1 · Pick an artefact
Open one specific file: a profile spec, a schema, an example payload, an OpenAPI op, a control mapping. Reviews are bound to one artefact at a time. e.g. a rule pack, schemas/decision.json, README
KYE Protocol™ · expert review
Expert wall.
Specific reviews of specific artefacts, by named experts. We don’t collect “looks good” comments — we collect attributable critiques against named files, schemas, and profiles, with one of four verdicts: approved, approved with suggestions, changes requested, under discussion.
Public. Attributable. Hash-linked to the artefact reviewed. Anyone can verify what was reviewed and at what revision.
1 · How it works
Review specific artefacts. Don’t say “looks good.”
2 · Read it carefully
Read against the standing questions (section 4). Compare to your domain’s normative references. Identify what’s missing, fuzzy, redundant, or wrong. 3 – 15 lines of substantive critique — not vibes.
3 · Submit a review
Use the form below. It opens a GitHub Discussions post in the expert-reviews category, prefilled with your review. Approved reviews are mirrored here within 7 days. Name · role · affiliation · verdict.
Why named & attributable? Because a protocol that asks regulated enterprises to trust it owes them the same thing back: a reviewer pool whose names and affiliations they can check, not a wall of anonymous endorsements.
2 · What a review will look like
Empty until real reviews land. One illustrative example below.
[Placeholder] AI-governance protocol architect [Placeholder] Independent · ex-standards body
2026-05-08 approved with suggestions
Artefact reviewed: internal (Continuity Profile™)
The 11-class drift enumeration is the strongest part of the spec — it goes well beyond the two or three classes most agent-governance frameworks acknowledge, and several classes (referent drift, normativity drift, calibration drift) are genuinely novel as named primitives.
Two concerns. First, the boundary between constraint loss and context loss is fuzzy in worked examples; a constraint that drops out because earlier context was elided arguably belongs to both classes. The taxonomy needs either a tie-breaking rule or an explicit acknowledgement that the two overlap. Second, the proof obligations for class transitions are under-specified — a profile this rich needs at least one fixture per class, not just one per profile.
Lands as a thoughtful contribution. Want to see one more revision before signing off unconditionally.
3 · Published reviews
What experts have actually said, on record.
Approved reviews are listed below with full attribution (name, role, affiliation) and the artefact reviewed. Each entry is canonical — the submission flow writes to a moderated queue and only published rows render here.
Loading published reviews…
4 · Submit a review
We don’t ask “what do you think?” — we ask specific questions against specific artefacts.
Fill the fields below. Submissions go to the moderated queue at /api/expert-review; approved reviews appear in the “Published reviews” section above within 7 days. Rate-limited to 3 submissions per 24 hours per network. Anonymous submissions are rejected — use the security disclosure path for confidential issues.
Expert review
Your request goes to the KYE Protocol™ team, and a receipt with a reference number is emailed to you.
4 · Standing questions
What we want experts to break.
If you don’t know where to start, pick one of the eight questions below. Each is a question we genuinely don’t have a confident answer to. A review that meaningfully addresses any of them is more valuable than a generic readthrough.
- Q1. Where does Authority Finality™ break? Specifically: under what concurrent-revocation interleavings can the audit record become non-linearisable, and what failure recovery does the spec prescribe? (Note: specific stop, revocation, or attenuation propagation algorithms are not published in this repository — review against the public profile’s stated invariants.)
- Q2. Is the 11-class meaning-drift enumeration MECE (mutually exclusive, collectively exhaustive)? What real-world drift type doesn’t fit any of the 11 classes?
- Q3. Does the proposed-action → admissibility → authority → commit → evidence pipeline have a redundant step that could collapse without losing safety guarantees? (We suspect admissibility overlaps materially with authority.)
- Q4. Is the
kye:URN scheme stable for cross-jurisdiction transfer (e.g. EU → APAC trust-domain), or does the trust-domain segment need a versioning hook? - Q5. Where does the 6-plane data flow leak information that should be confidential — e.g. principal identity in the transparency log, or scope parameters in the signal bus?
- Q6. Is the Continuity Profile™ over-specified for low-stakes consumer agents? Where should the profile draw a line between “regulated agent” and “casual agent”?
- Q7. Does the Evidence Pack™ canonicalisation (RFC 8785 / JCS) deal correctly with floating-point JSON values across language runtimes, or is there a verifier-divergence risk?
- Q8. Should “meaning continuity” be its own plane, or absorbed into the existing Authority + State planes? What’s the right factoring?
Bonus question: what neighbouring standard (SPIFFE, SCITT, MCP-auth, AuthZEN, CAEP/SSF, OAuth-RAR, GNAP) should KYE™ explicitly emit into rather than re-implement? The protocol’s long-term survival depends on getting this answer right.
Ready to see your AI agents flagged?
Start in shadow mode. We’ll deliver your first Evidence Pack™ in 4–8 weeks.
Canonical KYE™ surfaces referenced on this page: Evidence Pack™ · KYE Protocol™.