Paper 07 · The standard
The Irrecoverable Institution: Why Replayability — Not Explainability — Is the Governance Standard for Agentic Banking
In an agentic institution, the failure mode that matters most is not error. It is irrecoverability.
What this paper argues
Banks are deploying AI systems that decide faster than any governance framework can keep up with. When those decisions fail — and some will — the institution pays for what it cannot reconstruct.
The paper establishes replayability, not explainability, as the correct governance standard. Explainability tells you what the model tends to do. Replayability tells you what the institution actually did — this decision, this context — while there is still time to change what follows.
It also puts a number on the cost of not knowing. Across four analysed institutional failures, an estimated 35–60% of the eventual loss accrued after the original error, in the window where the institution held a signal it could not act on. The sample is small; the range needs wider testing.
Errors can be corrected. Irrecoverable decisions cannot: they compound, cascade, and become the institution.
Terms used here are defined once in the glossary: Decision replayability · Irrecoverability · Decision Integrity Chain™
Start with one decision
Name an important decision made in the last ninety days that your institution could not fully replay today — the rule that applied, the data that was live, the authority that permitted it.