The source_id uniquely identifies the individual CMIP7 models. Please take note that a CMIP7 source_id:

1. Must be constructed solely using the following characters: a-z, A-Z, 0-9, and the hyphen ("-").

2. Must be limited to 25 characters and generally should be much shorter. Note that the source_id appears in filenames, and a shorter source_id makes the filenames easier to read. Of the 132 CMIP6 models, the median source_id length is 11 characters, with only 10% of the models exceeding 15 characters.

3. Typically includes a model name followed by a version number. The source_id's listed in CMIP5 and CMIP6 serve as examples. Note that for a version like "3.2.4", the decimal points are forbidden by rule 1 above and usually replaced with hyphens: "3-2-4".

4. Does not need to include the name of an institution because the institution_id is recorded separately in CMIP7. Note that if more than one institution adopts the same model for its CMIP simulations, the same source_id should be assigned to all results, but the institutions must coordinate so that when the same experiment is run by two institutions they take care to assign different variant_labels to the files they produce.

5. Must be changed when a model is run at a different horizontal or vertical resolution; models run at two different resolutions must be assigned different source_id's.

6. Should not differ when multiple variants of a model perform experiments for exploring model sensitivity to parameter (or parameterization) changes (as in perturbed physics ensembles); in this case the physics_index, which is part of the variant_label, is used to distinguish the different model versions.

7. Should remain consistent (unchanged) for experiment results considered to have come from the same model, even if some model components are prescribed rather than fully coupled in a particular experiment. If, for example, the oceanic and sea-ice components in an ESM or AOGCM are replaced by prescribed conditions, as in AMIP experiments, the source_id should not change. By using the same source_id across both coupled and uncoupled experiments, users will be able to know definitively whether the differences between two simulations are due to the different experimental conditions (achieved by prescribing one or more components) or due to differences in model physics. When, on the other hand, results from different experiments should be considered as coming from different models (whether because of structural differences, different coupling strategies, or some other reason), assign different source_id's.