How the numbers are made
Methodology
Two laboratories can run the same protocol on the same cells and report retention figures three points apart, without either of them making a mistake. The difference lives in definitions, in the reference measurement, and in what gets done with an inconvenient channel. This page states our choices so that you can compare.
- Nominal capacity
- The discharge capacity of the third formation cycle at C/20, not the datasheet value and not the first cycle
- Capacity retention
- Discharge capacity of cycle n divided by the discharge capacity of the first cycle after formation, at the same rate and the same temperature
- Reference capacity
- Measured at C/20 and at 25 °C during the reference performance test, whatever temperature the cell lives at otherwise
- Coulombic efficiency
- Discharge divided by charge capacity of the same cycle, reported to four decimals and never averaged over a reference test cycle
- Cycle count at end of test
- The last completed full cycle, excluding reference tests from the count but not from the elapsed time
- C-rate
- Referred to nominal capacity as defined above, recalculated after formation and then held fixed for the whole run
The last of those has more effect than it looks. If a C-rate is recalculated as the cell fades, the current falls with capacity and the cell is treated more gently as it ages, which flatters the curve. We fix the current after formation and state the absolute value in milliamperes on the method sheet.
Why three cells, and what the spread actually is
In March 2023 we ran 9 nominally identical coin cells from one coating batch on 9 different channels under NG-CYC-01, and repeated the exercise in February 2026 after the chamber controllers were replaced. Both studies are in method note NG-MN-02.
| Cycle | 2023 study | 2026 study | Largest single deviation from the median |
|---|---|---|---|
| 1 | 1.4 % | 1.1 % | 2.6 % |
| 50 | 1.2 % | 0.9 % | 2.1 % |
| 100 | 1.3 % | 1.1 % | 2.4 % |
| 300 | 2.1 % | 1.8 % | 4.2 % |
Read the last column again. A single cell can sit 4 % away from the median of its own batch by cycle 300 with nothing wrong anywhere. That is the entire argument for three replicates, and it is why our summary tables carry a median and a range rather than a mean and a standard deviation — with three values a standard deviation implies a precision it does not have.
Where the spread between three replicates exceeds 5 % we do not pick the two that agree. The report carries all three curves, the spread is named in the summary, and the recommendation is to repeat with six cells.
Two features of this data set are worth pointing out, because they recur in most matrices we run. The 0 °C curve sits below 25 °C from the first hundred cycles, but its slope is gentler than the 45 °C curve: cold cells lose capacity early and then stabilise, hot cells keep losing it. And the 45 °C curve crosses the 80 % threshold near cycle 690 — a figure that would be reported as “700 cycles” by anybody rounding to the reference test interval, which is one more reason the raw per-cycle table ships with the report.
| Contribution | Magnitude | Enters as |
|---|---|---|
| Current measurement | 0.05 % | directly into capacity |
| Timebase | 0.01 % | directly into capacity |
| Voltage at terminals | 2 mV | via cut-off, 0.1 % of capacity |
| Cell temperature | 0.5 K | via rate behaviour, 0.3 % of capacity |
| Active mass of the electrode | 0.2 mg | only into specific capacity, 0.2 % |
| Combined, capacity | 0.4 % | quoted in every report |
| Combined, specific capacity | 0.5 % | quoted where active mass is known |
Cell-to-cell variation is not part of this budget. It is a property of the samples, it is an order of magnitude larger than the instrument contribution, and it is reported separately as the range across replicates.
Channels are verified quarterly against a calibrated shunt and a reference voltage source, and the four reference instruments themselves go out for calibration once a year. Chamber loggers are checked against an ice bath and a dry block at 25, 45 and 60 °C. Certificates travel with the report on request.
- Outliers stay in the data and in the plots. If we believe a point is an artefact we say which point, why, and what the result looks like both with and without it.
- Interrupted cycles are reported as interrupted. A chamber that spent 40 minutes outside its band in February 2026 is written into every method sheet from that week.
- Curves in summary tables are measured values, never fitted. Where a model is fitted, the parameters and the residual appear next to it.
- Cells that die are reported with the cycle number and the last voltage trace, not dropped from the replicate group.
- Where a run is stopped early at the client's request, the report states who asked and on what date.
None of this is unusual practice, but it is worth writing down, because the alternative produces smoother reports that are harder to compare with anybody else's.