Skip to content

VISION_v1.md

Latest commit

 

History

History
130 lines (86 loc) · 6.35 KB

VISION_v1.md

File metadata and controls

130 lines (86 loc) · 6.35 KB

Mbongo Vision v1 — Deterministic AI Inference Verification Layer

Status: Strategic direction document Last updated: 2026-02-16

1. What Mbongo IS

Mbongo is a blockchain specialized in verifying deterministic AI inference results on-chain. It does not execute AI models. It verifies cryptographic receipts submitted after off-chain inference, then settles the economic outcome on-chain.

Core properties:

  • Deterministic verification. Given the same receipt and the same chain state, every node reaches the same accept/reject decision. No probabilistic outcomes.
  • On-chain settlement. Payment for compute, reward distribution, and slashing are executed as atomic state transitions within block application.
  • Economic security. Proof of Stake (planned for v0.3+) binds executor honesty to capital at risk. Fraud is economically irrational.
  • Replayability and auditability. Every receipt, every verification decision, and every settlement is recorded in the block history and can be replayed from genesis.

2. What Mbongo IS NOT

  • Not a cloud GPU network. Mbongo does not schedule, route, or manage GPU hardware. Execution happens off-chain on infrastructure the executor controls.
  • Not a training platform. Model training is out of scope. Mbongo verifies inference results, not training runs.
  • Not a generic L1. Mbongo does not compete with Ethereum, Solana, or general-purpose chains. It has no smart contract VM and no plans to add one.
  • Not an AI marketplace. Mbongo does not match buyers with sellers. It verifies that work was done correctly and settles payment.
  • Not a DeFi protocol. Mbongo is infrastructure that DeFi protocols consume, not a DeFi product itself.

Mbongo is infrastructure. It provides a single primitive — verified inference — and does it well.

3. v1 Scope (Verification Layer)

Execution model:

  1. AI inference runs off-chain on the executor's hardware.
  2. The executor generates a ComputeReceipt containing the output hash, execution metadata, and a cryptographic proof blob.
  3. The receipt is submitted on-chain as a transaction.
  4. The chain verifies the receipt deterministically using the rules defined in the protocol.
  5. On success, settlement occurs: the submitter pays, the executor is rewarded, and the receipt is finalized.

What v1 does not include:

  • No Proof of Useful Work. Validators do not execute AI workloads.
  • No on-chain model execution. Models never run inside the chain runtime.
  • No heavy compute inside validators. Verification is lightweight and deterministic.

v1 deliverables:

  • Receipt structure and SCALE encoding (see COMPUTE_INTERFACE_v0.1.md).
  • Deterministic verification rules for receipt acceptance and rejection.
  • Economic fee model for task submission and executor compensation.
  • SDK for submitting tasks and retrieving receipts programmatically.
  • Explorer support for visualizing compute task lifecycle.

4. v2+ Evolution

Version Milestone Scope
v0.3 Minimal PoS security Stake-weighted validator set. Slashing for equivocation. No compute integration yet.
v0.4 Receipt standardization Canonical receipt format finalized. Challenge mechanism for disputing fraudulent receipts.
v1.0 Verified inference primitive Receipt verification live on mainnet. SDK stable. Adopted by initial DeFi and oracle integrators.
v2+ Optional on-chain execution PoUW as an opt-in extension. Validators may execute lightweight inference if staked and hardware-attested.

On-chain execution is an optional future expansion. It is not a v1 goal. The verification layer must be proven, adopted, and economically stable before execution is considered.

5. Target Initial Market

First integrations where verified inference creates immediate value:

  • AI-powered DeFi oracles. Price feeds, risk scores, and sentiment analysis where the consuming protocol needs cryptographic proof that the inference was performed correctly.
  • On-chain risk engines. Credit scoring, collateral valuation, and liquidation triggers that must be auditable and reproducible.
  • Parametric insurance. Claim adjudication backed by verifiable AI assessment of real-world data.
  • Governance systems. DAOs that use AI analysis for proposal evaluation and need proof that the analysis was not tampered with.

Positioning: Mbongo is the verified inference layer for decisions that move capital. If the output of an AI model affects money, governance, or automation, that output should be verified on Mbongo.

6. Long-Term Ambition

A neutral, global verification layer for AI decisions that affect capital, governance, and automation. Any AI model, any executor, any consumer — one chain that settles whether the inference was honest.

The end state is not a compute marketplace. It is a trust primitive: a single, credible answer to the question "was this AI output computed correctly?" that any protocol, enterprise, or institution can rely on without trusting the executor.

7. Long-Term Vision (3-10 Years)

The Settlement Layer for Verifiable AI-Driven Capital Allocation

Mbongo's ultimate ambition is not just to verify AI computations. It is to become the global settlement layer for any AI-driven decision that affects capital.

Target Domains

Domain Use Case
Finance Verified trading signals, portfolio rebalancing, market predictions
Insurance Automated claims assessment, risk pricing, fraud detection
Credit AI-based credit scoring with auditable decisions
Risk Engines Real-time risk assessment for DeFi and TradFi
DAO Governance Verifiable AI-assisted proposal analysis and voting recommendations

What This Means

Every AI decision that moves capital will need:

  • Auditability — Who made the decision? What inputs were used?
  • Verifiability — Can the computation be independently verified?
  • Accountability — Is there economic stake behind the decision?

Mbongo provides all three.

The Path

Phase Focus
v1 (Now) Verification layer foundation
v2 PoUW + zkML research partnerships
v3 Native zkML proofs for inference
v4+ Settlement primitive for global AI-capital infrastructure

This is not a 12-month goal. This is a decade-long mission.