Could you reproduce this figure in eighteen months?

Your assurer signs off the FY2025 disclosure. Eighteen months later they are back for FY2027, and the FY2025 figure is sitting there as a comparative. They pick it and ask you to walk them through how it was built.

The analyst who built it left in June. The workbook has been through four revisions since. Nobody remembers which grid factor release was applied.

This is the test that catches teams who passed all the others.

The question your assurer asks next cycle

Reproducibility is not a separate control. It is what happens when the other three tests — source, sign-off, change history — are examined at a distance of time, by people who were not there.

The distinguishing feature is that you cannot pass it by explaining. Either the retained material regenerates the number or it does not.

What reproducing a figure actually requires

The source data, as received. The invoice, meter export or supplier statement in the form it arrived. Not a summary someone typed into a tab, because a typo in transcription is invisible once the original is gone.

The emission factor, with its version. Publisher, publication year, and the exact value applied. Covered below, because this is where most failures happen.

The boundary and consolidation approach. Which entities were in, which were out, and on what basis. A figure calculated on operational control is a different figure to one on equity share.

The method, written down. Location-based or market-based. Any apportionment applied to shared premises. Any estimate used where actual data was unavailable, and how it was derived.

The four things that go missing first

In rough order of frequency: the factor version, the reason a site was excluded, the apportionment basis for shared premises, and the estimate methodology used where real data was not available by the deadline.

Notice they are all decisions rather than data. The data usually survives because it lives in a system. The decisions live in someone's judgement at the time, and judgement does not persist unless it is written down.

A worked example: rebuilding one Scope 2 number

Illustrative figures, realistic shape.

A company applies a grid emission factor of 0.585 to its FY2025 consumption in October 2026 and discloses 4,796.3 tCO2e. In February 2027 the publisher issues a new release, with the factor at 0.571. Applied to identical consumption, that produces 4,681.5 tCO2e.

Neither number is wrong. They are answers to slightly different questions.

Now imagine your assurer recalculates using the current release, gets 4,681.5, and asks why you disclosed 4,796.3. If you recorded the factor version, this is a thirty-second conversation. If you did not, you are attempting to prove a negative about a decision nobody documented.

Why the emission factor version is the usual failure

Three reasons compound.

It feels like a constant rather than a variable. People treat "the grid factor" as a fact about the world, not as a published figure with a release schedule.

It is applied once, early, by one person, often without comment. There is no natural moment where anyone is prompted to record it.

And it changes the answer materially. A 2.4% movement in a factor moves your headline number by 2.4%, which is comfortably enough to matter.

The IFRS S2 measurement approach requirements expect you to disclose the approach, inputs and assumptions used. The factor and its vintage sit squarely inside that.

Keeping enough to pass the test

The bar is lower than it sounds, and it is almost entirely about capturing decisions when they are made rather than reconstructing them later.

Record the factor source and release year alongside the figure, not in a separate methodology document that drifts out of date. Keep the source file itself, attached to the number it supports. Write down exclusions and apportionments at the moment you decide them, in one line each.

Do that and reproducibility is a by-product. Skip it and reproducibility becomes an archaeology project conducted under time pressure, which is the failure mode described throughout The audit trail behind your climate numbers.

If you want to see what a record that cannot be quietly revised behaves like, you can try to alter ours.

The opinion: reproducibility is the only one of the four tests that gets harder with time rather than easier. Source documents can be dug out later. Approvals can, at a stretch, be reconstructed from calendars and emails. But an undocumented decision made two years ago by someone who has left the company is simply gone, and no amount of effort at the point of testing will recover it.

Common questions

What does reproducibility mean for emissions data?

That someone can take what you retained and rebuild the disclosed figure independently, arriving at the same number. Not approximately the same. The same. It requires the source data, the emission factor with its version, the boundary and consolidation approach, and the method, all retained rather than remembered.

Why do emission factors cause reproducibility failures?

Because publishers revise them and nobody records which release was used. A grid factor applied in October may be superseded by a new release in January. Both are legitimate values. Without a record of which one you applied, you cannot show which figure you disclosed or explain the difference when your assurer recalculates and gets the other one.

How long do I need to be able to reproduce a figure?

Long enough to cover the next assurance cycle at minimum, since your assurer will return and may test comparatives. Retention requirements vary by jurisdiction and by regulator, so check what applies to you rather than assuming a default. The practical planning horizon most teams use is the period until the figure stops appearing as a comparative.

What is the fastest way to fail the reproducibility test?

Staff turnover. The analyst who built the model leaves and takes the undocumented decisions with them: why a site was excluded, which apportionment was applied, which factor release was used. None of it was written down because at the time it did not need to be.