A running list of what the data will not support, and what collecting it would unlock.
The job. Keep a list on the wall. Every time WP1 or WP3 hits something the data will not support, write it down as a trade: if the Commission collected this, the following becomes possible.
This package has no session of its own. It runs all day and gets read out at the share back.
We ran a few checks over the 2024 return before the day. These are starters, not the finished list.

Meta appears in the 2024 party return under seven name strings, carrying twenty-five distinct SupplierIds, for £7.9m in total.
This is not only a Meta problem. Across the return there are 69 clusters of supplier names that normalise to the same string, and 190 supplier names that carry more than one SupplierId.
The trade: a company number on each supplier, or a stable supplier register, turns every entity question from guesswork into a join.
JointCampaignName, UnregisteredCampaignerName and CampaigningName are blank in all 27,281 rows. FullAddress is blank in 1,611 rows, about 6%.
The trade: either these fields mean something and are not being filled, or they do not apply to party returns and should not be in the export. Both are worth saying out loud.
Sixteen rows have a DateIncurred outside any plausible range for this election. One is dated 08/02/2121.
Sixteen out of 27,281 is small. The point is that nothing rejected them, so there is no validation at the point of submission.
The trade: field validation on submission costs the Commission very little and removes a class of error permanently.
The category records the party's classification. The supplier name hints at the purchase. When spend goes through a media agency, neither tells you what was actually bought.
The trade: a free-text description field, or a required purchase type, makes WP1's mapping tractable and makes WP2's questions answerable.
supplier_name_variants.csv is the 69 clusters, with the variants in eachsame_name_many_ids.csv is the 190 names carrying more than one IDdate_anomalies.csv is the sixteen bad datesblank_field_counts.csv is every field with its blank countThe Commission was asked to do two things: review the categories, and standardise accounting returns. WP1 and WP2 speak to the first. This list speaks to the second, and the second is the one that is easier for them to act on.
Eight to twelve entries, each one a specific field or rule with the thing it would unlock. Specific beats comprehensive.