Architecture
The principles Float is built on, and how it proves what it issued
Design principles#
Float assesses, funds and collects. The infrastructure provider keeps running its service and takes no credit risk. Float's funding model will be published before launch.
Portable reputation is built only from repayment outcomes that Float's chain observer verifies on-chain, never from data an infrastructure provider reports.
An infrastructure provider sees its own data, and only aggregate eligibility counts for its users. Individual assessments happen only after a user opts in by connecting its wallet and signing a message. The one deliberate exception is reputation, which is shared across providers with the user's consent.
Devnet and mainnet are fully separate. Devnet data never mixes with mainnet data.
Provider-reported state vs Float-verified evidence#
The most important rule in Float.
| Provider-reported data | Float-verified evidence | |
|---|---|---|
| Source | Subscription records and usage data an infrastructure provider shares with Float | User repayments, seen on-chain by Float's chain observer |
| Used for | Assessment: eligibility and limits | Portable reputation |
| Affects reputation? | ❌ Never | ✅ Yes, the only input |
Subscription and usage records help Float assess a user. They never become reputation evidence.
Because users repay Float on-chain in USDC, Float's chain observer sees repayments directly rather than relying on anyone's report.
The rule is built into how the data is stored, so it holds no matter how the rest of the system behaves.
Float still records every repayment it collects. But a repayment counts as Float-verified reputation evidence only when Float's chain observer sees it on-chain. Where observation is off, repayments are marked as not verifiable and reputation stays empty. See the Roadmap for what is available today.
Verifiable assessments#
Float can prove it issued a specific assessment, the exact outcome and terms, without putting any private user or invoice data anywhere public.
Only a small cryptographic fingerprint of the assessment's public-safe facts ever becomes public. The private detail stays with Float and the owning infrastructure provider. With the user's consent, it can be shared selectively with a third party, such as an auditor or another infrastructure provider. That party can then check the detail they were shown against the fingerprint, without taking Float's word for it.
flowchart TB A["Assessment created"] --> B["Fingerprint of the<br/>public-safe facts<br/>generated and stored"] A --> C["Full private detail<br/>available only to Float and<br/>the owning provider"] B --> G["Many fingerprints batched<br/>into one small value"] G -.->|"public footprint stays tiny<br/>(coming soon)"| P["Anchored on Solana"] C -->|"shared with the<br/>user's consent"| T["Third party<br/>auditor or another provider"] T --> V["Recomputes the fingerprint<br/>from the shared detail"] B --> V V --> OK["Match: the detail is exactly<br/>what Float committed to"]
Fingerprints are batched, and each one gets a proof that it belongs to its batch. Only the single value that represents the whole batch is anchored publicly, so the public footprint stays tiny no matter how many assessments Float issues.
Lifecycle#
Every attestation moves through these states.
stateDiagram-v2 [*] --> Pending Pending --> Submitted: batch sent for anchoring Submitted --> Anchored: confirmed final Submitted --> Pending: submission fails, retried in a later batch Pending --> Revoked: assessment withdrawn Submitted --> Revoked: assessment withdrawn Anchored --> Revoked: assessment withdrawn Anchored --> [*] Revoked --> [*]
On-chain anchoring is coming soon. Fingerprints, batching, proofs and checking shared detail against a fingerprint are already in place. The on-chain step is still in the works, so every attestation currently stays at Pending. Revocation is part of the design and is coming soon too.