Cross-domain action primitives

KFD Candidate: Cross-Domain Action Primitives

• Candidate status: promoted to KFD-7
• Kind candidate: principle
• Current slot hint: 7
• Slot binding: non-binding

Promotion

This candidate was promoted on 2026-07-18 to KFD-7. The numbered decision is now the authoritative rule. The independent KFD-7 formal reference owns the current non-normative action responsibility geometry.

This document remains as source-candidate lineage. Its original slot hint was non-binding until the explicit promotion allocated KFD-7. The Atlas, Pursuit, and Warrant elaboration candidates were later promoted as numbered drafts KFD-8, KFD-9, and KFD-10 and remain preserved as source lineage.

One sentence

Real-world action must begin from independently addressable, fact-bound direction, perspective, and authority, and return as independently replayable causal experience.

Candidate rule

KFD-1 through KFD-3 establish a fact-grounded world. KFD-4 through KFD-6 make perspective transformation, Primitive discovery, and autonomous correction possible over causal experience. This candidate asks how an intelligent participant acts inside that world.

The current substrate hypothesis is:

Fact     -> what holds under a declared authority and cut
Episode  -> what happened through a bounded encounter with reality

Fact and Episode are not proposed here as ordinary domain Primitives. They are the minimum operational substrate on which action coordinates can be discovered, qualified, and used.

Consequential action requires three independently addressable semantic roles:

direction             -> Pursuit
perspective and cut   -> Atlas
authority boundary    -> Warrant
realized experience   -> Episode

Pursuit, Atlas, and Warrant are the first cross-domain action-Primitive candidates. They direct, observe, and permit action. Episode preserves what actually occurred. None may silently imply another.

This candidate established the category, separation rule, and action loop. It did not fully define the three coordinates. Their later numbered drafts are:

• KFD-8 Atlas, with its candidate lineage;
• KFD-9 Pursuit, with its candidate lineage;
• KFD-10 Warrant, with its candidate lineage.

The promoted formal model, conditional non-derivability argument, valid-action intersection, object/path interpretation, and confidence boundary now live in docs/KFD-7-formal.md.

Action loop

The proposed closure is:

Fact cut
  -> Atlas declares what can be seen
  -> Pursuit declares what continues
  -> Warrant declares what is permitted
  -> action encounters reality
  -> Episode preserves what happened
  -> claims, review, correction, and admission
  -> successor Fact cut

The order is explanatory, not a mandatory user-interface sequence. Low- consequence work may derive inspectable defaults. Simplification may reduce ceremony but must not fuse semantic responsibility.

Low-complexity session limit

KFD-7 is more general than a session model only if it conservatively recovers the familiar session experience when work is locally simple. A representative limit has one local Pursuit, one Atlas adequately represented by the bounded context, one stable Warrant, one contiguous Episode, and sparse Fact change.

Under those conditions, a product may project:

Pursuit  -> goal
Atlas    -> context
Warrant  -> tool permissions
Episode  -> session, run, or transcript
Fact     -> input state and result

The participant should not manage five explicit objects in this limit. The projection must preserve the goal, context boundary, effective permissions, realized execution, and result. It must expand when work crosses Pursuits, perspectives or freshness cuts, authority states, Episodes, or material Fact branches. The session remains a valid local object; it stops being sufficient as the sole model of complex real-world work.

Why the candidate exists

The conditional distinguishability argument shows that direction, perspective, authority, and occurrence can change action or audit conclusions independently. It is a conditional information-separation result, not proof that the current names or object set are universally final.

Treating all four as one mutable task record creates recurrent invalid inferences:

• an active intention appears to authorize action;
• available context appears to be complete reality;
• permission or planning appears to prove occurrence;
• occurrence appears to prove success or completion.

The candidate preserves these differences so that higher-domain Primitives can reuse a stable action basis instead of rebuilding direction, observation, and authority for every field.

Generative role

This candidate does not close Primitive discovery. It creates a recursive ladder:

Fact and Episode substrate
  -> discover cross-domain action coordinates
  -> action coordinates structure richer Episodes
  -> richer Episodes expose domain boundary pressure
  -> KFD-4 through KFD-6 discover higher-domain Primitives

A future domain Primitive may declare its relevant Facts, Atlases, Pursuits, Warrants, and Episode history. It need not display every role in every interaction, but a consequential decision must not fabricate a missing role from another one.

Invalid compressions

The candidate rejects systems that infer:

• authority from intention, context, capability, or occurrence;
• reality completeness from available perspective;
• occurrence from a plan, permission, or expected transition;
• completion from execution or technical success;
• causal experience from before-and-after state alone;
• child authority from parent context without explicit derivation;
• durable direction from a provider session or mutable task body.

Activation handoff

Promotion allocated KFD-7 after the responsibility split, substrate boundary, conditional distinguishability argument, source lineage, and explicit maintainer judgment became sufficiently coherent for a numbered standard decision. Numbering does not certify activation.

The remaining activation gate is owned by decisions/KFD-7.md and requires:

1. deletion witnesses showing that direction, perspective, authority, and occurrence independently change real action or audit conclusions;
2. counterfactual tests that hold two action roles fixed while varying the third;
3. transfer across non-isomorphic work domains;
4. comparison with task, case, workflow, session, goal, approval, capability, and fused alternatives;
5. counterexamples where one or more roles are legitimately absent or safely defaulted;
6. product evidence that the separation lowers reconstruction, authority, or consequence risk without imposing greater total lifecycle burden;
7. a conservative-reduction witness in which a simple task round-trips through the action model and a session-compatible view without semantic loss or manual object ceremony;
8. complexity-breakpoint evidence showing that the product exposes the relevant roles when the low-complexity assumptions fail;
9. evidence that equal state endpoints can contain materially different causal experience;
10. composition evidence showing that causal paths preserve aligned fact-cut boundaries;
11. independent review of whether the rule is ready to become active.

Falsifiers

The candidate weakens or fails if:

• one role is consistently derivable from the others without loss;
• deleting a role does not change safe action, trust, or audit;
• valid work cannot vary the roles independently;
• another representation preserves equivalent decisions at lower total cost;
• progressive disclosure cannot prevent the model from burdening ordinary low-consequence work;
• a simple session cannot be projected into and out of the action model while preserving its bounded decision semantics;
• the product keeps all roles permanently explicit or fails to expand them when the session compression becomes insufficient;
• the three proposed action coordinates do not transfer beyond their originating software-work context;
• the state/path model cannot represent concurrency, compensation, or equal- endpoint causal differences without hidden reconstruction;
• product evidence cannot distinguish ontology value from implementation preference.

Numbering boundary

The original slot hint did not allocate or reserve a number. Explicit promotion created decisions/KFD-7.md, updated the numbered registry, and made KFD-7 the authority. This lineage document does not extend the later KFD-8, KFD-9, or KFD-10 numbered authority.