AI Ratings Agency for Tokenized Assets

Risk dossiers that
agents can read.

Deterministic scoring engine. On-chain anchored. Every number computed, never hallucinated. Built on Mantle.

Dossier Viewer

Asset Ratings

Loading ratings...

Side by Side

Asset Comparison

AssetGradeComposite CollateralRedemptionLiquidity ConcentrationTransparency

How It Works

Anti-Hallucination Architecture

The LLM never produces a score. Scores come from a deterministic rubric with published weights and quantifiable inputs. The LLM writes only the narrative layer, and every number in the prose is validated against computed values.

Deterministic Engine

Same inputs produce the same score, every time. Five dimensions, published weights, reproducible output. No randomness, no LLM in the scoring loop.

On-Chain Anchored

Every dossier is published to DossierRegistry on Mantle Sepolia. The methodology hash pins the rubric version — a scoring change is publicly visible.

Validated Narratives

Plain-English explanations are generated by AI, then validated: every number must match a computed value or the narrative is rejected and regenerated.

Unknown, Not Default

When data is missing, the engine flags it unknown and redistributes weight. It never silently defaults — gaps are visible in every dossier.