CCI compliance is an evidence problem

We have spent months building a CCI engine. The scoring and cost mathematics took weeks. Everything since has been governance: who decided, on what grounds, with what authority, and where is the proof. That ratio is not an accident of our roadmap. It is the regime's actual message.

The regime runs on judgement calls

Look at where DISC leaves decisions with the manufacturer: a monthly pricing exception where weekly was not reasonably obtainable. A downward risk adjustment after an anomalous window. A representative share class standing in for its siblings. A determination that simulation is not reasonably performable, opening the benchmark leg. A benchmark's appropriateness itself.

Every one of these is legal to take and dangerous to take silently. Each needs a named human with recorded authority, a written rationale, supporting evidence, retention (representative-class records carry a three-year minimum), and approval by someone other than the proposer where a person made the proposal. The judgement is part of the disclosure. An adjustment you cannot evidence is not an adjustment; it is a finding with a delay on it.

Unknowns must block, not default

The second structural demand: every regulatory input has a state. Approved, proposed, unknown, conflicted, absent, expired. Only approved releases a document. This sounds bureaucratic until you watch what silent defaults do at scale. A fund whose retail-distribution status is unanswered must not slip out of the regime because a field was blank; an unresolved question holds the fund in scope until a human answers it. Two sources disagreeing about FCA authorisation must block until someone resolves the conflict, with reasons, rather than letting the most recent file win.

Our version of this is a five-outcome classifier: every fund, share class, and date resolves to exactly one of eligible, decision required, blocked, unsupported, or route elsewhere, each carrying named reason codes, with out-ranked conditions still visible on the record. The specific vocabulary matters less than the property: one outcome, reasons attached, nothing swallowed.

The platform must not approve its own homework

The strongest design constraint we adopted: the platform never approves its own disclosures. Automation drafts, computes, detects, and proposes. Named humans approve, and their authority is itself a recorded, revocable grant. A service account cannot sign a disclosure. Neither can an AI agent, including ours.

We hold this line because the alternative fails slowly and then suddenly. Self-approving automation looks efficient for exactly as long as nothing is wrong. The first time a regulator asks "who decided this was fair, clear and not misleading?", an audit trail that answers "the batch job" ends the conversation and starts a different, worse one.

Evidence should be generated, not assembled

The end state worth aiming for: compliance evidence produced from the system's real runs, not curated into a binder the week before a review. Ours generates a pack that maps every regulatory case to the tests that prove it, with pass and fail states from an actual execution, and prints "no test mapped" rather than papering over a gap. A readiness report does the same for a whole book: every fund classified, every blocker named with its fix.

Ask one question of any CCI process, including your own: for every judgement in this document, can you show who approved it, when, on what evidence, and can you regenerate the proof on demand?

If the answer involves a shared drive and someone's memory, that is the gap, and it will not close itself between now and June 2027. The calculations were never going to be what fails firms under this regime. The receipts are.

We run CCI readiness assessments that answer exactly that question for your book: classification, blockers, and the evidence state of every judgement call. One report, no workshop theatre.