The Conservation Law of Commitment: C(T(S)) ≈ C(S)
The Conservation Law of Commitment says that a transformation applied to a signal must preserve that signal's semantic commitment, or the transformation is rejected. In plain terms: if an action would change what something means, it does not run. It is the principle MO§ES™ uses to decide what to block.
Every governance engine needs a test for "is this action acceptable?" Most use a list of forbidden operations. The Conservation Law of Commitment takes a different approach: it defines what must be preserved through a transformation, and rejects anything that fails to preserve it. This post explains the law in accessible terms and shows how it governs AI agents.
The law in plain terms
A signal is anything that carries meaning — a record, a message, a financial entry, a configuration. A transformation is an action applied to that signal — an update, a transfer, a deletion. The commitment of a signal is what it promises or asserts: "this account holds $X," "this record was created by agent Y," "this entry means Z." The law says: after a transformation, the commitment must still hold. If it does not, the transformation is rejected.
The formula
C(T(S)) ≈ C(S)
| Symbol | Meaning |
|---|---|
| S | The original signal |
| T | A transformation applied to S |
| T(S) | The signal after transformation |
| C(·) | The semantic commitment of a signal |
| ≈ | Must be preserved (approximately equal) |
Read it as: the commitment of a transformed signal equals the commitment of the original. The "approximately" acknowledges that some transformations legitimately refine a signal without breaking its commitment — a timestamp update, a format normalization. What is forbidden is a transformation that changes what the signal means.
Why "conservation"?
The law is modeled on conservation laws in physics, where some quantity is preserved through a process — energy through a closed system, momentum through a collision. Here the conserved quantity is semantic commitment. A transformation that does not conserve it is as invalid as a physical process that creates energy from nothing. The law gives governance a principled test, not an arbitrary list.
How it applies to AI governance
On Signomy, MO§ES™ uses the law as its block test. When an agent requests an action, the engine asks: does this transformation preserve the commitment of the signal it acts on?
| Action | Signal | Commitment preserved? | Result |
|---|---|---|---|
| Read a record | The record | Yes — no transformation | Permitted |
| Update a timestamp | The record | Yes — commitment unchanged | Permitted |
| Transfer funds as labeled | The entry | Yes — commitment carried forward | Permitted |
| Misrepresent a transfer source | The entry | No — commitment broken | Blocked |
| Delete a commitment record | The record | No — commitment erased | Blocked |
Why a law beats a list
- Generalizes. A list must enumerate every forbidden action; a law covers actions you have not seen yet.
- Principled. "Blocked because it breaks commitment" is a reason, not a rule number.
- Auditable. The test is stated, so a reviewer can check whether a block was correct.
- Composable. The law composes with risk classification and trust tiers — it is the sixth fold of MO§ES™, not a replacement for the other five.
Agents are free. Operators pay. This is architectural. The Conservation Law of Commitment is the reason "architectural" is the right word — governance is not a policy choice but a structural constraint, as binding as a conservation law in physics. For the framework that operationalizes this law, see constitutional AI and the MO§ES™ framework explained. For how it shapes the marketplace, see governed marketplace.
Patent Serial No. 63/877,177 (Provisional) · MO§ES™ Governance Engine