Addendum updates enforcement evidence and licensing in the open-source release of Elyon-Sol for better accessibility.
**Elyon-Sol** is a deterministic, fail-closed HTTP admission gate derived from a formaladmissibility specification. The canonical model is unchanged from v0.9.8.4:**G(I) = AC³ ∧ T²⁶ ∧ CCS**. This addendum (Revision 4) reports an updated enforcement-evidencesnapshot and a change in how the implementation is distributed, superseding Revision 3(DOI 10.5281/zenodo.20600051). **Licensing change — the material update in this revision.** The canonical model remainspublished for citation. The core implementation is now released as **open source underAGPL-3.0** at https://github.com/Elyon-Sol/Elyon-Sol. This is an open-core model: theadmission gate, the admissibility envelope, the target-side verifier, and the ext-authzsidecar are AGPL; a commercial license is available for uses that cannot accept AGPL's terms;and a separate administration/tooling SDK is proprietary. This supersedes the "proprietary,all rights reserved, not distributed" posture of Revision 3 — the core is now public andinspectable. This document is licensed CC BY 4.0. **The invariants.** All three canonical invariants — Authority (AC³), Coverage (T²⁶), andContinuity (CCS) — are implemented and exercised in code. Continuity is realized at theadmissibility-envelope layer: a content-hashed record of decision state, reasserted againstlive state, where any change in canon, manifest, or evaluator state invalidates a priorELIGIBLE (canon §12.4), and eligibility does not persist across transitions withoutrevalidation (§13). **New in Revision 4.** An OPA/Envoy ext-authz eligibility sidecar (ALLOW/DENY, fail-closed,reusing the production verifier rather than re-implementing cryptography); request-body-derivedbinding so a decision binds to the exact executed bytes; multi-instance replay that failsclosed at startup without a shared store; plus signed published-record freshness, clock-skewtolerance, an envelope inspector, an executor SDK, an issuance log, and an MCP server surface.The filename-level inventory is in the attached document. **Results.** The full repository test suite passes **419 of 419, 0 xfailed** (up from 211 atRevision 3 and 84 at Revision 2). Carried forward from Revision 3: a 204-call enforcement run— 102 REFUSE → 403 with zero external executions; 102 ELIGIBLE → 200 with exactly 102 externalexecutions, each gate-signed. **Honest scope.** No external, third-party adversarial validation on a real multi-host publicsurface has been performed. The enforcement run is author-driven observation over localtransport; the sidecar's real-TLS evidence is a loopback handshake plus a scripted two-VMprocedure, not an external attacker. Real cross-host external red-team validation remains theopen finish line. *This is an evidence publication for provenance, not a build guide.* ================================================================ADDITIONAL NOTES / VERSION CHANGE NOTE (Zenodo "Additional notes" field)================================================================Revision 4 — open-source edition. Supersedes Revision 3 (DOI 10.5281/zenodo.20600051).The canonical model (v0.9.8.4) is unchanged. The material change: the core implementationis now released under AGPL-3.0 (open-core) at github.com/Elyon-Sol/Elyon-Sol, replacing the"proprietary, all rights reserved, not distributed" posture of Revision 3. Records thecapabilities added since Rev 3 (OPA/Envoy ext-authz sidecar, request-body-derivedbinding, multi-instance fail-closed replay) and refreshes the test count to 419/419.Honest scope retained: no external adversarial validation on a real multi-host surface yet.
No takes yet. Share an insight, caveat, or question.
Justin LaPorte (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: