Outcome · Compliance evidence
A supervisor asks about one decision. The answer has to be reconstructed from systems that were never asked to agree.
The log the model wrote, the ticket the handler wrote, the note the reviewer wrote and the export the data team produced are separate accounts of the same week. Assembling them is the work. Defending the assembly is the risk.
One stream, eight renderings — specified, and labelled as such
One governed event stream, eight regulator emitters specified against it. The same threads compile into whichever packs the tenant enables — the record is written once, at decision time, and each regime reads its own rendering of it rather than its own reconstruction.
The first emitter is not starting from a blank page: its basis is the production audit-logging code, which resolves in the repository today. The emitters above it are projections, and the honest bound below says exactly that.
NX/services/nexus-workflows/src/services/audit-emitter.ts
The disclosure that limits itself
How we count this
The clause count is from the demo-data corpus specification’s formal clause lane: 2,900 of 41,000 specified clauses are formalised, and the specification carries the shortfall on its face rather than in a footnote. A certificate that names what it does not assert is worth more than one that implies it asserts everything — that is the design principle, applied to our own coverage figure first.We count the pilot metric from the request to the pack the supervisor accepts, including every follow-up question. A fast pack that triggers three more requests has saved nobody anything, so the follow-up rate is reported beside the days.The falsifier: if the pack is quick because it is thin, the follow-up rate rises and the result is negative. Across the vendor set our research corpus surveys, none publishes a per-decision log schema at all, which is why the comparison here is against your own current process rather than against a competitor’s number.A deadline that is not ours: the DORA register
How we count this
The Register of Information is required by Article 28(3) of DORA and submitted on the templates set by the EBA/EIOPA Implementing Technical Standards. The 2026 window runs 16 February to 13 March 2026 against a 31 December 2025 reference date. It is printed here because it is the nearest date on which “assemble the evidence afterwards” stops being a style choice and becomes a submission you owe.The mechanism: one governed event stream, log or refuse
Every decision emits its record, or it does not take effect. That is the ordering that makes the rest of this page possible: the evidence is not written about the decision afterwards, it is the condition of the decision happening.
The policyholder’s status timeline is a projection of the same log, so the account the customer reads and the account the supervisor reads cannot drift apart.
NX/services/nexus-workflows/src/services/governance.ts
One-Log, Many-Regulator Evidence Fabric
One governed event stream compiles into every regulator’s artefact, instead of four teams reconstructing four different stories from the same week.
NX/services/nexus-workflows/src/services/governance.tsSee it in the console
What the record covers today
One governed stream exists and is in production. The formal clause lane covers 7.1% of the specified clauses, which means 38,100 of 41,000 are checked by people, not by solvers. Today a pack is assembled from a complete record by a person.
REFUSED
A surface that cannot say whose data it is reading refuses to read — no placeholder identifier, no silent organisation-wide query.
This is the failure that produces the worst possible compliance artefact: a complete, confident record of a query that crossed a tenant boundary. Refusing is louder, and it is the only version that stays defensible when someone reads the log a year later.
What you can do next
Your sign-in is valid but names no workspace, and signing in again will not change that. Ask an admin of your organisation to add you to a workspace, then try again here.
Details for support
SCOPE_PREDICATE_MISSING
Start with the decisions you would be asked about first
You do not need the whole book on the record to answer a supervisor. You need the class of decision they ask about — adverse outcomes, referrals and overrides — carrying its own derivation from the day it happened. That is a first pilot, not a programme.
A person replies with two or three times. Not a sequence.