document_class string | edition timestamp[s] | evaluation_access dict | files dict | page_count int64 | privacy string | schema_version string | title string |
|---|---|---|---|---|---|---|---|
white_paper | 2026-08-11T00:00:00 | {
"available": true,
"commercial_terms": "paid engagement",
"continuing_access": false,
"credentials": false,
"implementation_access": false,
"mode": "supervised in person with operator physically present",
"remote_access": false,
"system_copy": false
} | {
"ouroboros_recursive_self_improvement_capability_report.docx": {
"bytes": 56711,
"sha256": "1b41e630feeade23f717a8867607cbf3f5da3bcd8e7a33a4582c857b13f51f63"
},
"ouroboros_recursive_self_improvement_capability_report.md": {
"bytes": 42865,
"sha256": "24c94ec33a8b6c90b606cafca6be39c5eff5d4169804a... | 16 | public architecture overview with private run evidence | ouroboros_recursive_self_improvement_capability_report_manifest_v2 | Ouroboros: Recursive Self-Improvement Without Model-Weight Updates |
Ouroboros: Recursive Self-Improvement Without Model-Weight Updates
This repository contains the white paper in Markdown, PDF, and DOCX formats.
The paper presents Ouroboros as a system-level recursive self-improvement capability: a persistent operational control plane that can improve the machinery around fixed-weight foundation models, verify and adopt durable changes, and reuse those changes in later improvement cycles. The detailed run evidence remains private.
Independent evaluation
Independent evaluation is available only as a paid, supervised, in-person engagement with the operator physically present. An evaluator may design and witness adversarial tests and inspect agreed outputs. No implementation access, system copy, credentials, remote access, or continuing access is provided.
Files
ouroboros_recursive_self_improvement_capability_report.md- web-readable paperouroboros_recursive_self_improvement_capability_report.pdf- publication-formatted paperouroboros_recursive_self_improvement_capability_report.docx- editable documentmanifest.json- SHA-256 integrity metadata
Evidence boundary
Run-specific claims are labeled as private evidence. This repository does not contain implementation details, credentials, private run artifacts, or a replication package.
- Downloads last month
- 24