Authoritative rules that make every model safe for banking
Palladium’s Rulebook holds the regulatory requirements you work under and the policies your institution has set, so a capable language model can explain a decision without ever being the one to make it.
The Rulebook
Palladium’s Rulebook encodes the decisions behind regulated customer work and the conditions that govern each one. It covers eligibility, dollar thresholds, regulatory deadlines and escalation paths.
- Written as versioned, testable code
- Checks every request automatically, before a model sees it
- Covers eligibility, limits, regulatory clocks and escalation
Average decision on a fully autonomous Reg E dispute
To compliance sign-off on the audit trail
Tell Palladium a customer is disputing a card charge, and it won’t simply pass the question to a model. It checks the claim against the Reg E window, issues provisional credit when the rule allows it, and starts the investigation clock, all before anyone on your team has to touch the case.
Rules + explanationDeciding beats guessing
The Rulebook spells out which policy applies to which product, customer and situation, and which version is in force. Decisions come from those rules every time, however a model might have phrased the answer.
| Model only LLM assistants | Palladium Rulebook | |
|---|---|---|
| Who decides | The model, from its training and the prompt | Versioned rules in code. The model only explains. |
| Consistency | Answers can drift between runs | Same input, same output |
| Thresholds | Written into a prompt and hoped for | Enforced in code, outside the model’s control |
| Explanation | Whatever the model generates | Screened to make sure it matches the decision |
| Record | Answers and sources logged | Sealed, replayable step by step, exportable as signed PDFs |
Anchored in policy, not tuned to please
General-purpose models lean toward telling people what they want to hear. Palladium holds to the rule as written, and output screening checks that every explanation matches the decision. Each outcome is sealed with a full record, so every one can be explained.
Grounded in your institution
Palladium turns your policies, thresholds, product terms and escalation paths into your own Rulebook, which sits on top of the regulatory requirements you already work under. Both apply to every request, so agents stay inside the regulation and your institution’s choices at the same time.
Your Rulebook belongs to you, and your data is never used to train models, by default and by contract. Over time, it grows into an asset your institution keeps building on.
Your policies
Your thresholds
Your product terms
Regulatory requirements
Questions about the Rulebook
What is deterministic AI?
A design where written rules make the decision, so the same input always gets the same outcome. A language model reads the request and explains the result, but it never decides. That is how the Rulebook works.
Who writes the rules?
Your team owns them. During the two-week diagnostic, Sphere maps your policy for one request type into the Rulebook with your service and compliance staff, and you approve every rule before it runs.
What happens when a rule changes?
Rules are versioned and testable. Every decision records the rule version that produced it, so an examiner can see which version applied on any date.
What if no rule covers a request?
Palladium does not guess. Anything outside your rules or above your thresholds goes to a person, with the identity check, the case details and the steps so far attached.
Is the Rulebook certified by regulators?
No. The Rulebook maps controls to the requirements you operate under, such as Reg E, ECOA and Reg B, UDAAP, GLBA and SR 26-2. That mapping is not a certification by, or endorsement from, any regulator.
Get more done with AI built for regulated work
Bring one request type, such as Reg E disputes or fee waivers. We map it to your rules and show you a sealed decision record on your own case.