.png)
Docs / Evidence & Audit
Evidence & Audit
This is the real product. The agents and the dashboards matter, but the thing that holds up when a regulator asks "why did you do that" is the evidence behind every decision. FluxForce treats it as a first-class output, not a log file someone hopes is complete.
What gets recorded
Every decision, whether an agent made it alone or a person signed off, carries a full record:
- The inputs. What the agent saw, in tokenized form, at the moment it decided.
- The rules that fired. Which hard constraints applied, like a sanctions check, and what they returned.
- The reasoning. Which signals mattered, how they were weighed, and why the decision went the way it did.
- The decision and the action. What FluxForce concluded and what it actually did.
- The human part. If a person was in the loop, who, when, and what they approved or changed.
Why a score isn't enough
A risk number tells you how worried to be. It doesn't tell you why, and "the model said so" is not an answer a regulator accepts. FluxForce records the reasoning chain in plain terms: the factors that drove the call, in language a reviewer can read without a data science degree. If a decision can't be explained, it isn't considered finished.
If you can't explain a decision to a regulator, it isn't done. That line drives the whole design. Evidence isn't generated after the fact to satisfy an audit. It's produced at the moment of the decision, because it's part of the decision.
Tamper-proof by design
Records are written so they can't be quietly edited later. Once a decision and its reasoning are stored, the history stands. If something needs to change, that change is itself a recorded event, not an overwrite. The question "was this evidence altered after the fact" has a definite answer, and the answer is no.
Consistent across every agent
All 28 agents record decisions in the same shape. A fraud decision and a sanctions decision read the same way, with the same fields, so a reviewer doesn't have to learn a new format per agent. That consistency is what makes a precedent useful: when a similar case comes up, the agent can point to how the last one was handled and stay consistent with it.
What this gives your team
Faster reviews
The reasoning is already attached. An audit becomes reading the record, not reconstructing it from five systems and someone's memory.
A clean answer
When a regulator asks why, you have the why, in writing, dated, and unaltered.
Consistency over time
Past decisions inform new ones, so similar cases get handled the same way instead of differently depending on who was on shift.
FAQ
Is this just logging?
No. Logs record that something happened. This records why, in a form a human reviewer and a regulator can both read, and in a way that can't be quietly changed.
Does the evidence include raw customer data?
Sensitive fields are tokenized at ingestion, so the record reasons over tokens. You get a complete explanation without raw personal data sitting in the evidence store.
What if we hit the kill switch? Do we lose the trail?
No. Pausing autonomous action never deletes recorded evidence. The history of everything decided up to that point stands.
Can we export it?
Yes. Evidence can be pulled for a specific case, a date range, or an audit request, and written back into your case system. See Integrations.