Skip to content
Indwel

Sovereign Cognition

The whole cognitive act is the system.

A model can be governed. A tool can be governed. A workflow can be governed. The undertaking can still fail in the seams between them.

The category

Govern the undertaking—not only its parts.

Modern AI stacks can contain excellent model controls, access controls, tool permissions, approval steps, and logs while still leaving the cognitive act itself fragmented. Each component may be locally governed while the transitions between them silently change the meaning of the work.

Whole-act governance keeps one authority-bearing undertaking intact from Objective through Evidence, judgement, permission, Effect, observed outcome, Settlement, and continuity.

Read the research argument in Papers →
ObjectiveEvidenceJudgementAuthorityEffectSettlementContinuity

A governed component does not by itself produce governed Work.

Where governance breaks

The seams carry consequence.

The dangerous moments are usually transitions: a document becomes Evidence; a model’s recommendation becomes a decision; a permitted tool call becomes an institutional action; an API response becomes “complete”; a new conversation forgets an unresolved obligation.

Indwel treats those transitions as first-class constitutional boundaries rather than implementation glue.

Evidence seamRetrieved is not admitted.

Source authority, purpose, provenance, and contradiction determine standing.

Authority seamRecommended is not authorised.

Inference can propose. Institutional authority determines consequence.

Settlement seamCalled is not completed.

The observed result—not merely a tool invocation—settles the obligation.

Commercial boundary

Use it where the work is more consequential than the prompt.

Not every AI task needs constitutional machinery. Drafting, ideation, and low-consequence exploration can often remain ordinary model use.

Whole-act governance earns its cost when evidence, authority, continuity, provider independence, auditability, or controlled action materially affect the organisation.

Good fitDecisions with evidence obligations.

Vendor risk, release readiness, policy interpretation, regulated review, benefits administration.

Good fitActions that change institutional state.

Approvals, exceptions, transactions, notifications, records, and controlled external effects.

Good fitWork that survives the conversation.

Multi-day, multi-person, recurring, or stateful undertakings with unresolved obligations.

Apply it

Map a consequential process to the whole act.

The public Ask experience can walk through the architecture deterministically. A real implementation conversation begins with one process and its present sources, decisions, authorities, and effects.