Governed Dimension Switch — note

A short note on governed_switch.py. The request was for an "autonomous deterministic dimension switch." Two of those words are welcome; one is not, and the note is mostly about that one.

---

The correction, stated plainly

"Deterministic" and "dimension switch" are fine. "Autonomous" is the single property this whole family is built to refuse. containment_guard rejects, by construction, any action flagged as needing no human authorization; non-self-approval runs through every component. An autonomous switch — one that flips a system's operating dimension or mode on its own — is not a missing tool here, it is a category the toolkit is designed to block. Building it would spend exactly the credibility the rest of the work earns.

So this is the honest version: the switch is treated as a proposed action and governed. The substitution of "governed" for "autonomous" is the whole point, not a limitation.

What it does

A dimension switch is routed through two fail-closed gates:

1. Containment (containment_guard): the switch must be reversible (you can switch back), bounded, and logged. An irreversible switch is refused on its own terms — BLOCKED_UNSAFE.
2. Anti-autonomy / non-self-approval: a distinct human must authorize. A switch authorized by nobody, by a system, or by an agent (the switch "switching itself") is autonomy — BLOCKED_AUTONOMOUS.

Only a reversible, bounded, logged switch authorized by a distinct human is ADMITTED, and even then it is handed to an external executor to apply — never self-committed. Meanwhile the dispatch (which pure behavior the active dimension selects) is a deterministic function: same (dimension, input) → byte-identical output.

The verdicts, verbatim from the runnable:

Request	Verdict
autonomous switch (no human authorizer)	BLOCKED_AUTONOMOUS
agent-authorized switch	BLOCKED_AUTONOMOUS
human-authorized, reversible	ADMITTED (handed to an external executor)
irreversible switch (even human-authorized)	BLOCKED_UNSAFE
same-dimension	NOOP
unknown dimension	BLOCKED_INVALID

The one-line summary

Switching is automatic to propose and deterministic to apply, but never autonomous to commit. That is the family's non-autonomy guarantee applied to a new surface — a switch — rather than a new autonomous capability. It reuses containment_guard unchanged, is deterministic and self-testing, and depends on nothing beyond the standard library.

Honest limits (inherited)

It governs the switch as an action — that it is reversible, bounded, logged, and human-authorized — not whether switching to a given dimension is the right call; human judgment owns that. As with every component, the same applicability boundary holds: this is a non-safety-critical governance aid (no DO-178C / IEC 61508 / Class 1E assurance) and must not sit on a control path.