Skip to content

Latest commit

 

History

History
220 lines (123 loc) · 2.61 KB

File metadata and controls

220 lines (123 loc) · 2.61 KB

GLYPH_COMMITMENT_TRUST_MODEL_V1

Status: DRAFT

Purpose:

Define exactly who must be trusted, who must not be trusted, and what can be verified independently.


PARTIES

Publisher

Auditor

Verifier


PUBLISHER

Role:

Creates:

  • corpus
  • suffix array
  • commitment
  • proofs

Publisher is NOT automatically trusted.

Publisher may be:

  • honest
  • mistaken
  • malicious

AUDITOR

Role:

Rebuilds deterministic artifacts.

Checks:

  • corpus fingerprint
  • suffix array correctness
  • commitment correctness

Auditor produces:

AUDIT ATTESTATION

Auditor is independent from Publisher.


VERIFIER

Role:

Checks proof.

Verifier should not need:

  • original GLYPH engine
  • original machine
  • original repository

Verifier consumes:

  • commitment
  • proof
  • audit record

Verifier outputs:

TRUE or FALSE


TRUST ASSUMPTIONS

Trusted:

SHA256

Deterministic build specification

Public commitment format

Untrusted:

Publisher

Network

Storage provider

Transport layer

Repository host


CURRENT MODEL

Evidence Replay

Trust path:

Publisher → Evidence → Replay → Verification


FUTURE MODEL

Commitment Verification

Trust path:

Publisher → Commitment

Auditor → Audit Attestation

Verifier → Independent Verification


MALICIOUS PUBLISHER CASE

Publisher may:

  • modify corpus
  • modify index
  • fabricate absence claims
  • fabricate count claims

Goal:

Detection must occur.


MALICIOUS AUDITOR CASE

Single auditor may fail.

System should allow:

Multiple independent auditors.


MALICIOUS VERIFIER CASE

Verifier cannot create truth.

Verifier only evaluates proof.


CORE PRINCIPLE

Trust should move away from software.

Trust should move toward proofs.


LONG TERM GOAL

A verifier should eventually validate:

Presence

Absence

Count

Interval

without running GLYPH retrieval.


NON-GOALS

Trusting authority.

Trusting vendor.

Trusting company.

Trusting repository.

Trusting AI.


AXIOM

Do not trust the producer.

Verify the claim.