In every other high-stakes domain, trust is earned through visibility. Financial modelling remains a notable exception - sadly.
Trust is not given. It is constructed — carefully, incrementally, through repeated demonstration that a system behaves the way its operators claim it does.
This is well understood in medicine, in aviation, in engineering, and increasingly in software. In each of these fields, the mechanisms of trust-building are not informal. They are structural. They are embedded in the process itself. And crucially, they are epistemic — meaning they are built on the ability of an external party to actually know, with evidence, that the thing they are being asked to rely upon is sound.
Financial modelling always has had trust problem. Not because the people building models are untrustworthy. But because the infrastructure for epistemic transparency, which means the ability to show your working in a way that can be independently *and continuously *verified simply does not yet exist in any meaningful form.
What epistemic transparency actually means
The term deserves a precise definition before it is applied.
Epistemic transparency, in the context of a complex system, means that the parties relying on that system can access not just the output, but the reasoning, the assumptions, the structure, and the change history that produced it. It means that trust does not have to be extended on the basis of reputation alone. It can be grounded in direct, traceable evidence.
In software, this looks like open repositories, documented APIs, automated test results, and deployment logs. A well-governed software system leaves a trail. Anyone with the appropriate access can follow that trail backwards from any output to understand exactly how it was produced, what changed, and when.
In aviation, it looks like black boxes, maintenance logs, regulatory inspection regimes, and incident databases that feed continuously back into the design process. Trust in aircraft is not based on the manufacturer’s word. It is based on a vast, interlocking system of documented evidence.
Financial modelling currently offers none of this. What it offers instead is something quite different: authority-based trust and it’s non-existent.
Authority-based trust and its limits
Authority-based trust is the oldest form of trust. It says: you cannot verify this yourself, so trust the person who can.
It is not without value. In a world of genuine complexity, we rely on experts constantly and appropriately. But authority-based trust has a structural vulnerability that epistemic transparency does not: it is entirely dependent on the integrity and competence of the authority, with no independent mechanism to verify either.
When a financial model is presented to a board, a lender, or an investment committee, the trust extended to that model is almost entirely authority-based. The model was built by reputable professionals. It was audited by an independent firm. The numbers have been checked oh and wait “Karen just left to join Goldman Sachs, sorry you’ll have to just work it out.”.
But checked against what? On what date? With what understanding of how the model has evolved since? And with what mechanism for flagging if any of those things change tomorrow?
The honest answer, in most cases, is: against the file as it existed at a point in time, by a team who no longer has visibility of it, with no ongoing mechanism of any kind.
This is not rigorous trust. It is delegated faith.
The model audit as a trust instrument, and why it falls short
The model audit exists, in part, precisely to address this gap. It is the industry’s attempt to introduce a layer of independent verification — to convert authority-based trust into something with slightly more epistemic grounding.
And to its credit, a well-executed model audit does provide real value. Errors are found. Logic is challenged. Assumptions are tested. For a moment, the model is genuinely scrutinised by someone with no stake in its conclusions.
But the audit’s structural limitations undermine its function as a trust instrument in three specific ways.
The first is temporal. The audit reflects the model as it was, not as it is. In a transaction environment where models are revised continuously — sometimes daily — a report issued at signing may be materially misleading about the model that actually informed the final decision.
The second is observational. The auditor sees what is in the file. They do not see what was in the file last week, or what changed between the draft they reviewed and the version that was circulated. The absence of version history means the audit cannot distinguish between a model that has been stable and one that has been substantially rebuilt since the review was commissioned.
The third is structural. The audit is a one-time engagement. Once the report is issued, the auditor’s involvement ends. There is no monitoring function, no ongoing validation, no mechanism for surfacing new errors introduced after sign-off. The model continues to be used — in some cases for years — on the basis of an assurance that was time-limited by design but is treated in practice as indefinite.
The result is a trust instrument with a known expiry date that nobody stamps.
What genuine epistemic transparency would look like
The question is not whether financial models can be held to a higher epistemic standard. They clearly can - blind Freddy can see that. The question is what the infrastructure for doing so would need to look like.
At minimum, it would require three things.
First, continuous rather than point-in-time verification. Validation should not be an event that happens once before a transaction closes. It should be an ongoing state — a system that knows what the model looked like yesterday, flags what has changed today, and surfaces anything that warrants attention before it compounds into something material.
Second, structural transparency. The model’s logic, its dependencies, its key assumptions, and its change history should be accessible in a form that does not require a human to manually trace every cell reference. The audit trail should be automatic, not reconstructed.
Third, independence that persists. The value of independent verification is not only in the initial review. It is in the ongoing relationship between the model and a system that has no interest in the model being correct — only in accurately reporting whether it is.
Platforms beginning to work in this space — Formuly.ai among them — are attempting to operationalise exactly this: embedding the logic of continuous, evidence-based verification into the financial modelling workflow, so that trust in a model is not a one-time credential but a living, demonstrable property of the model itself.
Final thoughts
The industry needs to step up but it needs the willingness to adapt to the technology that is already available that allows it to do so.
First published on Nick Crawley’s Substack.
