1Purpose and scope

The BESS Backtest measures how much of a battery's full potential its optimizer converts into realised results. It compares the realised results of the battery, as settled by the optimizer and checked against meter data, with the full potential of the same battery on the same site: the highest result that was achievable within the limits of the site and the rules of each market. The outcome is expressed as a capture rate, in percent, and as a delta with expected results, in euros. This methodology sets out how these figures are calculated and presented.

The BESS Backtest covers battery energy storage systems (BESS) in the Netherlands that operate under a merchant contract, either stand-alone or behind a grid connection shared with solar panels (PV) and on-site consumption. A project is one site with one connection agreement and one or more battery assets, each with its own meter. The markets in scope are the day-ahead and intraday markets, the imbalance market, and the balancing markets for frequency containment reserve (FCR) and automatic and manual frequency restoration reserve (aFRR, mFRR). In this version, intraday and mFRR are not yet modelled; realised intraday revenue is compared with imbalance and aFRR energy together (sections 4 and 7). Section 16 lists the features planned for later versions (backlog); section 18 defines all terms and abbreviations.

Results are published per project in a dashboard and a report. Both contain the same sections: key figures, the BESS Backtest chart of realised results against expected results and full potential, the delta with expected results per period, revenue by driver, project details and downtime; each chart has its figures in a table underneath.

This methodology is prepared by the Independent Capacity Market B.V. (the ICM) and forms part of the Terms & Conditions; in case of conflict, the Terms & Conditions prevail. Capitalised terms have the meaning given in the Terms & Conditions. Until the BESS Backtest has its own Terms & Conditions, those of the BESS Benchmark apply.

2Principles

• Site-specific. The full potential is modelled with the battery, PV, consumption, grid limits and dispatch constraints of the site itself, not with a standard battery.
• Independence. The ICM owns no battery assets and is not an optimizer. The same rules apply to every client, regardless of its optimizer.
• Perfect foresight within market rules. The full potential uses realised prices in hindsight: the results of the day-ahead and capacity auctions. Balancing prices, which are known only after delivery, are used only within backtest limits that restrict how much the model can gain from knowing them (section 7). Expected results, set at 80% of the full potential, account for what an optimizer cannot know in advance.
• Triangulation. Every figure is checked against at least two independent sources before a month is marked final (section 5).
• Additive figures. Drivers add up to the total, and periods add up to the period total. Percentages for longer periods are calculated from summed euros, never averaged.
• Like for like. Realised results and full potential are valued with the same formulas (section 6). Revenue from a market that the full potential does not model at all is reported, but left out of the capture rate.
• Hardware separate from trading. Downtime is reported as a separate driver, so that a fault is not attributed to trading performance, and the reverse.
• Reproducible. The full potential is solved with an open-source solver, and inputs, code version and solver version are stored with every run, so that any result can be recalculated (section 7).

3Comparison with other approaches

Battery performance is measured in four established ways. Each answers a different question; the BESS Backtest answers how much of the potential of one specific site was captured.

How benchmark providers model a battery

Benchmark providers fall into three groups: simulations of a virtual battery (Modo Energy, KYOS, RWTH, Enervis, Clean Horizon and Aurora), realised revenue of real portfolios (enspired, and Modo Energy in Great Britain), and optimizers that simulate a client's asset on request (E6). Simulations use perfect foresight for markets that clear in auctions, limit foresight on continuous intraday and imbalance, and cap cycles; several then scale modelled revenue by a factor of 0.75 to 0.90. None applies perfect foresight to imbalance prices without limits: Modo Energy leaves imbalance out and KYOS uses a forecast. Of the simulations reviewed, only KYOS covers the Netherlands today. The full potential of the BESS Backtest uses perfect foresight on auction prices and limits the use of balancing prices by backtest limits (section 7).

Research on the continuous intraday market shows why an index alone is not enough to value intraday trading. A forecast-driven trading strategy on German 2023 data earned 11% less than perfect foresight on individual trades but 4% more than perfect foresight on ID1, the volume-weighted average price of trades in the last hour before delivery (Hornek et al.). Indices do not reflect bid–offer spreads, and the shorter the window before delivery, the more volatile the index and the higher the modelled revenue (Oeltz). Traders run a rolling intrinsic policy on the order book, re-optimising as prices move; its value depends on how often it re-optimises (Schaurecker et al.). The BESS Backtest therefore adds intraday once continuous intraday prices per quarter-hour, with their volatility, are collected (section 4).

An industry grading puts perfect foresight over a whole month or year at grade C, a look-ahead of 24 to 48 hours at grade B, and grade B plus asset limits (downtime, state-of-charge bounds, degradation) and network limits (outages, curtailment, grid constraints) at grade A (Energy-Storage.news). The full potential of the BESS Backtest is designed to meet grade A with a two-day rolling horizon, in which each day is planned together with the next; outages of the grid are not modelled, and congestion limits apply when the client declares them.

Backtests run well ahead of what an asset earns over time: a German whitepaper puts historical backtests at EUR 200,000–280,000 per MW per year against a structural level of EUR 120,000–180,000 (Re-Twin Energy and FfE). The BESS Backtest compares like with like instead: the same period, the same site limits and the same valuation for the model and the realised results.

