A product carbon footprint can be arithmetically flawless and still come apart under review, because two of its emission factors were built on different versions of the same conversion table.
That table is global warming potential. Every non-CO2 gas in your footprint gets converted into CO2-equivalent using a GWP-100 value, and those values are published by the IPCC, and the IPCC revises them with each assessment report.
Methane is the clearest illustration. In AR4 its GWP-100 was 25. In AR5 it was 28 for biogenic methane and 30 for fossil methane, the first time the two were separated. In AR6 the figures are 27.0 and 29.8. Nitrous oxide went 298, then 265, then 273, moving down and then back up.
None of those revisions is large enough to notice in a spot check. All of them are large enough to make the same activity data produce two different answers.
How a single footprint ends up with two vintages
Almost nobody chooses to mix vintages. It arrives through the supply chain of the data itself.
You pull a national grid factor from one government dataset, a freight factor from a carrier's published figures, a materials factor from a commercial life cycle database, and a supplier's own PCF from a PDF they emailed you. Each of those four sources made its own decision about which assessment report to use, and each of them made it at a different time.
A 2022 vintage of a national factor set may still be on AR5 while the 2024 vintage of a life cycle database has moved to AR6. Your spreadsheet does not know that. It sees four numbers in kgCO2e and adds them up.
The failure is quiet by construction. Every input is defensible on its own. The problem exists only in the relationship between them, which is exactly the kind of thing a spreadsheet is built to hide and an assurance provider is trained to look for.
Why reviewers care more than the size of the error suggests
The absolute impact is usually small. If methane is a minor line in your footprint, moving from AR5 to AR6 might change your total by a fraction of a percent.
Reviewers do not read it as an arithmetic problem. They read it as a control problem, and a control problem generalises.
If a footprint mixes vintages, the organisation producing it does not have a documented methodology that binds every factor to one basis. That opens an obvious follow-up question: what else is inconsistent? Allocation? System boundaries? The treatment of biogenic carbon? A single mixed-vintage finding is cheap to fix and expensive to explain, because it invites the reviewer to test the things that are not cheap to fix.
The year-over-year case is worse. If your 2025 footprint used AR5 and your 2026 footprint uses AR6, the difference between them is partly real reduction and partly a change in the conversion table. Present that as progress without saying so and the comparison is not defensible. Restate the prior year on the new basis and say that you did, and it is.
What "one vintage" actually requires
Picking a vintage takes a minute. Enforcing one takes a system.
Declare it explicitly. The methodology note should name the assessment report, the time horizon (GWP-100 in nearly all corporate and product reporting) and whether climate-carbon feedbacks are included. Most corporate and product reporting uses the values without feedbacks. If your note does not say, someone will eventually have to reverse-engineer the answer from your numbers.
Tag the vintage on every factor, not on the report. A vintage recorded once in a document is a claim about the past. A vintage stored as a field on each emission factor, alongside its source and its own publication date, is a property you can query. The difference shows up the first time you need to prove which basis a two-year-old number was calculated on.
Check supplier PCFs before you accept them. A supplier's footprint arrives already converted to CO2e, which means the vintage decision has already been made for you by someone whose methodology note you may never see. Ask which assessment report they used. It is a reasonable question and a short one, and the answer tells you something about their maturity beyond this specific issue.
Handle the transition deliberately. When you move vintages, move everything in the same reporting period, restate the comparative, and record the date of the switch. A staggered migration is how most organisations end up mixed in the first place.
The pattern underneath
Mixed GWP vintages belong to a family of findings that share one shape: the individual numbers are all correct, and the assembly is not.
Mixed factor vintages generally. Allocation methods that differ between two product lines in the same report. A boundary drawn one way in the model and described another way in the methodology note. In each case there is no false input to find, so nothing looks wrong until someone reconstructs how the total was built.
That reconstruction is the thing worth designing for. A footprint whose lineage can be replayed step by step, with each factor carrying its source, its date and its GWP basis, answers a vintage question in seconds. A footprint that exists as a total in a cell answers it with a week of archaeology, assuming the person who built it still works there.
This is the problem CarbonSKU was built around: binding every number in a footprint to the evidence and the assumptions it came from, so the assembly can be inspected rather than reconstructed.
Whichever way you solve it, the vintage question is coming. It is one of the easiest findings for a reviewer to test and one of the easiest to prevent, and those two facts are why it keeps turning up.