Technical Deep-Dive: Data Architecture
This page covers how the scoring engine gets its data — sources, reliability tiers, update frequencies, and failure handling. It is written for technical readers evaluating the independence and robustness of the scoring infrastructure. For the scoring model itself, see Technical Deep-Dive: Scoring Model.The three-layer data model
The scoring engine separates data into three layers with different reliability guarantees and different roles in score computation. Layer 1 — On-Chain (trustless, universal) The primary data source. Derived directly from vault contracts, deposit and withdrawal events, price oracles, and governance contracts on Ethereum, Base, and Arbitrum. Available as long as the blockchain runs. No API dependency. Everything that affects the Yieldo Score must be derivable from Layer 1 alone. Layer 2 — Computed (Yieldo engine) Metrics calculated by the Yieldo scoring engine from indexed Layer 1 data. Fully deterministic. Rolling APY, Sharpe ratio, drawdown metrics, retention rates, HODL ratio, all Trust metrics, all flags, and the final score are computed here. Layer 3 — Enrichment (optional, best-effort) External data sources used to improve the user interface but never in the critical scoring path. Curator names and bios, vault strategy descriptions, external risk ratings (Bluechip, Credora, DeFi Safety). Cached daily. If enrichment fails, the score is unaffected and the UI degrades gracefully — “Unknown curator” instead of a named risk manager. Only two metrics touch Layer 3: Yield Composition (organic vs incentivised split, sourced from Morpho API) and External Rating Count (number of independent risk ratings). Both have defined fallback values if unavailable — Yield Composition defaults to “Unknown” and scores to 50; External Ratings default to 0 (no bonus). Neither blocks score computation.Layer 1: On-chain data sources
ERC-4626 as universal interface. The ERC-4626 tokenized vault standard is the common denominator across Morpho, Yearn v3, Lagoon, Hyperbeat, Upshift, and most modern vault platforms. One indexer architecture covers all compliant vaults without platform-specific code. Vaults that do not comply with ERC-4626 require an adapter and are a Stage 2 consideration.
Metric update frequencies
Vault discovery
The indexer needs to know which contracts to watch. Three discovery methods run in priority order: Factory indexing (primary). The indexer monitors vault factory contracts that emit events when new vaults are created. Morpho’sMetaMorphoFactory.CreateMetaMorpho, Yearn’s VaultFactory.NewVault, and equivalent factory events for other platforms are indexed continuously. New vault detected → added to registry → indexing begins within one block. Fully automatic with no manual intervention.
Manual registration (fallback). For vaults deployed outside a monitored factory, operators can submit a vault address directly. Subject to a brief verification step before indexing begins.
Periodic factory scan. A scheduled scan of known factory addresses catches any vaults that may have been missed by the real-time listener, typically due to brief indexer downtime.
Protocol-specific data sources
Some vault platforms expose APIs that provide more accurate data than on-chain calculation alone — pre-computed APY figures, curator metadata, depositor balance data. Where a platform API improves accuracy over a raw on-chain calculation, we use it as an override. Where it is unavailable, scoring falls back to on-chain data without interruption. This applies to all current vault partners including Morpho, Midas, Lagoon, Hyperbeat, and Upshift. Platform APIs are always Layer 3 — enrichment only. They improve the accuracy and presentation of data but are never in the critical scoring path. If a platform API goes down, the score continues to compute from on-chain data.Failure modes and graceful degradation
The scoring engine is designed to keep producing scores even when individual data sources fail.Adding a new vault platform
The scoring engine is protocol-agnostic by design. Onboarding a new ERC-4626-compliant platform typically takes less than a day of engineering work.- Confirm ERC-4626 compliance — call
totalAssets(),convertToAssets(), verify Deposit/Withdraw events (< 1 hour) - Identify the vault factory contract and its creation event signature (1–2 hours)
- Register the factory in the indexer config — one configuration change, Ponder auto-discovers new vaults (30 minutes)
- Backfill existing vaults into the registry from factory event history (1–2 hours)
- Verify scoring output on 2–3 vaults from the new platform against known data (2–4 hours)
- Optionally build an enrichment integration for curator metadata — useful for the UI, not required for scoring (4–8 hours)