How the Battery Replacement Data Is Built
Where the data comes from
Every part number is derived from linuxhw/DMI, a public corpus of dmidecode hardware reports contributed by Linux users, licensed CC-BY-4.0. We pin a specific upstream commit (57b21bc6fd) so results are reproducible. This snapshot was generated on 2026-07-25 and covers 1,229 published models out of 12,711 grouped.
We do not scrape battery manufacturers’ compatibility databases. Where we needed to confirm an individual model’s part number, we read the manufacturer’s own published service manual by hand.
Reading the DMI Portable Battery block
Each dmidecode report includes one Portable Battery block per physical battery slot detected, listing a Location, Manufacturer, Name (the reported battery identifier), Design Capacity, Design Voltage, and Chemistry. Machines with two slots — a common design on older business laptops — report two blocks, which we key by their Locationstring (typically “Front” or “Rear”).
Filtering out placeholder and fake batteries
A meaningful share of dmidecode output reports batteries that do not exist: virtual batteries synthesized by firmware (Location: Fake, a “Virtual Battery” manufacturer, a “CRB Battery” name, an impossible “Zinc Air” chemistry), unfilled BIOS template blocks (“Battery 1 Name”), and duplicate blocks describing the same battery twice under two locations. We discard all of these before aggregation. Placeholder names that carry no part number (“Primary”, a bare cell-maker name) still count as evidence a battery exists, but never get published as a part number.
Why we list several part numbers for one slot
Multiple part numbers sometimes appear in reports for the same slot and laptop model. The reports show that those identifiers were observed; they do not tell us why they differ or prove that the batteries are interchangeable. We list every observed part number with its report count rather than choosing one, and ask buyers to verify the replacement against the original battery and the manufacturer’s documentation.
When we refuse to answer
We publish a model only when enough independent reports agree on a stable set of part numbers. Where reports contradict each other in a way that looks like genuinely different hardware under one marketing name, we withhold the result rather than guess. Passing this threshold reduces noisy results, but it does not turn an observed part number into a fit guarantee for every unit.
What this data does not know
- Whether your specific unit’s battery has been replaced with a different part since it left the factory.
- Battery health or cycle count — this data records reported identifiers and specifications, not the condition of any individual battery.
- Whether two observed part numbers are interchangeable or physically fit every variant sold under the same model name.
- Coverage is uneven. Reports come from people running Linux, which over-represents business laptops and under-represents consumer lines.
- The upstream corpus is a snapshot. A laptop released after it was captured will not appear.