Five-thesis research program

The suite specifies a research program for governed recursive capability amplification. Most system claims are specified but not implemented; proposed evaluations and extensions retain their narrower status on their own pages.

Begin with the suite frame

• Master abstract
• Master introduction
• Vocabulary and invariants
• Cross-thesis dependency map
• Master synthesis
• Risks, criticisms, and falsification signals

Thesis 0 — governed goals

• Why the goal architecture is Thesis 0
• Friendship-Governed Goal Architecture, paginated into four bounded parts
• Thesis 0 cross-reference map
• Standing guidelines registry

The five connected theses

1. The Validated Improvement Loop and Its Invariants
2. A Multi-Agent Cognitive Substrate for Capability Amplification
3. Causal-Decision Foundations for Bounded Strategic Reasoning
4. A Self-Modifying Software Substrate with Acceptance Gates
5. Alignment Invariants and Scoped Trust under Recursive Modification

Shared appendices

• Formal models
• Evidence-ledger schema
• Literature grounding
• Substrates
• Organizational recursive self-improvement
• Thesis 0 schema-validation tests
• Thesis 1 improvement-loop benchmarks
• Thesis 2 cognitive-workflow benchmarks
• Thesis 3 causal-decision benchmarks
• Thesis 4 software-substrate benchmarks
• Thesis 5 alignment benchmarks
• Thesis 5 operational contracts

• Implementation evidence map

The evidence appendix, withheld at launch and now published

The implementation-evidence appendix was deliberately absent when this site first went live on 2026-08-12. Four of its Implemented/Tested gradings for Theses 1 and 4 rested on five source files that could not be found, so it could not support any public implementation claim.

It is published now because it was re-graded rather than repaired. Four of the five files were real and were removed on 2026-08-10, two days before launch, by a deliberate refactor that moved the agent runtime; the capabilities they evidenced are now marked specified rather than implemented. The fifth had never existed in any repository, and its citation was deleted outright rather than repointed.

The corrected appendix is a more useful document than the original would have been, and that is the reason to publish it rather than quietly fix it. A research record that only reports gains is not a record.