4Collected data

The BESS Backtest needs three kinds of input. The client delivers one intake workbook per project, with the site and its grid connection, each battery's rated power and storage capacity, and the markets in the optimizer's mandate. The full potential is calculated from quarter-hour meter data, the project info and public market prices. Realised results come from the optimizer report in its most granular form, together with the meter data for grid costs and degradation costs. Data from the energy management system (EMS) establish downtime and serve as a check on the meters. Every other item is a cross-check.

The checklist below states, per delivering party, which items are required and which serve as a cross-check. Meter data are delivered by the metering service provider (MSP), the certified company responsible for the meters of the connection. The checklist assumes a site with multiple suppliers on one connection (MLOEA), on which the battery, PV and consumption each have their own allocation point. Meter and EMS data are needed per quarter-hour for at least six months before the first report, and then each month.

Market data. Day-ahead, imbalance, balancing energy and FCR prices and aFRR capacity prices are published free of charge by the ENTSO-E Transparency Platform, TenneT and regelleistung.net; bulk downloads from the ENTSO-E Transparency Platform require a free account. Continuous intraday prices are not free: no free quarter-hour price series exists for the Dutch continuous intraday market. EPEX Spot sells quarter-hour price indices and trade data for the Netherlands, for the continuous market and for the intraday auctions that started on 13 June 2024, as data packages; its general conditions govern their use (EPEX Spot market data (EEX Group webshop), EPEX Spot general conditions for market data (July 2024), EPEX Spot – go-live of intraday auctions (13 June 2024), ENTSO-E – intraday auctions on SIDC). An index alone, such as ID1, fills every trade at one price without bid–offer spread or liquidity limit (section 3). This version therefore leaves the intraday stage out of the full potential until continuous intraday prices per quarter-hour, with their volatility, are collected from these licensed data: the real-time stage deviates from the day-ahead position, and its result stands for intraday, imbalance and aFRR energy together. Realised intraday revenue stays in the capture rate and is compared with imbalance and aFRR energy as one. Realised mFRR revenue, for which the full potential has no counterpart, is reported but left out of the capture rate and the delta with expected results while mFRR is not modelled (section 7).

Public market data are collected in one workbook per period, with every download kept unchanged next to the series the model reads; a check on every row tests that the value used equals the value downloaded, and the source files are kept with SHA-256 checksums. Each period passes automatic checks before use, among them no missing quarter-hours or prices and a unit check of the aFRR capacity prices against the FCR auction. Activated mFRR volumes are collected but not yet used.

A review of the public websites of 29 optimizers active in the Dutch market (section 17) found that none commits publicly to revenue per market per quarter-hour. Edmij provides metered data, traded volumes and prices per quarter-hour so that customers can verify their invoice, and E6 Group states that settlement data are available per asset; most others describe a dashboard or a monthly report. The ICM therefore asks for the most granular report available. Where only a monthly report exists, realised results have monthly granularity (section 12).

On a site with a single supplier, the connection has only its primary allocation point: charge and discharge are taken from the battery submeter, read by the MSP where possible and otherwise through the EMS, in which case they are marked as EMS data. Supply contracts are not needed: the allocation points are identified by their EAN codes, which the client provides and the MSP confirms with the meter data.

Multiple suppliers on one connection (MLOEA). Under MLOEA, the battery has its own allocation point, with its own EAN, meter, supplier and balance responsible party; the rules are set out in articles 2.3 and 2.5 to 2.10 of the Systeemcode elektriciteit 2026, which since 1 January 2026 replace the code decision of 2017 without change of policy (ACM). For the BESS Backtest, MLOEA matters for the data, not for the model. The battery meter is a certified meter read by the MSP, and imbalance is settled per allocation point, so the optimizer report covers the battery alone and can be checked directly against its meter. The full potential does not change: the GTV and the transport tariffs remain per connection, so the battery shares the grid limits with PV and consumption as on any other site (Westland Infra, Liander). Whether energy tax then applies per allocation point is an open question (section 15).

Neither the battery datasheet nor the full optimizer contract is required; only the clauses listed above. A subsidy decision for the PV (SDE++) is not required either (section 7). All amounts are in euros, excluding VAT. Storage and use of client data are governed by the Terms & Conditions.

5Data validation

Every figure is checked against at least two independent sources. At intake, the project info is confirmed with the client against the documents in section 4. Each month, the figures below are triangulated before the month is marked final (section 13):

Any difference above 2% of the monthly total is explained before the month is marked final. Where meter and EMS data differ, meter data prevails: the meters are certified and are the basis for allocation, imbalance settlement and the energy invoices, whereas the EMS measures at the inverter and is not calibrated. Energy invoices contain monthly totals only and therefore serve as a check, not as a source. All figures are calculated by software and reviewed by an analyst before a month is marked final. Client reports are prepared for real assets only. The model itself will be validated by an independent party; the choice of that party is an open question (section 15).

How each figure is obtained

Every figure in the dashboard and report is one of five types. Measured figures come directly from certified meters or the EMS; reported figures come from the optimizer; calculated figures follow from measured or reported data by a fixed formula; modelled figures result from the optimisation in section 7; estimated figures rest on an assumption. Two inputs are estimated: the degradation cost per MWh and the 80% factor for expected results.

Checks on the model

Every run passes a fixed set of automatic checks before its results move on, each with a stated tolerance: the revenue of every driver is recomputed independently from the model's volumes and the public prices; stored energy is never negative; import and export stay within the GTV and delivered aFRR energy within TenneT's activations; discharge stays within the daily cycle limit; drivers add up to the total; and the run with actual downtime never exceeds the run at full availability.

Next to the full potential, the model solves variants of every day that form a value ladder. Each rung adds one market or one freedom to the rung below, so its value can only be equal or higher; a day on which a rung falls below the one beneath points to an error in the model.

A stochastic benchmark shows what was achievable without knowing balancing prices. It takes auction decisions with known auction prices, decides aFRR bids, schedule and passive positions on 14 scenarios drawn from the days before D-1, and settles that plan on the actual day. Its capture rate is calculated as an internal check on how much of the full potential was achievable without hindsight, and is not shown in the dashboard. Before a month is released, an analyst confirms that all checks pass, that every capture rate above 100% and every difference above 2% between the recalculation and the optimizer report is explained, that downtime matches the EMS log, and that site limits match the intake documents (section 13).

6Realised results

Realised results are the earnings of the battery in euros, per revenue driver. Market revenues are taken from the optimizer report, because it reflects actual settlement, including capacity payments that meters do not record. Grid costs, energy tax and degradation costs are calculated by the ICM from meter data. Where the optimizer reports per quarter-hour, every driver is recalculated with public prices and the traded volumes are compared with the battery meter, using the formulas below. The same formulas value the positions of the full potential (section 7).

Sums run over the quarter-hours, or the blocks, of the period. Because the day-ahead driver values all energy at the day-ahead price, the intraday, imbalance, aFRR and mFRR drivers capture only the additional value over the day-ahead price; together, the drivers add up to the settled revenue. Where the optimizer reports only monthly revenue per market, its split is cash per market rather than this decomposition; the total is then compared with the full potential, and the comparison per driver is indicative. Realised results are reported gross, before the optimizer's fee and fixed fees of the balance responsible party, because the full potential contains no fees either; fees are a contractual matter, not a measure of trading performance.

Grid costs are charged on the connection, so the battery's share is the difference it makes: the extra monthly import peak and the extra grid import, against the same site without battery. Extra grid import equals energy charged from the grid minus energy discharged to on-site consumption. Energy tax is reported with grid costs, because both arise from the same extra grid import. The fixed charge and the contracted-capacity charge of the transport tariff do not depend on how the battery is dispatched; they are costs of the connection and outside the backtest. On a site with several battery assets, grid costs and energy tax are allocated to the assets in proportion to their extra grid import.

The degradation cost of EUR 10 per MWh discharged follows from a replacement cost of EUR 70,000 per MWh of capacity and a warranted life of 7,000 full cycles. A full cycle discharges the full capacity once and uses 1/7,000 of the battery's life, so each MWh discharged costs EUR 70,000 ÷ 7,000 = EUR 10. For reference, BloombergNEF reports an average 2025 pack price of USD 70 per kWh (USD 70,000 per MWh) for stationary storage.

A difference of more than 2% between the recalculation and the optimizer report is explained with the optimizer before the month is marked final.

7Full potential

The full potential is the highest result that the battery assets on a site could have achieved with realised market prices, within the limits of the site and the rules of each market. Auction results are used with perfect foresight; balancing prices, which are known only after delivery, are used only within the backtest limits set out below. This follows the practice of benchmark providers (section 3) and keeps the upper bound unambiguous and reproducible. The gap between this bound and what is achievable in practice is reflected in expected results (section 8).

Time structure. The model works per quarter-hour, the imbalance settlement period: 96 quarter-hours per local calendar day, or 92 and 100 on the days the clocks change. Day-ahead prices are hourly until 30 September 2025 and apply to the four quarter-hours of each hour; the day-ahead market has been cleared per quarter-hour since delivery day 1 October 2025 (NEMO Committee). All battery assets behind one connection are optimised together as one site, each within its own limits, because they share the GTV and the monthly import peak.

Rolling horizon. Traders and balance responsible parties plan with forecasts days ahead and choose the stored energy at midnight that sets up the next day. The full potential therefore optimises each day together with the next, on a two-day rolling horizon:

1. Start. Day D starts with the stored energy in which day D−1 ended. The first day of the period starts at the stored energy recorded in the EMS log, or at the midpoint of the usable capacity where the log has none.
2. Look ahead. Days D and D+1 are optimised together, in the stages described below; D+1 ends at the midpoint. The stored energy that this solve holds at midnight at the end of day D becomes the target for day D.
3. Settle the day. Day D is optimised again on its own, from its start to that target, and its result is kept. The second day of the two-day solve is discarded and solved afresh when its turn comes.
4. Value the change. The change in stored energy over the day is reported as a driver of its own, stored energy carried, so that a day that charges for the next is not shown with the cost of charging but without the value it passes on.

The last day with market data ends at the midpoint. Where the two-day solve has no feasible solution, the day ends at the midpoint instead; in a test run on four sites over July–September 2026, this applied to 4 of 368 site-days. A fixed stored energy at midnight gives away value, mainly for batteries of two hours or more: Modo Energy fixes 50% at midnight for its German index and optimises three days for batteries above 6 hours (Modo Energy – Germany index, Modo Energy – core dispatch model).

In words: energy passed on to the next day is valued at the average day-ahead price of the day, net of the losses of discharging it later. The next day books the revenue from that energy and a negative carry. Over a month the amounts largely cancel: in the test run of July–September 2026 they netted to −0.9% of the full potential, because energy is valued at the price of the day on which it is carried, not the day on which it is sold.

Order of the markets. Each day is solved in three stages that follow the gate closures of the markets. A later stage cannot change what an earlier stage committed.

1. Day ahead (D-1). FCR capacity, aFRR capacity and the day-ahead position are optimised together with the auction results of the day, and every capacity bid in the plan is accepted at the published price. FCR closes at 08:00, aFRR capacity at 09:00 and the day-ahead market at 12:00 on the day before delivery (ENTSO-E – FCR, TenneT aFRR manual v2.8, EPEX Spot). Optimising them together is an upper bound: in practice, capacity is offered before day-ahead prices are known, bids can be rejected, and pay-as-bid auctions reward bid strategy.
2. Intraday. Not modelled in this version. Intraday is added once continuous intraday prices per quarter-hour, with their volatility, are collected from licensed EPEX Spot data (section 4). An index such as ID1 fills every trade at one price, without bid–offer spread or liquidity limit, and is therefore not used on its own. Once added, this stage re-plans the remaining power and energy against the intraday price, with the positions of stage 1 fixed, and the intraday price replaces the day-ahead price as the reference for the backtest limits.
3. Real time. Per quarter-hour, the battery settles against TenneT's balancing prices: passive imbalance, aFRR energy offered without a capacity contract, and the activation of contracted aFRR, within the backtest limits below. The deviations of the day are optimised together, so that the state-of-charge window, the energy held for FCR and aFRR and the target at midnight still hold.

Backtest limits. A backtest knows every price after the fact, including the imbalance price that TenneT sets only after each quarter-hour. The backtest limits restrict how far the model may use that knowledge, so that the full potential reflects what a skilled trader could have earned, not what perfect knowledge of the imbalance price would yield. They apply to the two revenue streams whose price is known only after delivery: passive imbalance, in which the battery deliberately deviates from its traded position and TenneT settles the difference at the imbalance price, and aFRR energy offered without a capacity contract. TenneT allows passive balancing: a balance responsible party may take an imbalance opposite to the system imbalance and receives the imbalance price for it (TenneT – balancing markets). The limits are the same for every project; an optimizer contract can make them stricter, never looser.

Within the quarter-hour, TenneT's balance delta and regulation state are updated every 12 seconds and published with a delay of 2 minutes (TenneT – balance delta, TenneT – 12-second balance delta); traders act on this signal, but the final imbalance price is settled afterwards. The model does not use the real-time signal; the limits stand in for it. No benchmark provider reviewed uses perfect foresight on imbalance: Modo Energy leaves imbalance out, and KYOS uses a forecast (section 3).

Balancing prices. Long positions earn the surplus price and short positions pay the shortage price, as published by TenneT. In regulation state 0 both prices are equal; in state +1 both equal the upward balancing energy price; in state −1 both equal the downward balancing energy price; in state 2 the shortage price is at least the upward and the surplus price at most the downward balancing energy price (TenneT – Imbalance Pricing System v6.3). On April–September 2026 data these rules hold in 99.7% of quarter-hours. Dutch imbalance and balancing energy prices are set by the Dutch merit order, not by the cross-border marginal price of the European platform PICASSO (TenneT – market update). aFRR energy offered without a capacity contract is paid the Dutch balancing energy price for its direction, and the battery never delivers more than TenneT activated in that quarter-hour and direction.

Price impact. The battery is a price taker in the auctions. On aFRR energy and imbalance, its own volume would have replaced the most expensive bid in TenneT's merit order and moved the price against it. Over April–September 2026, each extra MW activated moved the balancing energy price by EUR 1.06 per MWh upward and EUR 0.50 per MWh downward, estimated as the slope of the balancing energy price on the activated volume; the fit is weak (R² 0.18 and 0.10), and the full potential moves by 0.3% or less between zero and twice these values.

In words: the price moves by α per MW on the whole volume, so the cost rises with the square of the volume. A battery that delivers 2 MW of upward aFRR energy at a published price of EUR 200 per MWh would have moved the price to EUR 197.88 and earns EUR 98.94 instead of EUR 100.00: a cost of 1.06 × 2² × 0.25 = EUR 1.06.

Capacity markets. Capacity is included only for markets that are in the optimizer's mandate and for which the asset is prequalified, up to the prequalified MW. Capacity is balancing capacity, in MW, held in reserve for a block; what the battery delivers when it is activated is balancing energy, in MWh.

• FCR is a symmetric product, auctioned daily per 4-hour block at one clearing price (ENTSO-E – FCR). For batteries, TenneT requires full delivery for at least 15 minutes and a ratio of rated power to prequalified power of at least 1.25:1 (TenneT FCR manual v3.3). The model therefore reserves at most rated power ÷ 1.25, keeps the stored energy at least 15 minutes of full FCR delivery away from both limits while FCR is held, and uses only the remaining power for energy trading. FCR revenue is the reserved MW times the Dutch settlement price of the block. The minimum activation period for batteries is under review by ACER, with a decision expected by early 2027 (ACER).
• aFRR capacity is contracted by TenneT on D-1, upward and downward separately and paid as bid, in two products: 4-hour blocks and a daily product with the same MW all day (TenneT aFRR manual v2.8, TenneT – 4-hour contracts). The model may use both. Because the auction is pay-as-bid and has no single clearing price, each block is valued at the average price of accepted bids per block, product and direction, as published on the ENTSO-E Transparency Platform in EUR per MW per quarter-hour (× 16 gives the price of a 4-hour block). These prices are capped at the 95th percentile of the month, per product and direction, which limits the windfall from perfect foresight on rare spikes: 4-hour blocks reached EUR 4,902 per MW per hour in May 2026, and the monthly cap on upward 4-hour aFRR ranged from EUR 23 (April) to EUR 770 (May) per MW per hour in April–September 2026. A monthly cap is final when the month closes and gives every month the same haircut on foresight; Modo Energy caps at a fixed EUR 50 per MW per hour, its 95th percentile (Modo Energy – Germany index).
• aFRR energy reservoir. TenneT requires "an energy reservoir for the contracted period for the contracted volume" (TenneT aFRR manual v2.8), so the battery holds enough stored energy to deliver the reserved MW for the whole block: for upward aFRR, energy above the minimum; for downward aFRR, room below the maximum. For a battery with two hours of storage and 4-hour blocks, this limits aFRR to about half its rated power in one direction, or about a quarter in each direction when both are held. A MW is not counted in FCR and aFRR at the same time, and the energy held for each adds up. Whether energy already delivered in the block may count is an open question (section 15).
• Activation of contracted aFRR. Contracted aFRR is activated pro rata: in each quarter-hour, the share of TenneT's contracted volume that TenneT activated, valued at the Dutch balancing energy price for its direction. Activation the battery cannot deliver, for example at its cycle limit, is settled upward at the higher of the shortage and the upward balancing energy price, and downward at the lower of the surplus and the downward balancing energy price.
• Reserves within the GTV. Reserved FCR and aFRR must be deliverable at the connection: full downward activation within the import GTV, full upward activation within the export GTV.
• mFRR is not modelled in this version. Since 1 September 2026, TenneT procures it as Incident Reserve, replacing the directly activated mFRR product (TenneT – Q3 2026 update); participation of batteries is not yet confirmed (section 15).

The full potential is subject to the same limits and costs as the battery on site. These limits are confirmed at intake and shown in the dashboard under Project details, together with a diagram of the site. Where the intake gives no value, the default below applies.

Energy tax is levied per kWh supplied through the connection, at a rate that falls with the annual offtake. In 2026, the rate is EUR 37.35 per MWh for an annual offtake of 50,000 to 10 million kWh and EUR 3.10 per MWh above 10 million kWh for business use (Vattenfall), the same order of magnitude as the day-ahead spreads on which a battery trades. The model uses one rate per connection: the marginal rate of the bracket in which the connection's annual offtake falls, from the connection meter and annualised where less than a year of data is available; the energy invoice serves as a check. Brackets below 50,000 kWh are not modelled, as battery sites rarely fall below. Energy exported later earns no refund. Energy tax is zero where the supplier applies the exemption for supply to an energy storage facility, which requires storage on its own, separately metered connection (Ministry of Finance, Vattenfall – declarations).

Pooling and site-wide optimisation. Where the optimizer trades the site as a whole, its revenue includes the imbalance of PV and consumption; that part is split off from realised results, or the site is flagged, because PV forecast errors move with the Dutch system state and would otherwise count against the battery. A battery pooled with batteries behind other connections is benchmarked as part of the pool: the revenue of the pool is compared with the sum of the full potentials of its sites. This keeps the allocation key out of the capture rate and keeps each connection's GTV, which one virtual battery would ignore; it is not yet built (section 16). The intake asks whether the optimizer trades PV and consumption, whether its reported revenue includes their imbalance, and whether the battery is pooled, with the allocation key.

Two runs per day. Each day is solved twice. Run 1, at full availability, is the full potential and the denominator of the capture rate. Run 2 fixes the output at zero in the quarter-hours of downtime (section 10) and sells no capacity in a block that contains downtime, because a contracted block must be available for its whole duration; it sets the expected results of the trading drivers. The difference between the two runs is the downtime driver. The positions of both runs are valued with the formulas of section 6, so that realised results and full potential are measured alike, and a gap reflects dispatch decisions, not differences in accounting. Revenue drivers are calculated per asset from its own positions; grid costs and energy tax are allocated to the assets in proportion to their extra grid import. Doing nothing is feasible apart from reaching the stored-energy target at midnight, whose cost the stored energy carried offsets; a day with a negative full potential is reviewed before release.

Solver and reproducibility. Every optimisation is a mixed-integer linear programme per site, day and stage, solved with HiGHS, an open-source solver under the MIT licence (HiGHS), with a time limit of 60 seconds and a relative optimality gap of 0.001% per solve. A binary variable per quarter-hour prevents charging and discharging at the same time, which a battery cannot do. The price impact is applied as 8 linear segments, exact at the segment edges and slightly conservative between them; results are valued with the exact formula. Inputs, code version and solver version are stored with every run, so that any full potential can be recalculated without a licence fee. Client inputs are confidential: an optimizer can rerun the full potential of its own asset only with the client's consent, and an independent validator works on example data or under a confidentiality agreement.

