Roadmap
Seven phases, ordered by what each unlocks rather than by difficulty. Everything before the last is deliverable without a deployed contract.
Argued in full in the whitepaper at §12.
Phases#
| Phase | Contents | Unlocks |
|---|---|---|
| 0 — Silent calibration | Ingestion, features and the survival ensemble run daily; attestations logged privately | The ability to state confidence honestly |
| 1 — Public attestation | EIP-712 signed attestations over REST, reliability diagram published from day one. No contracts | A public record |
| 2 — Derived and agent | dTVL, emission efficiency, expiry calendar, MCP server | Distribution and agent-path integration |
| 3 — On-chain registry | Attestation commitment, permissionless checkpointing, deposit guard, collateral module. Still unbonded | Contract-level consumption |
| 4 — Bonding | Bond vault, challenge contract, insurance reserve | The guarantee moves from reputational to economic |
| 5 — Multi-attester | k-of-n median with independent bonds and published divergence | Removal of the single-attester assumption |
| 6 — Address-level indexing | Direct measurement of HHI, median tenure and tenure Gini | Cohort decomposition, tighter conformal intervals |
Phase 3 deliberately ships the registry unbonded. That lets integrators build against the interface while the accuracy record keeps accumulating, and avoids sizing a bond before there is data to size it from.
Why Phase 0 is silent#
Why indexing is last#
Address-level indexing is the largest single accuracy improvement available, and it is deferred to the end. It is the only phase needing substantial infrastructure spend, and aggregate estimates are enough to establish whether the thesis holds at all.
If the ensemble does not beat the campaign-expiry baseline on aggregate features, better features will not save it — and that finding is much cheaper to reach without an indexer.
Out of scope#
Stated so that absence reads as a decision rather than an oversight.
- Custody or execution. A durability oracle that also allocates has an interest in its own outputs.
- Price or yield prediction. The horizon is about liquidity persistence; conflating them imports a much harder forecasting problem.
- Protocol quality ratings. Cleaton does not rate teams, code or governance.
- Private or licensed data. A guarantee that cannot be independently recomputed is not a guarantee.