Three Paradigms Compared
| Approach | Decision-Making | Guarantee |
|---|---|---|
| LLM as Brain | Probabilistic decision-making | Cannot guarantee constraints |
| Workflows | Explicit branch coding | Inflexible and difficult to maintain |
| Apollo-1 Truths | Declarative statements | State computation with deterministic application |
How Truths Work
Rather than step-by-step instructions, operators declare what must be true:- “Block payment processing without explicit user confirmation”
- “Require ID verification for refunds over $200”
- “Block ticket cancellation unless passenger is Business Class and Platinum Elite”
Key Benefits
- Composability — Truths combine naturally without requiring explicit integration
- Maintainability — Add, modify, or remove truths independently without cascading changes
- Flexibility without lost control — Natural conversation with deterministic enforcement
- Auditability — Every decision traces back to specific truths and state
- Separation of concerns — Domain experts define truths; the model handles execution
Truth Conflict Resolution
When multiple truths apply to the same entity, Apollo-1 resolves conflicts through per-entity evaluation and priority by strictness:- Block — highest priority
- Ask Question — medium priority
- Activate Tool — lowest priority