↔
Public challenge
Selected sanitised receipt fixtures with honest limits. A reference exercise is not full native qualification.
Explore →EVIDENCE CUT-OFF · 14 OCTOBER 2026
Measured results and honest status
These are separate source-bound results, not a cumulative test count. Both genuine success and unresolved failure remain part of the record.
986Historical offline casesPassed original pinned source. Not rerun on the later research generation.
341Reconstructed receiptsFrom current four-phase research Smoke.
54.97 MBRetained physical historyAuthenticated locally; four extents, 31,677 frames.
FAILCurrent restricted Reader INITZero native effects; full current native Smoke not run.
The successful four-phase research demonstration exercised real ordinary-account governance and storage but used simulated model and OS-effect endpoints. Earlier Windows native/model demonstrations apply to their distinct historical revisions.
WHAT IS KNOWN / UNKNOWN
From concept to externally challengeable systems
The meaningful difference is between source existence, actual integration, tested behaviour and institutional assurance.
| Evidence class | What is established | What is not established |
|---|---|---|
| Integrated constitutional research | Four phases, six governed calls, permission/refusal, correction, forgetting, retirement and separate reconstruction under simulated endpoints. | Complete current restricted-native execution or production fitness. |
| Historical real-model work | Selected local model/engine/client substitution, warm/cold behaviour and receipt continuity. | Universal model/provider API compatibility or current GPU performance. |
| Historical restricted Windows work | Bounded native containment and Gate 1 protected-storage success in original profiles. | Passing the later restricted Reader Gate B, which failed INIT. |
| Historical P1 systems | Substantial original implementation and identified passing subsets. | Complete forest, portability, storage, custody or product qualification. |
| Independent review | Separately authored code readers and hostile challenge fixtures on founder-hosted evidence. | Certification or independently operated third-party validation. |
| Sustained campaign | Planned 24 hours of credited active time and 100,000 qualifying transitions. | The campaign has NOT passed and must not be presented as complete. |
18 PAGES · COMPLETE DEVELOPMENT & EVIDENCE OVERVIEW
Read the institutional white paper
This paper incorporates the October source/IP investigation and distinguishes integrated P0, developed P1 and separate portfolio technology.
EXECUTIVE SUMMARY
Summon is implemented software for finite permission, governed memory and examinable AI transactions. Historical results are substantial; current full native qualification and independent institutional assurance remain open. The proposed partnership is a defined requirement, controlled diligence and funded qualification.
SOURCE-GROUNDED PUBLIC DISCUSSION
Constitutional Governance for Artificial Intelligence
Institutional White Paper V3 · 19 October 2026 · 18 pages
Browser-native PDF display varies by device. Use the download or full-screen link if it does not render. The paper describes supported architecture and qualification limits; confidential technical records are available only through controlled engagement.
NOT A CERTIFICATE · AN EVIDENCE INVITATION
Open the claims to scrutiny
The source-grounded dossier reconciles preserved engineering records; a distinct institutional reviewer has not yet validated the full system.
◇
Confidential technical review
Relevant source, manifests and evidence can be considered under founder-approved confidentiality and rights terms.
Explore →✓
Joint native challenge
The next material work includes resolving Reader INIT, executing actual current native flows and external hostile challenge.
Explore →A CREDIBLE TEST INVITATION
A pass is evidence. A failure is evidence.
Nothing in this paper asks a partner to accept production readiness, universal safety or independent certification in advance.