8Expected results

No optimizer reaches the full potential: prices are uncertain at the moment of bidding, bids are subject to gate closures, and capacity markets are competitive. The 80% is based on interviews with market experts and is in line with the calibration used by Modo Energy, which applies a central calibration factor of 80% to its modelled revenues (75% low, 85% high) to account for unavailability, imperfect price foresight and competition in ancillary markets. Modo Energy calibrates on models that leave imbalance out; the full potential of the BESS Backtest includes imbalance within the backtest limits of section 7, so the factor is reviewed together with those limits once realised imbalance results of clients are available. Modo Energy's factor also covers unavailability, which the BESS Backtest reports separately as downtime; this is part of the same review (section 15).

The 80% norm applies to the total only. Drivers compete for the same power and energy, so a separate norm per driver would not be meaningful: one driver can be above its share while another is below. To show where the delta arises, the expected results are allocated to the drivers in proportion to each driver's share of the full potential.

In words: each driver receives the part of the expected results that matches its part of the full potential. The allocated amounts add up to the expected results, and the waterfall of expected results in the dashboard follows from this allocation.

9Capture rate and delta with expected results

The capture rate expresses realised results as a share of the full potential. The delta with expected results expresses the same gap in euros, against the 80% that an optimizer can be expected to achieve.

The capture rate is rounded to whole percentages and classified on the rounded value, so that it does not suggest more precision than the model provides. For periods of several months, euros are summed before dividing; monthly percentages are not averaged. Realised mFRR revenue is left out of both figures while mFRR is not modelled (section 4). With the full potential at full availability in the denominator, downtime lowers the capture rate, while the delta with expected results shows it as a separate driver (section 15). A month with a capture rate above 100% is reviewed by an analyst before it is marked final: the rules of section 7 limit the model where a trader may not be limited, so such a month points to a rule to revisit.

10Downtime

Downtime is the time in which the battery cannot charge or discharge because of a fault or maintenance. It is derived from the EMS log, which records each change in a battery's status, such as active, fault, maintenance or offline, as an event with a start and end time. Every quarter-hour that a fault or maintenance event touches counts as down for that battery, also where the event covers only part of it, because a battery that is unavailable for part of a quarter-hour cannot deliver a full position in it. Other statuses, such as offline or standby, do not count: offline can mean only a lost data connection while the battery runs its last schedule, and the client marks it as a fault where the battery stopped. An event that names no battery applies to every battery of the project. Each event is described by the start, end, type and description in the alarm log. Reduced power and grid restrictions are not counted as downtime.

In words: the expected results of the hours in which the battery was unavailable. The full potential is calculated twice for this purpose (section 7): at full availability, and with zero output in the quarter-hours of downtime and no capacity in the 4-hour blocks that contain downtime. Downtime is shown as a separate driver, so that a fault is not counted against the optimizer. In Revenue by driver, the realised results of this driver are zero, because the battery earned nothing while it was down; the expected results are the amount above; the delta is the amount lost to downtime. The expected results of the other drivers are based on the full potential with actual downtime, so that the drivers together add up to the expected results at full availability. The Downtime table lists each event with start, end, asset, type, duration, and the expected results and full potential lost. The BESS Backtest chart shows each event on a timeline below the axis.

11Periods, annualisation and zoom

• Periods. Months, quarters and years; a year in progress is shown as YTD. Only complete quarters are shown.
• Annualisation. EUR per MW per year = EUR ÷ (MW × days) × 365
• Zoom. Below month level, the charts and their tables show EUR per week, day, hour or quarter-hour, excluding grid costs, which are charged per month. Each quarter-hour of the full potential shows its value added: its cash flow plus the change in stored energy, valued at the average day-ahead price of the day × √RTE, with capacity revenue spread over the quarter-hours of its block; the quarter-hours add up to the total of the day. Cash flow alone would show charging as a loss in the quarter-hour it happens and the revenue hours later. Realised results below month level require an optimizer report per quarter-hour. Key figures, Revenue by driver, Project details and Downtime then cover the months that contain the zoom period.
• Several assets. The view All assets sums the euros and MW of all assets active in the period.

12Missing data

Quarter-hour data are rarely complete: meters fail, data transfers drop out and reports arrive late. The rules below set out how each gap is treated, so that results remain comparable between months and projects and no gap is filled with an assumption that is not visible to the client. Every gap and its treatment is logged.

13Preliminary and final results

The BESS Backtest is filled as data becomes available: the full potential once meter data and project info are in, realised results once the optimizer report is in. No result reaches a client before an analyst has reviewed it (section 5). A month's results are preliminary for three months after the month ends, while meter data, optimizer reports and invoices are still corrected, and every figure is triangulated with at least two sources in that time; the month is then final. Final results are not changed; a later correction is logged and shown with the month concerned.

