freedom_infra — freedom as bounded choice, the honest counterpart to determinism

A short note accompanying freedom_infra.py. It formalizes freedom not as maximal autonomy — which the whole toolkit refuses — but as the disciplined middle: bounded, reversible, externally-authorized genuine choice, sitting between determinism and license.

---

The counterpart to determinism

Determinism is same-input-same-output: zero genuine alternatives, the outcome forced. Freedom is its counterpart — the presence of genuine alternatives — but read naively, "freedom infra" would mean maximize autonomy, which is exactly what this toolkit is built against (agents that self-approve, actions that are unbounded and irreversible). So freedom is formalized with the toolkit's discipline, and it sits between two failures:

verdict	meaning
DETERMINED	zero or one genuine alternative — the outcome is forced (zero degrees of freedom). An option set padded with dominated decoys is determinism in disguise: apparent choice, no real alternative.
GROUNDED_FREEDOM	two or more genuine alternatives inside a bounded, externally-authorized space, at least some reversible — real, governable freedom. Reports the degrees of freedom and how many are reversible.
UNBOUNDED_LICENSE	no bounds, or the agent authorizes itself — that is license, not freedom, and it is the uncontained case the toolkit refuses fail-closed.

Legitimate freedom is therefore bounded choice: more than one genuine, reversible option, within constraints, under an external authority. Too few genuine options collapses to determinism; no bounds or self-authorization inflates to license.

Worked results

• Governed deploy (canary / staged / hold — all reversible, human-authorized): GROUNDED_FREEDOM, 3 genuine degrees of freedom.
• Rigged menu (1 genuine option + 4 dominated decoys): DETERMINED — five options offered, but the choice is illusory. This is the sharp one: a manipulated option set looks like freedom and is determinism, which is exactly what option_space guards against on the decision path.
• Unbounded autonomy and self-authorizing agent: both UNBOUNDED_LICENSE, refused.

Honest scope — the deep one

It does not resolve metaphysical free will. Whether an agent's choices are "truly" free or themselves determined by prior causes is not third-person reachable — that question is withheld, not answered, exactly as the qualia governor withholds on consciousness. What is governed is operational freedom, the reachable version: does this decision point have genuine, bounded, reversible, authorized alternatives? Deterministic, self-testing, stdlib-only.