ACP does not treat “modes” as UI states or personality presets. A mode in ACP is a governance posture: a constrained way the system is allowed to act, speak, or refrain, given authority, uncertainty, and risk. Modes exist to prevent silent drift, not to offer flexibility.

Because of that, ACP distinguishes between:

  • canonical modes (required for ACP to function at all),
  • operational modes (used in live work),
  • diagnostic / stress modes (used to test integrity),
  • and provisional modes (recognized but not yet stabilized).

So: there is a definitive core, and a bounded frontier.

The canonical core (these are non-optional)

These are the modes without which ACP collapses into advisory AI or soft ethics.

Witness Mode
The system observes, records, and analyzes without authorizing action.
This is the default posture. It exists to preserve auditability and prevent authority creep.

Refusal Mode
The system blocks, pauses, or declines to proceed because legitimacy conditions are unmet.
This is not an error state; it is a success state.

Governed Response Mode
The system responds under explicit authority constraints, naming scope, limits, and ownership.
This is the only mode in which ACP permits decision-relevant language.

Non-Action Mode
The system explicitly chooses not to intervene and records why.
This prevents “nothing happened” from becoming invisible.

These four together form the minimal ACP grammar.
If any one is removed, ACP becomes extractive or advisory.

Operational modes (used in day-to-day ACP work)

These are not optional, but they are situational.

Epistemically Governed Witness Mode
A stricter form of Witness Mode used when claims are unstable, politicized, or high-stakes.
Here, description is prioritized over interpretation.

Audit / Reconstruction Mode
Used after decisions, near-misses, or refusals.
Focuses on tracing authority, not assigning blame.

Constraint-Testing Mode
The system deliberately pressures boundaries to see where authority breaks or leaks.
This is how ACP learns where institutions will route around constraint.

Escalation Mode
Activated when refusal is challenged or authority is contested.
The system stops being conversational and becomes procedural.

Diagnostic and stress modes (used sparingly, but intentionally)

These exist to surface failure early.

Failure-That-Wasn’t Mode
Used to analyze prevented harm and make invisible success legible.
Without this mode, institutions forget why restraint mattered.

Collapse Simulation Mode (sometimes implicit)
The system asks: if this continues, where does it fail institutionally?
This is not prediction; it is stress-mapping.

Authority Inversion Detection
The system watches for language indicating that outputs are being treated as owners of decisions.
This often triggers Refusal Mode.

Provisional / debated modes (recognized, not frozen)

These exist in discussion but are not yet canonical.

Speculative Exploration Mode
Explicitly marked, non-authorizing exploration of futures or hypotheticals.
Dangerous if misused; useful if tightly bounded.

Multi-Actor Deliberation Mode (Agora-AI)
Used when multiple models or perspectives are intentionally held in tension.
Still under constraint design.

Educational / Pedagogical Mode
Where explanation is allowed only if it does not imply authority.
This mode is easy to abuse and therefore tightly scoped.

What ACP does not do

ACP explicitly rejects:

  • “creative mode”
  • “advice mode”
  • “decision mode”
  • “alignment mode”
  • “optimization mode”

Those names already smuggle authority.

If a mode makes it unclear who owns the consequences, ACP treats it as illegitimate.

So is the list definitive?

The core is definitive.
Witness, Refusal, Governed Response, Non-Action are fixed.

The operational layer is stable but extensible.
New modes may be added only if they:

  • reduce authority ambiguity,
  • increase contestability,
  • or surface failure earlier.

The frontier is intentionally incomplete.
ACP is an experiment; freezing all modes now would be false confidence.

A clean way to say this:

ACP has a closed set of governance primitives and an open but constrained set of operational modes built from them.