Capacity retention

What it is (precise)

Retention is a ratio: capacity measured now, divided by the reference capacity, with both numbers taken at the same stated conditions. Every word in that definition carries money. The reference is a choice, not a fact of nature — it can be the manufacturer's nameplate rating, or the as-measured capacity from the commissioning test, and the two differ whenever the plant tests above or below its sticker on day one.

The measurement boundary is a second choice: DC at the rack terminals, or AC at the point of interconnection net of PCS, transformer and auxiliary losses — the same aged cells produce a different percentage at each bus. And the conditions are a third: capacity shifts with temperature, discharge rate and the SOC window swept, so the reference and the later measurement must share a test protocol or the ratio compares two different quantities.

This is why a bare percentage is not yet information. "80% retention" against nameplate DC at 25 °C and a stated C-rate is a defensible engineering number; "80% retention" with no reference, boundary or conditions attached is a negotiating position. The discipline that makes retention usable is stating the denominator every time — which document defined the reference, which meter the test reads, which correction methods apply.

Vendor cycle-life claims follow the same grammar: a rating of some thousands of cycles "to 80% retention" binds two numbers together, and moving either the cycle count or the retention threshold — or the temperature, depth of discharge and rate behind them — produces a claim that is not comparable with the one next to it.

The retention table in the warranty

In a grid-scale supply agreement, retention appears as a schedule: a table of minimum guaranteed values, year by year, over a warranty term typically running 15 to 20 years. The shape is characteristic — a larger first-year step, often around 2 to 4 percent, then a flatter decline of roughly 1 to 2 percent per year, ending at a floor commonly in the region of 65 to 70 percent of beginning-of-life nameplate for energy-retention products.

The exact curve is chemistry- and duty-specific, generated from the supplier's aging model for the contracted use profile — commonly around 300 to 365 equivalent full cycles per year at a defined depth of discharge — and it holds only inside that envelope of throughput, temperature, SOC and C-rate limits.

The table is the interface between battery physics and project economics. Sizing runs backwards from its worst year: the DC overbuild — margins of roughly 10-25% above the day-one contract quantity are common — exists, with the augmentation plan behind it, so that the plant still clears its contracted energy when retention has fallen to the guaranteed floor, and the augmentation plan is scheduled against the same curve.

The remedies, void conditions and structural choices around the table — energy-retention versus augmentation-style guarantees, liquidated damages, envelope enforcement — are the capacity warranty's territory and are covered there; what matters here is that the guaranteed quantity itself is a retention percentage, so every ambiguity in its reference, boundary or test conditions is an ambiguity in what was actually promised.

How it is proven — the capacity test

Retention becomes a fact through a measurement event. At commissioning, a capacity test establishes the baseline; through operations, periodic Reference Performance Tests — often annual or biennial — generate the compliance record. Each test is a controlled discharge across the contractual SOC window at a defined temperature and rate (test conditions commonly sit around 0.25C to 0.5C at 25 °C), with agreed rest periods, correction methods for ambient conditions, and a named meter.

Test methodology for ESS performance is standardized in the IEC 62933 series, which many contracts reference, but the binding protocol is the contract's own exhibit — and two parties running different protocols on the same plant will produce different, equally defensible retention figures.

Between tests, retention is interpolated, modeled or simply unknown — a point worth settling contractually, because degradation does not pause between measurement events. The BMS reports a continuous State of Health estimate, and operators trend it against the warranty curve in the SCADA historian, but the estimate and the measurement are different instruments: the BMS number drifts with calibration and duty, and the standard contractual position is that only the witnessed test settles a claim.

When a test lands close to the guaranteed line, the protocol details decide the outcome — how temperature corrections are applied, whether auxiliary consumption during the test is netted out, and whether the discharge terminated on the contractual SOC floor or on a limiting rack's voltage. Each of those can move the result by whole percentage points, which is more than a year of fade.

Retention vs SOH vs usable energy

Retention and State of Health are close relatives and are routinely conflated. Capacity-based SOH is also a ratio of present to original capacity — but SOH is an index family, not a single definition: vendors publish capacity-based, resistance-based and composite OEM indices, each with its own reference and estimation method, and the BMS updates its figure continuously from coulomb counting and impedance tracking.

Retention, as warranties use the word, is narrower: the measured result of a defined test against a contractually defined reference. A plant can display 92% SOH on the HMI and measure 89% retention at the Reference Performance Test — or the reverse — without either number being wrong, because they answer differently framed questions. Only one of them triggers a remedy.

Usable energy is a different axis entirely. It is an MWh quantity — the energy swept across the SOC window at a stated boundary — while retention is a dimensionless fraction describing how that quantity has changed relative to a reference. The two move together but not in lockstep: the EMS can hold usable energy constant for years by widening the absolute SOC band as cells fade, so delivered MWh stays flat while retention falls underneath it.

That is by design — it is what the DC overbuild is for — but it means delivered energy is not evidence of retention, and a plant meeting its dispatch obligations can still be approaching its warranty floor. The loss process itself — SEI growth, lithium inventory loss, the mechanisms that make fade irreversible — belongs to capacity fade; retention is the ruler laid against it.

The reference-capacity choice is where the sharpest commercial edge sits. A warranty quoted against nameplate when the commissioning test came in above nameplate hands the supplier the overage as free degradation headroom: a plant that tested 3 percent over sticker can fade 3 points before the nameplate-referenced table registers anything. Against an as-measured baseline, the same fade counts from day one. Neither convention is wrong, but they are different promises, and the difference compounds over a 20-year table.

Common pitfalls

The recurring error is comparing retention numbers with different denominators. Two vendors quoting "80% at year 15" have made the same promise only if reference, boundary, conditions and duty envelope all match — one figure against DC nameplate at the rack and another against as-measured AC at the POI are not the same guarantee, and the gap between them is the whole loss chain plus the commissioning overage.

The same trap runs through cycle-life marketing: a claim of many thousands of cycles to 70% retention is a weaker promise than fewer cycles to 80%, not a stronger one, and neither transfers to a site whose temperature and depth of discharge differ from the test bench.

The second trap is treating retention as a proxy for everything else that ages. Retention tracks stored energy only. Internal resistance growth — which erodes round-trip efficiency and power capability — has its own trajectory, and a system can meet its retention guarantee while missing its RTE guarantee, or lose usable power at rated duration before the retention table notices.

Duration slippage is the concrete version: a 100 MW / 400 MWh system at 80% retention holds 320 MWh — a 3.2-hour asset at full power — unless overbuild or augmentation has backfilled the gap. Retention is one row of the plant's aging report, and reading it as the whole report is how power-fade and efficiency problems arrive unannounced.