01Ask who was unaware of which information

NC3Rs’ in vivo guidance describes masking across people conducting the study, caring for animals or analysing results. It recognises that complete masking may be impractical while partial masking can still protect particular stages.NC3Rs — Using masking/blinding in in vivo experiments (opens in a new tab)

For a peptide experiment, the relevant hidden information might be the assigned condition, sample origin or another characteristic that creates an expectation about the result. A coded label is helpful only if it actually conceals that information from the relevant person.

The labels single blind and double blind can mean different role combinations. A concrete description is more informative than counting how many people were said to be blinded.

02Consider a measurement involving judgement

In an original hypothetical image-analysis experiment, an assessor chooses which fields are usable before counting cells. Knowing which images belong to the peptide condition could influence borderline choices even if the final counting software is automated.

Masking image labels during selection addresses that particular opportunity for influence. It does not establish that the images were collected without bias or that the segmentation algorithm measures the intended biological property.

NC3Rs’ DRIVER guidance discusses masking in in vitro work and distinguishes influences during conduct, outcome assessment and analysis. It also notes that contextual details can reveal group allocation despite a nominal code.NC3Rs — DRIVER recommendations: risk of bias (opens in a new tab)

03Check whether the code genuinely concealed allocation

A filename such as peptide-treated-01 is not masked by replacing the spreadsheet’s group column with A. The information remains visible elsewhere. Similarly, a consistent date or plate-location pattern can make a code easy to infer.

Visible differences between conditions may make some stages difficult to mask. A useful report identifies that constraint and the remaining protections instead of implying that masking was either perfect everywhere or impossible everywhere.

Keep masking distinct from allocation concealment and randomisation. Concealing an upcoming assignment protects the assignment process; masking later assessment protects a different decision. A random sequence alone does not conceal the resulting labels.

If unmasking occurred early, read why and which subsequent decisions were exposed to the information. A documented exception is more interpretable than a broad blinded claim that omits it.

04Use the report to assess risk, not assign blame

Missing masking details leave uncertainty about the design. They do not prove that bias occurred, just as a masking statement does not prove that every relevant source of bias was removed.

In the fictional image example, the evidence note should distinguish masked field selection from unmasked image acquisition if that is what the methods describe. Combining them into a fully blinded experiment would overstate the protection.

Also check whether analysis choices were fixed before the labels were revealed. Masking through one stage can be undermined if later selective reporting uses the revealed groups to choose a favourable endpoint.

A clear conclusion identifies which decisions were protected, which were not and how much those decisions matter to the outcome. That supports a more precise appraisal than treating blinded as a universal quality badge.