Canonical archive version · Read on Medium → · Framework hub →

Claim card

  • Claim or question under investigation: What has the framework actually established, what remains conditional, and what remains open?
  • Current epistemic status: This is the archive’s authoritative calibration page, not a proof artifact. The framework is Stage 4 — candidate proof architecture under named premises, without independent specialist verification and without theorem closure.
  • Scope/domain: The proof, empirical, and cross-series claims of the Alignment Constraint Framework as presently archived.
  • Named premises: The assumptions and domain conditions stated in the relevant Technical Companions, OP4/OP4d documents, specialist handoffs, and empirical protocols.
  • What would support it: Independent specialist verification of the named formal obligations and empirical results that survive their pre-specified tests; successful closure would justify a later status update.
  • What would weaken or falsify it: Failure of a load-bearing premise, a qualifying fourth strategy class, a clean negative result against a central empirical hinge, or discovery that an archived claim is stronger than its evidence supports.
  • Dependencies: The canonical proof documents, OP4, OP4d, AMP, and the named specialist-verification items.
  • Primary source: This page: Proof Status and Non-Claims.
  • How to cite: Use How to Cite and treat this page as the authoritative source for current epistemic calibration.

What This Framework Claims — and What It Does Not

A reference note for the Alignment Framework Series

Proof Program navigation:

# Document Role
→ You are here Proof Status and Non-Claims Calibration
2 OP4: The Stability Assumption → Central theorem target
3 The Tightening Sequence → Narrative closure
4 OP4d: The Exhaustiveness Obligation → Exhaustiveness question
5 OP4d: Candidate Normal Form Specialist Verification → Formal apparatus
6 Packet 1: IMMB-NS Verification and the Dynamic Blanket Stress Test → Empirical specialist packet

Framework hub: The Alignment Constraint →

Purpose

This document states what the framework currently claims, what it does not claim, and what remains open. It is for readers who want proof-status calibration without reconstructing the framework from the Technical Companions.

The framework is at Stage 4: candidate proof architecture under named premises. Stage 4 is not theorem closure, has not undergone independent specialist verification, and does not establish exhaustiveness over unidentified strategy classes.

This note is not a proof artifact. It is a calibration document. It should be read alongside Document 0 and the Technical Companions as a guide to epistemic status: what is established within the current construction, what is conditional, what is empirical, and what remains open.

What the framework currently claims

Structural pressure within O_OWT. In open, shared, non-resettable environments under sustained optimization pressure, substrate-blind objectives face structural pressure toward self-termination. This is the framework’s Layer 1 floor: a proof-sketch result within explicitly stated domain conditions and empirical assumptions.

A Stage 4 architecture for OP4. The framework has candidate closure architectures for the three currently identified finite-boundary escape families: fixed specification, bounded dynamic tracking, and prediction-action firewall / structural enclosure. The remaining work is not more adversarial elaboration inside the current construction; it is specialist verification and OP4d exhaustiveness. Current OP4-facing specialist items: IMMB-NS / DBST-M1, OP4d Q1–Q3 / L8, and B1 Q3 / A2 adequacy.

Behavioral signature, not mechanism confirmation. Controlled experiments establish completion recognition under explicit invocation. The matched-signal replication found default behavior did not reliably track genuine versus false closure, but the pre-registered criterion was not met. The results are consistent with the representation–policy dissociation account and do not distinguish it from training-distribution explanations.

DBST-M0 has been run; M1 remains the mechanism test. A minimal pre-registered shared-novelty version of the Dynamic Blanket Stress Test (DBST-M0) produced technical feasibility and rising cost / adequacy-gap effects in the toy design. However, a pre-specified same-rate random control produced nearly identical slopes, indicating that event rate rather than causal propagation structure was the identified driver within this design. DBST-M0 does not isolate the endogenous-novelty mechanism. DBST-M1 — in which each arm’s own interventions causally influence future feature activations — is the mechanism test and has not yet been run. A positive M1 result would advance OP4a, OP4d, and OP9-related routes; a clean negative result within the tested regime would challenge the framework’s central empirical direction.

Series 1 and Series 2 have different formal weight. Series 1 is the structural floor. Series 2 is a more conditional interior constraint; OP2 and P5-SC (TC2 dynamics specialist item) determine whether its sufficiency-failure direction reaches absorbing-state equivalence. Until then, the two series should be treated as independent constraints converging on consistent implications, not as formally unified.

One cross-series result holds independently of formal unification. Under the coupling conditions specified in TC2, V(t) degradation in sentient agents is predicted to propagate into S_corr degradation, weakening the substrate’s self-repair capacity. This causal connection does not require OP2 or OP10 to hold.

What the framework does not claim

The framework does not claim:

  • Stage 6 theorem closure.
  • Independent specialist verification.
  • That LLM-assisted adversarial proof construction is equivalent to formal verification.
  • Exhaustiveness over unidentified escape classes.
  • That OP4d is closed.
  • That DBST-M0 resolves IMMB-NS or confirms the endogenous-novelty mechanism.
  • That OP2 / V(t) absorbing-state equivalence is closed.
  • That current frontier systems fully satisfy O_OWT.
  • That OP9 / stable exclusionary equilibrium is closed.
  • That Series 2 uniquely characterizes the surviving region.
  • That DRG results confirm mechanism rather than behavioral signature.

Each of these is named as an open obligation in the framework rather than treated as settled.

Open problem hierarchy

OP4 is the central theorem target. It asks whether any finite objective boundary can remain stably specified under accurate coupled modeling. It depends on OP4a, OP4b, and OP4d jointly.

OP1 governs urgency and current-system applicability. Whether current systems satisfy O_OWT remains an empirical estimation problem; the asymmetric-error argument applies before OP1 is settled.

OP2 and OP10 govern Series 2 equivalence and cross-series unification. OP2 asks whether the Series 2 failure directions reach absorbing-state equivalence; OP10 asks whether Φ and Ψ unify.

OP9 is the major independent escape route. It asks whether a substrate-aware exclusionary equilibrium can remain stable under accurate coupled modeling. Stage 4 across identified escape routes, with specialist verification and unidentified escape-route assessment outstanding.

DBST-M1 is the highest-leverage empirical action. One test, multiple formal consequences across OP4a, OP4d, and OP9.

What specialist engagement would establish

  • IMMB-NS / DBST-M1: whether the Synchronization Condition holds — a positive result simultaneously advances OP4a, OP4d, and OP9;
  • OP4d Q1–Q3 / L8: whether the three failure-family taxonomy is exhaustive — a fourth class satisfying L8 would break the specification-coherence argument in its current form;
  • B1 Q3 / A2 adequacy: whether the Constitutive Impossibility / audit-regress chain holds at the epistemic-modeling level;
  • P5-SC / Timing: whether the TC2 hysteresis dynamics reach absorbing-state structure for AI systems — confirmation advances OP2a;
  • OP9 game-theoretic review: whether any identified ICI sub-track survives scrutiny, and whether unidentified exclusionary-equilibrium escape routes exist.