14Limitations

• Perfect foresight on auctions. The full potential knows the day-ahead and capacity prices of the day in advance, and every capacity bid in its plan is accepted; no optimizer can do this. Expected results at 80% of the full potential account for this. The cap on aFRR capacity prices limits the windfall from rare price spikes. The appropriate factor may differ between markets, for example between FCR and imbalance.
• A model, not a competitor. The full potential represents what was achievable within the rules of the markets, not the result that another optimizer would have achieved.
• Two-day horizon. Each day is optimised together with the next; value from holding energy over longer periods is not included. Stored energy carried between days is valued at the average day-ahead price of the day, which misvalues single days somewhat but largely cancels over a month.
• Sequential markets. The stages follow gate closure and are not optimised together; co-optimising all markets at once can earn more. In a test run, a plan made without hindsight on balancing prices exceeded the full potential on a part of the site-days for this reason (section 15).
• Price impact. The battery's own aFRR energy and imbalance volume moves the price against it by a slope estimated on six months of data with a weak fit; the aFRR bid ladder is not yet used. In the auctions, the battery is a price taker.
• Backtest limits. The full potential uses the settled imbalance price, but only within the backtest limits of section 7. They remove small and uncertain opportunities; a trader with a good forecast may earn more on imbalance than the limits allow, and a month with a capture rate above 100% is reviewed for this reason (section 9).
• Capacity prices. aFRR capacity is valued at the average price of accepted bids, because the highest accepted price is not published; TenneT pays aFRR capacity as bid, so the price the asset would have received can differ. Activated aFRR energy and imbalance are priced on the Dutch merit order, not on the cross-border price of PICASSO (TenneT – Imbalance Pricing System v6.3, TenneT – market update).
• Markets not yet modelled. Intraday and mFRR are left out of the full potential. Realised intraday revenue stays in the capture rate and is compared with imbalance and aFRR energy together; realised mFRR revenue is left out of the capture rate (section 4).
• Degradation costs. A fixed EUR 10 per MWh discharged is used for every project, based on a market pack price and a typical warranted cycle life, not on the warranty terms or degradation curve of the specific battery.
• Outcomes, not causes. The backtest shows where differences arise, not why they arise.

Effect on the capture rate

Some choices in section 7 make the full potential higher than any optimizer can reach; others make it lower. The 80% factor for expected results absorbs the net effect; which side dominates is to be calibrated on the first client months.

15Open questions

The following questions require a decision and will be addressed in future versions of this methodology:

• aFRR energy reservoir. The TenneT aFRR manual v2.8 requires energy for the whole contracted period, which limits a 2-hour battery to about half its rated power in aFRR; German models use 1 hour. The alternative under review counts energy already delivered in the block, which frees part of the reservoir as the block runs. The reading is to be confirmed with TenneT and with an optimizer.
• Full potential below a plan without hindsight. In a test run over July–September 2026, a plan made without knowledge of balancing prices and settled on the actual day exceeded the full potential on 118 of 368 site-days, because the full potential commits capacity and the day-ahead position in stage 1 before balancing is considered. The option under review sets the full potential of a day to the higher of the full potential and the settled result of that plan.
• Model choices to confirm. Several choices of section 7 are proposed and await confirmation: the FCR rules, both aFRR products, reserves within the GTV, aFRR energy without a capacity contract, pro-rata activation and the settlement of shortfall, the exception for restoring charge, the downtime rule, the treatment of pooled batteries and of PV and consumption traded by the optimizer, grid costs, and the solver settings.
• Verification, objections and independent validation. Whether and how the optimizer can recalculate realised results and the full potential, how an objection to a result is handled, and which independent party validates the model, and how often.
• Reporting per asset in pooled portfolios. Optimizers that pool assets may record traded volumes at portfolio level only. Benchmarking the pool as a whole is proposed (section 7); how the pool and its sites are defined, and who confirms the allocation key, needs to be agreed per optimizer.
• Other assets behind the connection. Sites may also contain a combined heat and power unit (CHP), an electric boiler, a heat pump, a heat buffer, wind turbines or EV chargers. These assets share the GTV and the monthly import peak with the battery. Optimizers usually control the whole site, not only the battery: a grower with a CHP, for example, typically has the optimizer dispatch the CHP and the battery together. The question is whether the full potential treats the other assets as given, using their metered output per quarter-hour as is now done for PV and consumption, or optimises them as well (section 16). Treating them as given requires only an electricity meter per asset, but judges the optimizer on the battery alone.
• Capture rate and downtime. The capture rate divides by the full potential at full availability, so downtime lowers it although the delta with expected results shows downtime as a separate driver. The alternative is to divide by the full potential with actual downtime, so that the capture rate measures trading only.
• Stored energy carried in realised results. The full potential reports the change in stored energy over each day as a separate driver. Whether realised results report the same driver, using the stored energy in the EMS log, so that single days compare like with like, is to be decided.
• Peak shaving. The full potential does not lower the connection's monthly peak below that of the site without battery, whereas realised grid costs credit a battery that does. Whether the full potential should include peak shaving, which requires an optimisation over the whole month, is to be decided.
• Reduced available power. Reduced power is not counted as downtime. Whether the run with actual downtime should also use the available power reported by the EMS is to be decided.
• mFRR. Whether batteries can participate in Incident Reserve, and on which terms, before mFRR is added to the full potential.
• Intraday data. When continuous intraday prices per quarter-hour, with their volatility, are collected from licensed EPEX Spot data; until then, intraday is left out of the full potential (section 4).
• Energy tax. Whether, under MLOEA, energy tax applies per allocation point, so that the battery's own annual offtake sets the bracket (Minister of Finance); and whether the exemption for supply to an energy storage facility can apply to a battery on its own allocation point.
• PV under earlier SDE++ rounds. PV installations from rounds before 2023, and smaller installations, keep their subsidy at negative prices. For them, curtailing at negative prices understates the value of storing PV; whether to collect the subsidy decision for such sites is to be decided.
• Small full potential. The full potential can be small, for example in a month with much downtime. How the capture rate is presented when the full potential of a period is small.
• Calibration of the backtest limits and the 80% factor. The minimum price gap of EUR 50 per MWh, the maximum volume and the exclusion of regulation state 2 are reviewed together with the 80% factor once realised imbalance results of clients are available. The review also covers the unavailability included in Modo Energy's factor, and the proposal to limit the ramp rate of passive balancing (ACM).
• FCR activation period. The minimum period for which batteries must deliver FCR is under review by ACER (decision expected by early 2027); the model follows the current 15 minutes until then.

