corrective actions. These fields tell residents more about emergency capability than a repeated nameplate figure accompanied by a photograph of installed equipment.
The specification should also define how a changed operating configuration is treated. A software update, a replacement inverter or a revised protection setting may leave the advertised capacity unchanged while altering the behavior that was originally tested. The purchaser should know which changes require review or repeat testing and who decides that the evidence remains applicable. This does not imply that every minor change demands a complete recommissioning exercise. It means that the link between the current configuration and the demonstrated service should remain traceable. Otherwise a test report can become an artifact of a configuration that no longer exists, even while everyone continues to cite it.
The ultimate standard is straightforward: describe the service at the level where someone relies on it. A regional grid planner, a local facility manager and a household facing an outage may need different answers. The battery's value becomes clearer when those answers are separated. Its promise becomes credible when the energy inventory, dispatch plan, equipment condition and electrical pathway all support the same claim.