Compute Substrate

A Permissionless Computation Layer Without Authority


Note: This document describes a network that records speculative computation only. It does not determine correctness, execute actions, or resolve outcomes. Interpretation and action are external.
Contents

Abstract

We present Compute Substrate, a permissionless proof-of-work network for producing and persisting speculative computation without authority, execution, or resolution. Participants may propose computational outputs and attest to them at a cost. The network aggregates these signals deterministically and commits them to a public ledger. Compute Substrate does not attempt to determine correctness, trigger actions, or assign outcomes. Its sole purpose is to record computation under fixed cryptographic rules. By explicitly separating computation from authority, the system allows unbounded speculative computation while remaining safe to ignore.

1. Introduction

Modern distributed systems increasingly rely on large-scale computation: predictions, rankings, simulations, and analyses generated by many independent actors. In practice, the outputs of these computations are often treated as decisions, triggering actions or reallocating resources. This collapse of layers—allowing computation to accumulate authority—introduces fragility. Errors become catastrophic, incentives distort behavior, and systems attract adversarial pressure proportional to the consequences of being “right.”

This paper explores a narrower question: can computation be scaled independently of authority? That is, can a system persist the results of computation without implying correctness, execution, or obligation?

We present Compute Substrate as a minimal affirmative answer. It is a network that computes and records speculation only. Authority, interpretation, and action are explicitly external. The system remains complete even if all recorded outputs are wrong or ignored forever.

Compute Substrate is deliberately named as a substrate rather than a system. It does not aim to compute answers, resolve uncertainty, or produce outcomes. Instead, it provides a neutral, persistent layer on which computation may be recorded under fixed cryptographic rules. Like a physical substrate, it is indifferent to meaning, correctness, and use. Computation exists on the substrate, not within its authority. Interpretation, validation, and action are explicitly external.

2. Design Goals

Compute Substrate is designed around the following goals:

These goals intentionally exclude many features common to “intelligence” or “governance” systems. The exclusions are essential to safety.

3. System Overview

3.1 Network Model

Compute Substrate is a peer-to-peer network secured by proof-of-work. Nodes perform three roles:

All roles are permissionless.

3.2 Blocks and Consensus

Blocks are linked by hash pointers and ordered by cumulative chainwork. The consensus rule is simple:

The canonical chain is the valid chain with the highest total chainwork.

Proof-of-work provides Sybil resistance and ensures that history is costly to rewrite. Block rewards and transaction fees incentivize inclusion and liveness, not correctness of computation.

3.3 Transactions

Transactions follow a UTXO model. Each transaction may include an optional application payload. Two payload types are defined:

Transactions pay fees. Minimum fees are enforced for PROPOSE and ATTEST payloads to bound spam.

3.4 Proposals

A proposal consists of:

The network does not interpret proposal contents. Meaning exists only to observers.

3.5 Attestations

An attestation references a specific proposal and contributes weight to it. Attestations are also opaque to the protocol; they simply increase a proposal’s score within an epoch.

3.6 Epochs and Aggregation

Time is divided into fixed-length epochs. Within each epoch, proposals are ranked by the sum of attestation weight they receive.

At the end of an epoch, the network derives a deterministic Top-K list for each domain. These lists are part of the canonical state and are queryable by any node.

Importantly:

4. Incentives and Fees

Compute Substrate uses incentives only to ensure participation and liveness.

There are no rewards for being correct. Fees are paid regardless of outcome. This prevents computation from becoming authority-bearing.

5. State, Forks, and Reorganizations

Like any proof-of-work chain, Compute Substrate may experience temporary forks. Nodes track block headers, cumulative chainwork, and maintain undo logs for state transitions.

On reorganization:

This ensures that aggregated outputs always correspond to the canonical chain.

Although the system does not execute actions, correctness of reorg handling is essential for determinism and verifiability.

6. Security Properties

6.1 Adversarial Computation

Malicious or incorrect proposals are allowed by design. They are harmless because they carry no authority. Compute Substrate is not designed to be correct; it is designed to be reproducible.

6.2 Spam Resistance

Spam is bounded by transaction fees and proof-of-work costs.

6.3 No Oracle Attacks

Because outcomes are never resolved, there is no oracle to corrupt and no incentive to do so.

6.4 Miner Behavior

Miners may reorder or exclude transactions, but can only affect inclusion, not interpretation or execution.

7. Non-Goals

Compute Substrate explicitly does not attempt to:

These functions belong to higher layers.

8. Applications (External)

Compute Substrate is intended to be observed, not relied upon. External systems may use its outputs as:

Any authority to act on these outputs must exist entirely outside the protocol.

9. Conclusion

Compute Substrate demonstrates that a pure computation layer can exist independently of authority, execution, or resolution. By committing only speculative outputs under fixed cryptographic rules, the system enables unbounded computation while remaining safe to ignore. We leave all interpretation and action to external systems.


End of paper. HTML edition for readability and citation. This page is static and carries no authority.