An audit trail an examiner will accept
A third of your agent decisions cannot be explained after the fact
A regulator does not ask whether the agent was right. They ask why it did what it did, for a specific case, eighteen months ago. Justify records the four things that make an action defensible and tells you which ones are missing.
The problem
Logs record what happened, not why, and the gap only appears under examination
Agent actions in a regulated workflow get logged the way any service call gets logged: inputs, outputs, a timestamp. None of that answers the question an examiner actually asks, which is what evidence the decision rested on and what would have changed it. By the time anyone discovers the record is insufficient, the actions are already eighteen months old.
The insight
Completeness is a conjunction, and averaging it hides exactly the wrong thing
An action with three of four elements is not 75% defensible. It is indefensible, in the only sense that matters to the person asking. Averaging across a workflow produces a comfortable number that conceals the specific actions with a hole in them — so Justify reports the conjunction, separates consequential actions from routine ones, and names the missing element rather than scoring it.
Per-action completeness across decision, evidence chain, confidence and counterfactual, evaluated as a conjunction; consequential actions isolated; gaps attributed by element and by workflow so the instrumentation fix is specific.
How it works
Four steps, no data science team
What it did, what it read, how sure it was, what would have changed it.
Completeness per action, not per quarter.
Missing confidence on the fast path, missing counterfactuals before the July change. Both are code fixes, not process ones.
A per-case record in a form an examiner reads without help.
Who it is for
The compliance officer who will sit in the examination
Financial services, healthcare and insurance deploying agents into decisions with regulatory consequence.
Pricing
- –Completeness audit
- –Gap attribution
- –Workflow breakdown
- –Continuous audit
- –Case file export
- –Gap alerts
- –Retention policy
- –Self-hosted
- –Regime templates
- –Validation documentation
- –Examination support
Competition
What exists, and what it does not do
| Who | What they do | The gap |
|---|---|---|
| Model risk management platforms | Govern model inventory, validation and documentation. | Built for models that get validated annually. An agent takes thousands of individual actions a day, and these platforms have no per-action unit. |
| Observability and tracing | Record every call with full inputs and outputs. | Records what happened. An examiner asks why, which requires evidence and a counterfactual that nothing is currently capturing. |
| GRC platforms | Track controls, policies and attestations. | Operate at the control level. They will happily record that you have a policy requiring explainability while every action violates it. |
| Exporting logs when asked | The current plan at most companies. | Discovers the gap during the examination, which is the worst possible time. |
This needs regulated-industry access — an introduction, a compliance review and a reference customer — which is the same wall that stopped the vertical products in the first thirty, and nothing about this one is different. The demand is also driven by enforcement timing that is outside our control: an examiner asking the question makes this urgent overnight, and until then it competes with work that has a deadline. Selling anticipatory compliance to a buyer whose regulator has not yet asked is the hardest sale in this document.
Market
On the compliance budget, which is the one budget that does not get cut when the quarter is bad
High contract values, long cycles, structural growth as enforcement arrives. Wrong first company for a solo founder and a strong second one with a design partner.