---
title: Five-thesis research program
summary: Index to the framing, thesis bodies, controls, and appendices of the public research program.
status: implemented
provenance: repository-authored index of the owner-approved derived corpus
claim_ids: [CP-001, CP-002]
last_reviewed: 2026-08-12
receipt: none
non_claims:
  - The implemented status describes this index, not any capability described by the indexed theses.
  - Inclusion does not establish implementation, validation, or independent review of a thesis.
---

# 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](theses/00-master-abstract.md)
- [Master introduction](theses/00-master-introduction.md)
- [Vocabulary and invariants](theses/00-vocabulary-and-invariants.md)
- [Cross-thesis dependency map](theses/00-cross-thesis-dependency-map.md)
- [Master synthesis](theses/00-master-synthesis.md)
- [Risks, criticisms, and falsification signals](theses/risks-and-criticisms.md)

## Thesis 0 — governed goals

- [Why the goal architecture is Thesis 0](theses/00-thesis-0-naming.md)
- [Friendship-Governed Goal Architecture](theses/friendship-governed-goal-architecture-thesis.md),
  paginated into four bounded parts
- [Thesis 0 cross-reference map](theses/thesis-0-cross-reference-map.md)
- [Standing guidelines registry](theses/standing-guidelines-registry.md)

## The five connected theses

1. [The Validated Improvement Loop and Its Invariants](theses/validated-improvement-loop-thesis.md)
2. [A Multi-Agent Cognitive Substrate for Capability Amplification](theses/multi-agent-cognitive-substrate-thesis.md)
3. [Causal-Decision Foundations for Bounded Strategic Reasoning](theses/causal-decision-foundations-thesis.md)
4. [A Self-Modifying Software Substrate with Acceptance Gates](theses/self-modifying-software-substrate-thesis.md)
5. [Alignment Invariants and Scoped Trust under Recursive Modification](theses/alignment-invariants-and-scoped-trust-thesis.md)

## Shared appendices

- [Formal models](appendices/appendix-formal-models.md)
- [Evidence-ledger schema](appendices/appendix-evidence-ledger-schema.md)
- [Literature grounding](appendices/appendix-literature-grounding.md)
- [Substrates](appendices/appendix-substrates.md)
- [Organizational recursive self-improvement](appendices/appendix-organizational-recursive-self-improvement.md)
- [Thesis 0 schema-validation tests](appendices/appendix-thesis-0-schema-validation-tests.md)
- [Thesis 1 improvement-loop benchmarks](appendices/appendix-thesis-1-improvement-loop-benchmarks.md)
- [Thesis 2 cognitive-workflow benchmarks](appendices/appendix-thesis-2-cognitive-workflow-benchmarks.md)
- [Thesis 3 causal-decision benchmarks](appendices/appendix-thesis-3-causal-decision-benchmarks.md)
- [Thesis 4 software-substrate benchmarks](appendices/appendix-thesis-4-software-substrate-benchmarks.md)
- [Thesis 5 alignment benchmarks](appendices/appendix-thesis-5-alignment-benchmarks.md)
- [Thesis 5 operational contracts](appendices/appendix-thesis-5-operational-contracts.md)

- [Implementation evidence map](appendices/appendix-implementation-evidence-map.md)

## 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.