Skip to content

SUBMITTED_APPLICATION_REVIEWER_BRIDGE.md

Latest commit

 

History

History
124 lines (88 loc) · 4.04 KB

SUBMITTED_APPLICATION_REVIEWER_BRIDGE.md

File metadata and controls

124 lines (88 loc) · 4.04 KB
Raw

Submitted Application Reviewer Bridge

This page routes reviewers from already-submitted grant materials to the correct current repository, implementation surface, and evidence path.

NGI TALER application 2026-08-00b

Project: ProofPath Agent Payment Guard

Fund: NGI TALER

Requested amount: EUR 50,000

Canonical repository:

https://github.com/safal207/ProofPath

The submitted application may contain this earlier or incorrect URL:

https://github.com/ProofPath/AgentPaymentGuard

That URL may return 404. The canonical applicant-controlled repository is safal207/ProofPath.

This is a discoverability correction only. It does not change the project scope, budget, or submitted deliverables.

Fast NGI TALER reviewer route

  1. Read GRANT_EVIDENCE_INDEX.md.
  2. Read NGI_TALER_REVIEWER_PATH.md.
  3. Read TALER_ALIGNMENT.md.
  4. Run the Agent Payment Guard demos.
  5. Inspect BUDGET_AND_MILESTONES.md.

Reviewer commands:

git clone https://github.com/safal207/ProofPath.git
cd ProofPath
bash examples/agent-payment-guard/run_demo_check.sh
bash examples/agent-payment-guard/run_service_check.sh
bash examples/agent-payment-guard/run_e2e_evidence_demo.sh
bash examples/agent-payment-guard/run_mock_rail_demo.sh

Expected transition evidence:

valid signed intent        -> ACCEPT
replayed intent             -> BLOCK / INTENT_REPLAYED
policy violation            -> HOLD or BLOCK
ACCEPT                      -> reaches mock rail
HOLD / BLOCK                -> never reaches mock rail
evidence export             -> portable bundle
offline verification        -> OK
tampered evidence           -> verification failure

Current implementation focus

ProofPath Agent Payment Guard treats model-generated payment instructions as proposals rather than authorization.

The current repository demonstrates:

  • signed intent envelope checks;
  • recipient, asset, budget, purpose, and scope policy checks;
  • freshness and expiry checks;
  • persistent nonce replay protection;
  • deterministic ACCEPT, HOLD, and BLOCK decisions;
  • hash-chained audit records;
  • portable evidence export;
  • offline evidence verification;
  • mock payment rail execution only after ACCEPT.

GNU Taler relationship

ProofPath does not replace GNU Taler.

The intended relationship is:

AI agent proposes payment
  -> ProofPath validates human intent and policy
  -> ProofPath emits ACCEPT / HOLD / BLOCK
  -> ACCEPT may be passed to a GNU Taler adapter or another payment rail

A production GNU Taler integration is not currently claimed. The grant-funded transition is to stabilise schemas, privacy-aware evidence, adapter boundaries, fixtures, and integration documentation.

Other historical project names

Some prior materials may have used names or descriptions such as:

  • PythiaLabs / open evidence gates;
  • Liminal Stack / deterministic oversight;
  • LTP + CML / causal trace oversight;
  • Compute Witness;
  • agentic AI evidence gates;
  • verifiable intent for high-risk AI-agent actions.

These names should not be treated as exact aliases. They are related research or implementation surfaces around one broader direction: high-risk agent actions should be reviewable before execution and auditable afterward.

For application 2026-08-00b, the relevant implementation surface is specifically Agent Payment Guard, not the broader ProofPath research portfolio.

Claim boundary

This bridge does not claim that ProofPath currently provides:

  • real wallet custody;
  • real token or bank transfers;
  • private-key management;
  • production GNU Taler integration;
  • regulatory certification;
  • model truthfulness;
  • production security certification;
  • proof of external real-world outcomes.

Reviewer phrase

ProofPath Agent Payment Guard is the current executable implementation for application 2026-08-00b: a pre-execution authorization and evidence layer that treats AI-generated payment instructions as proposals, not authorization.