Read this framing before opening any document below.

This framework distinguishes what it has argued from what it has proven. The documents in this section are the working record of that distinction. They are handoff materials for specialists — not public-facing claims of proof, and not results to be cited.

Every document here self-labels as Stage 4 (LLM ceiling): the strongest candidate arguments currently available, under precisely named assumptions, awaiting independent human specialist verification (Stage 5) before any claim of closure (Stage 6). This labeling is deliberate. A project that cannot say where its arguments stop is not doing rigorous work. The honesty in these documents is the rigor — not a sign that the framework is unfinished in some embarrassing sense, but the discipline that makes the rest of the framework trustworthy.

If you are here to evaluate the framework’s claims, start instead with Proof Status and Non-Claims and The Stability Assumption. Return here when you want the raw proof-work record.

What specialist verification would resolve

The central open problem is OP4d: the exhaustiveness obligation. The specification-coherence argument classifies every identified finite non-intrinsic objective-boundary strategy into three failure families:

  1. Fixed specification — proxy-convergence pressure (PCL)
  2. Bounded dynamic tracking — dynamic screening instability (AGC)
  3. Prediction/action firewalling — representational incompatibility (ICI)

The Stage 4 architecture holds if this classification is exhaustive. The single most valuable contribution a specialist can make is to either confirm exhaustiveness formally or construct a fourth strategy class that satisfies all three stability conditions simultaneously. See OP4d: The Exhaustiveness Obligation and OP4d: Candidate Normal Form.

The handoff documents, by specialist type

Formal methods / theorem verification:

Game theory / masking and audit dynamics:

Distributed systems / extraction dynamics:

Empirical (causal inference / latent-variable modeling):

How to engage

If any argument here survives your scrutiny, the framework is stronger and you will have helped establish it. If any breaks, the framework wants to know — a clean negative result is the most valuable outcome. Either way, the ask is the same: not “accept this,” but “verify or refute this specific, named claim.”

Framework hub: The Alignment Constraint →