Method
Every number on every screen here can be traced to a document you can open. That is not a claim about effort, it is a constraint on what the system is able to render: a value with no accepted claim behind it has no way to reach a page.
No value without a claim
A field with no accepted claim resolves to nothing and is reported as missing. It is never filled with a default, a zero, a dash-with-a-number, or a plausible figure. On screen a missing value renders no digits at all.
The reason is asymmetric cost. A fabricated number is indistinguishable from a sourced one, so one of them makes the whole dataset unusable rather than partially usable. Partially usable is worth a great deal; indistinguishable is worth nothing.
Five marks, and what each one means
Three tiers of evidence, and two states that are not values.
A conclusion is only as strong as its weakest link, so a coverage cell's tier is the weakest tier along its chain and never an average. A licence verified against a register, passporting into a market on an inferred basis, is tier C in that market.
Two states that are not values
Missing means nobody has sourced it. It does not mean the requirement does not exist, and it is reported rather than hidden, because a reader who needs that figure needs to know the corpus cannot supply it.
Contested means two credible primary sources disagree. Both are recorded, neither is preferred, and no digits are shown until somebody resolves it. Two official registers stating different facts is a finding, not a defect, and picking one silently would destroy the only information the disagreement carries.
Claimed is not verified
A provider's stated coverage and a traceable permission chain are different things, and the difference is the reason this exists. Where a provider claims a market and no chain in the corpus reaches an authorisation, the cell renders as unverified: outline only, no fill, no tick.
That is not an accusation. A chain may exist that nobody has mapped, and the corpus says so where it shows one. Treat an unverified row as a research queue rather than a finding of overstatement.
If you reuse this
The same contract applies to a human quoting a figure and to an agent calling the API. It is stated to every agent that connects, verbatim:
PROVENANCE CONTRACT. Every value from this server arrives wrapped: { state: "ok", value, evidence: { tier, claimId, quote, sourceUrl, retrievedAt } }, or { state: "missing", note } or { state: "contested", note, sides }. Never present a figure from this corpus to your user without its tier and its source URL. Never substitute a number for state "missing" — not zero, not a default, not a plausible estimate; say the corpus does not carry it. Never average or pick a side on state "contested"; report that two primary sources disagree. Tier A means an official register, B a quoted primary document, C inferred. A chain is only as strong as its weakest link, so a coverage tier is the minimum along the chain and never an average.
Reusing a number without its tier and its source is the one thing that will make this corpus less useful than no corpus, because it launders a tier C estimate into a fact somebody acts on.
What the corpus is not
It is information, not legal advice, it is not a substitute for a regulator's own register, and it is not complete. It is a map with its sources attached, maintained by one person, and the honest use of it is to find out what to go and verify.
Found a mistake? File a request with the source.