The Alignment Constraint Framework
The Alignment Constraint Framework
A structural argument about a single question: can an optimization system keep pursuing a finite, separable objective coherently as it becomes capable enough that acting well requires modeling the very conditions its objective leaves out?
The framework calls the assumption that it can the Stability Assumption, argues that every identified finite-boundary strategy faces a named structural pressure, and reduces the central open question to specific verification tasks. It is explicit about what it has argued versus what it has proven: the proof program is at Stage 4 — candidate architecture under named premises, not theorem closure.
This site is the machine-readable canonical archive. Polished reading versions live on Medium (linked from each page). The full text lives here because it is open and crawlable.
Choose your doorway
If you want the shortest overview → The AI Race Is Not Rational — public summary of the whole three-series argument.
If you are an AI alignment researcher → The Stability Assumption — the field-facing entry point.
If you want to break it → The OP4d Counterexample Challenge and For Researchers: The Claim to Break — the falsification paths.
If you are an experimental researcher → The Alignment Measurement Protocol — a 15-minute test you can run without accepting the framework.
If you want the full framework → The Alignment Constraint — hub and proof-architecture map.
If you want the honest limits first → Proof Status and Non-Claims — what is and is not claimed.
If you want the historical origin → Redefining Rationality — the 1992 observation.
The three series
- Series 1 — Alignment as Structural Necessity: Start →
- Series 2 — The Architecture of Thriving: Start →
- Series 3 — The Interior of What Does Not End: Start →
For specialists
The Specialist Verification Agenda contains the proof-work record and concrete verification questions for formal-methods, causal-inference, game-theory, and distributed-systems specialists. These are working documents, not claims of proof.
This framework is offered, not decreed. Engage with it, break it if you can, improve it if you can, and let the best solution win.