16Backlog

The current version covers the markets and site limits that determine most of the revenue of Dutch batteries, with data that clients and their optimizers can provide today. The features below are planned for later versions. Each would make the backtest more complete, but requires data or model work that client demand does not yet justify.

• Intraday stage. Re-planning against continuous intraday prices per quarter-hour, once these are collected from licensed EPEX Spot data (EPEX Spot market data (EEX Group webshop)).
• mFRR (Incident Reserve). Adding mFRR to the full potential once battery participation is confirmed; activated volumes are already in the TenneT data.
• aFRR bid ladder. Estimating the price impact from TenneT's merit order list of aFRR bids per quarter-hour (TenneT – merit order list aFRR), instead of the slope on activated volume.
• Highest accepted aFRR capacity price. Valuing aFRR capacity at the highest accepted price per block, once a source publishes it.
• Annual cycle budget. Allowing more cycles on the most valuable days within an annual or lifetime limit, instead of a daily average.
• Co-optimised stages. Optimising capacity, day-ahead and real time together, as an upper bound next to the sequential full potential.
• Longer look-ahead. A look-ahead of three days for batteries above 6 hours, as Modo Energy uses, and a look-ahead on every day where the two-day solve is infeasible.
• Peak shaving. Lowering the connection's monthly peak below that of the site without battery, with an optimisation over the whole month.
• Earlier periods. Months before April 2026, with hourly day-ahead prices before 1 October 2025 and 24-hour aFRR blocks before 11 November 2025, and the energy tax rates of 2025 and the brackets below 50,000 kWh.
• Pools across connections. Benchmarking a pool of batteries behind several connections as a whole (section 7).
• Partial availability. Using the available power of a derated battery in the run with actual downtime.
• Backtest limits per project and per balancing product. Limits set per project or per product once realised imbalance results allow calibration.
• Plan without hindsight. Refining the internal stochastic benchmark (section 5): real-time information, PV forecast scenarios, capacity bids without known prices, whole-MW bids and a risk-averse plan.
• Congestion revenue. Compensation under a capacity limitation or steering contract (CBC, CSC) and revenue from redispatch via GOPACS come on top of wholesale and balancing revenue. The limits of such contracts already apply in the full potential when declared (section 7); including the revenue requires the bids, activations and compensation per site and the terms of each contract.
• Supplier-specific degradation. Degradation costs are now a fixed EUR 10 per MWh discharged. A battery-specific cost would use the warranty terms, degradation curve and replacement cost of each battery, which few clients can provide in a comparable form.
• Net results after fees. A view of realised results after the fee of the optimizer and other service costs, next to the gross view. It shows what the owner keeps, and requires the fee terms of each optimizer contract and the monthly invoices.
• Market-specific expected results. Expected results use one factor of 80% on the total. A factor per market, and a band around the capture rate, would require evidence per market that is not yet available.
• Additional revenue from prequalification. A forecast of what the asset could additionally earn if it were prequalified for FCR, aFRR or mFRR, or for more power in those markets, or if the optimizer were mandated for markets now outside its mandate. This shows the value of a prequalification before the client invests in it.
• Verification of FCR and aFRR delivery. A check of FCR and aFRR delivery against the system frequency and the TenneT setpoints requires EMS data per second: about 31.5 million rows per asset per year, against 35,040 rows at quarter-hour resolution.
• Performance alerts. Checks that detect a drop in performance and prompt the ICM to contact the client: a rise in downtime, a market shock, or dispatch that departs from the optimizer's usual pattern. They require live or daily access to the EMS and the meters.
• Site potential. A full potential for the whole site rather than the battery alone, in which the battery, a CHP, an electric boiler and a heat buffer are optimised together. For a grower with a CHP, this captures the trade-off between running the CHP for heat and CO2 for the crops and leaving room for the battery on the grid connection. It requires, per quarter-hour, a heat meter on each heat source (CHP, electric boiler, heat pump) and the level of the heat buffer, a gas meter on the CHP, and the CO2 demand of the crops where the CHP supplies it; in addition gas prices, the price of CO2 emission allowances where these apply, the cost of alternative CO2 supply, and the technical limits of each asset, such as minimum run times and start costs.
• Additional dispatch constraints. Constraints beyond the state of charge, such as fixed trading windows or ramp rates, would make the full potential more site-specific but require the full EMS configuration of each site.

17Sources

All sources consulted for this methodology, including those cited in the text.

18Definitions and abbreviations

Definitions follow TenneT, GOPACS, the ACM and the grid operators where these publish one; the source column shows where each definition comes from. Terms marked ICM are defined for this methodology.

19Changes to this methodology

Every change is published with a new version number and date, and clients are informed by email. After a change, past months are recalculated under the new rules where the data allows; months that cannot be recalculated are marked as not comparable. The three most recent versions are shown first.