Scopes and the World Scope¶

A scope is doxastica's unit of belief ownership. This page explains what scopes are, why doxastica has more than one (the original Kumiho architecture has exactly one), and why one particular scope, the world scope, is privileged in a way that forbids contraction.

What a scope is¶

A scope is a named belief-holder. Every belief lives in exactly one scope, identified by a scope_id string. Within a scope, each belief has its own revision history, its own current value, and its own retraction state. Two scopes are completely independent: revising a belief in agent-a has no effect on a same-named belief in agent-b.

You can think of a scope as a separate "mind." Each scope holds its own coherent picture of the world, and doxastica keeps those pictures from interfering with one another. The Scope model is a thin record (a scope_id and an is_world flag) because the interesting behaviour is in how the core routes operations by scope, not in the scope object itself.

Scopes come into existence on demand. There is no separate "register a scope" step you must remember; writing a belief into a scope creates it if it does not yet exist. You can also create one explicitly with get_or_create_scope, which we return to below.

Multi-scope as a Kumiho extension¶

The Kumiho architecture doxastica implements is described as single-agent: one belief-holder, one belief base. doxastica deliberately extends this to multi-scope.

Why add scopes? Because the systems that consume doxastica rarely model a single mind. A larger application may need to track what each of several agents believes, separate tentative working knowledge from established facts, or keep different subsystems' beliefs apart. Forcing all of that into one undifferentiated base would either lose the distinctions or push the partitioning logic up into every consumer.

Crucially, multi-scope is a superset, not a different model. If you use exactly one scope, doxastica behaves precisely as a faithful single-agent Kumiho implementation. The extension costs nothing when you do not need it, and it is there when you do. This is one of doxastica's two deliberate departures from the paper, covered alongside the other in The Kumiho Architecture.

The reserved world scope¶

One scope id is reserved: WORLD_SCOPE_ID, which has the literal value "__world__". The world scope represents shared, privileged knowledge: facts that are not the private belief of any one agent but hold "for the world."

The id is deliberately dunder-wrapped (__world__) rather than a bare word like "world", so it cannot accidentally collide with a scope a consumer genuinely wants to name "world." When you create a scope whose id equals WORLD_SCOPE_ID, doxastica derives is_world=True on the returned Scope; every other id yields an ordinary scope.

For a worked application of certain world facts alongside a provisional theory in its own scope, see The Detective's Notebook: Revising a Theory Under Contradiction.

flowchart TD
    op["get_or_create_scope(scope_id)"]
    op -->|"scope_id == WORLD_SCOPE_ID"| world["Scope(is_world=True)\nprivileged: contract forbidden"]
    op -->|"any other id"| normal["Scope(is_world=False)\nordinary: all operations allowed"]

Why world-scope contraction is forbidden¶

You can revise and expand in the world scope freely. But contract on the world scope is forbidden: it raises WorldScopeContractionError.

The reasoning is epistemic. Contraction means giving up a belief: ceasing to hold a fact without asserting anything in its place. For a private agent scope, that is sensible: an agent can become uncertain. But the world scope models facts that hold for the world. "Giving up" a world fact, leaving a hole where shared knowledge used to be, is not a coherent operation in this model. If a world fact changes, you revise it to a new value (superseding the old one); you do not retract it into a void.

Forbidding the operation outright, rather than silently doing something surprising, keeps the model honest. The guard is also strict about when it fires: it is the very first thing contract checks, before any storage access, so a forbidden world-scope contraction can never leak a partial write even if the world scope was never created. The practical handling, including how to catch the error and what to do instead, is in How to Retract a Belief with contract.

Reads on the world scope are unrestricted

The restriction is only on contraction. You can query_scope, get_scope_at, and get_revision_chain against the world scope exactly like any other scope. Reconstructing world history at a point in time is a perfectly valid read.

get_or_create_scope semantics¶

get_or_create_scope is the explicit way to ensure a scope exists. Its behaviour is worth understanding precisely:

It is idempotent: calling it twice with the same id returns an equal Scope and never creates a duplicate.
It derives is_world from the id (scope_id == WORLD_SCOPE_ID) rather than taking it as a parameter. You cannot ask for a non-world scope named __world__, nor make an ordinary id privileged; the id determines the privilege.
It does not flip an existing scope's privilege, because privilege is a pure function of the id.

You rarely need to call it directly: the write operations create scopes on demand. Reach for it when you want a scope to exist before you have anything to write into it, or when you want to confirm a scope's is_world status.

Key takeaways¶

A scope is an independent named belief-holder; beliefs in different scopes never interfere.
Multi-scope is doxastica's deliberate extension of single-agent Kumiho; using one scope recovers the original model exactly.
The world scope (WORLD_SCOPE_ID == "__world__") is reserved for shared, privileged knowledge.
Contraction is forbidden on the world scope (you revise world facts, you do not retract them), and the guard fires before any write.
get_or_create_scope is idempotent and derives is_world from the id.