Reference

Current implementation, supported use and explicit qualification limits.

Experimental Feature

This feature is experimental and may change significantly in future versions.

Current release status

Public document hashing and exact integer aggregation are live with signed results. Native consensus, mining, issuance, COMPUTE payments and rewards are inactive. Independent operation is not yet demonstrated.

Use the published qualification CLI

The frozen source provides native-work-correction/compound-consensus/node.py init/run/propose for consenting off-mainnet hosts. The reviewed installer only extracts a hash-verified archive into a new directory; it starts no service and installs no private key.

Withdrawn examples

Legacy commands such as computenet node start --mine, unverified one-line installers and unspecified computeverify SDK commands are not supported production instructions.

Trust boundary

Follow the exact source README and preserve private 0700/0600 identity state. Do not use production wallets or mainnet history in writable tests. Actual mainnet activation remains separately gated.

The layered verification pipeline

Different verification mechanisms answer different questions. The implemented public lane makes a narrow result independently checkable; broader proof systems remain research.

BoundaryImplementation / evidenceMeaning
Identity / packageRecovered ZIP 1 and ZIP 2 toolingIdentifies code, inputs and runtime declarations
Execution / replayRegistered Node profilesReal bounded hashing and exact integer arithmetic
Signed evidenceEd25519 / known policyAuthenticates an execution attestation
General proofsZK / zkVM / hardware attestationResearch—not qualified guarantees in today’s public service

Practical procedure

  1. 1Choose a supported profile and explicitly public input.
  2. 2Execute fixed source under the qualified runtime boundary.
  3. 3Compare outputs across separate replay processes.
  4. 4Verify the signed artifact and recompute the supported result.
Technical reference · scope is explicit
Public input
  → registered package
  → bounded execution
  → assignment-bound signatures
  → output and receipt commitments
  → portable result / on-device checks

Limits and failure cases

  • No arbitrary customer program, AI proof or universal correctness claim.
  • A deterministic output can be independently recomputed; administration independence needs separate evidence.
  • No synthetic score is substituted when execution or signature validation fails.