Series · Credentials as protocol · part 2

The credentialing trust stack

Jul 13, 2026 · 1 min read
Content note: Work in progress — outline only.

Work in progress. Public outline; prose to follow.

Why micro-credentials struggle

UNESCO, OECD, ENQA and a pile of 2023–2025 reviews agree: micro-credentials fail at trust — no standardization, fragmented regulation, employer skepticism, inconsistent assessment rigor. The credential itself is fine; the trust stack under it is missing.

The five layers a credential must stack

  1. Identity — who earned it? (secure, verifiable authorship; anti-impersonation)
  2. Assessment integrity — did the assessment measure the skill? (the AI-era problem: generative AI, remote proctoring, authorship verification)
  3. Evidence — what work demonstrates the skill? (verifiable artifacts, not just a pass/fail)
  4. Issuance — who says so, and why should anyone believe them? (issuer authority, reputation, revocation)
  5. Portability — can it move across employers, borders, and systems? (machine-readable, interoperable — Open Badges 3.0, verifiable credentials)

Where each layer is weakest today

  • Identity: remote proctoring is invasive and biased; the alternative (no proctoring) is gamed by generative AI.
  • Assessment integrity: the hardest layer now — AI writes the essay and passes the test. Assessment validity is the new security perimeter.
  • Evidence: most credentials assert, few evidence.
  • Issuance: issuer reputation varies wildly; no common quality bar (TrustEd framework is the leading attempt).
  • Portability: Open Badges 3.0 exists; adoption is patchy; employers don’t read metadata.

The protocol framing

In the agentic web, a credential that isn’t machine-readable is invisible. The credential that wins is the one that’s legible to agents — verifiable, portable, evidenced. That’s a protocol bet, not a product bet. (See strategy: aggregation theory for credentials.)

What Mneurix Lattice does here

  • Sealed-key custody for the issuer identity (the boot-guard pattern).
  • Multi-model council for assessment integrity (with deterministic disagreement resolution).
  • Evidence-first credentials.
  • Machine-readable, portable issuance.

Draft. Prose to follow.

Comments (Giscus) will appear here once the repo Discussions + giscus.app are configured.