AI alignment today faces a structural problem: binary safety classifiers (safe/unsafe) cannot express the nuanced ethical trade-offs that real-world deployment demands. The Eden Protocol proposes a ternary ethical routing architecture where every consequential AI decision passes through a three-state logic gate — Allow (+1), Investigate (0), Deny (-1) — rather than a binary safe/unsafe binary.
The mathematical foundation is balanced ternary arithmetic (Setun 1958, Brusentsov), applied here specifically to alignment routing. The core insight: ethical reasoning is inherently ternary (yes/no/defer-to-human), and forcing it into binary creates exactly the brittle safety behaviours observed in deployed systems.
| Category | Year 1 | Year 2 | Total |
|---|---|---|---|
| Researcher stipend | £90,000 | £95,000 | £185,000 |
| Compute and infrastructure | £40,000 | £45,000 | £85,000 |
| Open-source development | £35,000 | £35,000 | £70,000 |
| Evaluation and testing | £25,000 | £30,000 | £55,000 |
| Publication | £15,000 | £15,000 | £30,000 |
| Contingency (10%) | £20,500 | £22,000 | £42,500 |
| Total | £225,500 | £242,000 | £467,500 |
Bus factor of one. Independent research programme — no formal external peer review beyond documented institutional convergences. Prototype at TRL 0-1. No institutional affiliation.