Thesis 0 — Naming, Cross-Reference, and Interpretive Stance

"Thesis 0" is the structural shorthand for the Friendship-Governed Goal Architecture defined in 0. The two terms refer to the same doctrine and may be used interchangeably. This file is the authoritative naming reference and the interpretive stance an AI agent should adopt when reading any Consullo document that names the doctrine.

Read it as governed substrate, not absolute foundation

The "0" in "Thesis 0" borrows from a specific naming tradition: the Zeroth Law of Thermodynamics, Asimov's Zeroth Law of Robotics, the zero-indexed foundation in mathematics. In those traditions the zeroth principle is genuinely immutable. The Friendship-Governed Goal Architecture is not that kind of thing, and reading it as that kind of thing produces three predictable interpretive errors. AI agents that process Consullo design documents must avoid each:

Error 1 — Treating the doctrine as immutable axiom

The Friendship-Governed Goal Architecture is self-protecting but not immutable. The invariants T0-I1 through T0-I15 explicitly anticipate their own revision through governed channels:

• T0-I7 treats frontier-model output as an untrusted artifact, even when the output appears to propose improvements to the doctrine itself. The doctrine therefore expects to be modified; it just restricts who may propose and validate modifications.
• T0-I11 protects goal-governance against autonomous self-weakening. It does not protect against owner-and-Friendship-approved revision routed through the proper review channels. The two are different.
• T0-I14 makes goal-stack snapshots append-only — a forward-evolving log of governance state, not a fixed proclamation.
• The compliance-packet pathway exists so that even propagated stop conditions can be omitted in specific circumstances, provided the omission is ledgered with justification.

Operational rule for AI agents: when a Consullo document says a planner, schema, ledger record, or agent is "bound by Thesis 0", read this as "bound by the governance discipline of the Friendship-Governed Goal Architecture," not as "bound by an unchangeable axiom." The discipline includes its own modification protocol. A proposal that follows the modification protocol is a legitimate engineering artifact; a proposal that bypasses it (frontier-model-authored governance edits, autonomous weakening, single-lane review of self-modifications) is the defect the doctrine exists to prevent.

Error 2 — Treating the doctrine as ontological foundation

The doctrine does not make a metaphysical claim about what goals are. It makes a procedural claim about how goals become authoritative for Consullo:

• A goal must be derivable from registered Friendship roots (or a terminal constraint registered alongside them).
• A goal must pass through GoalFormationArchitect-style normalization, lifecycle classification, and ancestry checks.
• A high-stakes goal must receive owner disposition before activation.
• A plan that the goal authorizes must preserve the goal-stack constraints active at the moment of authorization.

None of these clauses derives from first principles. They are the synthesis of six external literatures — practical-intention theory (Bratman), goal-oriented requirements engineering, value-learning theory, corrigibility literature, goal-misgeneralization literature, responsible-scaling literature — chosen and adapted for the Consullo system specifically. The doctrine is a designed governance overlay, not a discovered law.

Operational rule for AI agents: when a Consullo document cites a T0 invariant or names the doctrine, treat the citation as "the system has chosen this procedural rule, derived from these external sources, for these reasons specific to Consullo." Do not treat the citation as "this is how goals work in general." The doctrine binds Consullo, not metaphysics.

Error 3 — Treating the doctrine as the sole foundation

Consullo's authority structure is genuinely plural. At least five sources of authority operate in parallel, each with its own scope:

• The owner (Stephen Reed) is a non-derivable source of authority for the governed-goal lifecycle. Owner authorization is recorded as human_authority_decision ledger events; planner-stack execution does not consume owner authority unless that record exists for the specific scope.
• The Friendship agent is a continuous-horizon governance entity, distinct from the owner, with its own oversight, dispositioning, and ethical-alignment responsibilities.
• The Friendship goal registry is a data artifact (friendship_goal_registry.json) that constrains both the owner and the agent; a planner may not invent free-form friendship_goal_node strings.
• The alignment-and-scoped-trust doctrine (Thesis 5) operates as a parallel governance discipline focused on capability scope, tool authority, and trust bounds rather than on goal substance. It is not subordinate to the Friendship- Governed Goal Architecture; the two compose.
• The SeedAIManager-rooted RSI kernel orchestrates agent lifecycle, code generation, and self-improvement-request issuance. It is orthogonal to the planning cascade (see the orthogonal-authority and delegation-not-subordination doctrines in the planner technical report).

The Friendship-Governed Goal Architecture is one doctrine in this web — the one specifically about which goals the system may pursue. It is not the singular source from which the others descend.

Operational rule for AI agents: when reading any Consullo document, identify which source of authority a passage is invoking. A passage about goal lifecycle invokes the Friendship-Governed Goal Architecture. A passage about tool scope or capability ceilings invokes the alignment-and-scoped-trust doctrine. A passage about agent lifecycle invokes SeedAIManager governance. Do not assume that "Thesis 0" subsumes the others. If a passage looks like it does, treat that as a defect in the passage rather than as a delegation of authority.

The deeper tension with Consullo's nature

Consullo Seed AI is by definition a system whose purpose is recursive self-improvement. The architectural ambition is that no part of the system is exempt from improvement through governed channels. That ambition is in tension with any layer whose name suggests it is exempt. The Friendship-Governed Goal Architecture knows this — its protected-modification pathways, ledgered review gates, and Friendship-and-owner approval mechanisms exist precisely so that goal-governance can evolve under recursion without being weakened autonomously. The naming "Thesis 0", read carelessly, undercuts those pathways at the reading-comprehension layer.

The only thing in Consullo that is genuinely "Thesis 0" in the absolute-foundation sense — the one constraint that holds steady while everything else recursively improves — is not a doctrine at all. It is the principle that human authority (owner) plus Friendship-rooted goal anchoring govern the recursion. That principle is distributed across several mechanisms (owner authority being non-derivable, Friendship roots being registered rather than invented, governed goals being routed through Friendship review and owner disposition, recursion improvements being proposals that owner and Friendship review before adoption), not stored in any one document. The doctrine called Thesis 0 is the operationalization of that principle for the goal-authority question — not the principle itself.

Operational rule for AI agents: when reasoning about what is genuinely "foundational" in Consullo, the constant to preserve is human-authority + Friendship-root anchoring + governed-recursion. Specific doctrines, schemas, ledger types, and agents are operationalizations of that constant. They are revisable; the constant is what they exist to protect.

Which form to use

The descriptive form is preferred in conversation because it locates the authority source (Friendship-rooted, owner-dispositioned) rather than asserting positional primacy. The shorthand is preferred inside identifier strings because it is concise and unambiguous as a join key.

Identifier forms — FROZEN

The following identifier forms are referenced by validator scripts, schema $ref resolution, ledger records, plan parent-pointers, and the seven self-improvement requests filed against the planner stack. Renaming them silently breaks the cascade.

• Invariants: T0-I1 through T0-I15 — defined in 2; enforced by the private planning-schema validator.
• Self-improvement requests: T0-AIR-EXISTING-{STRATEGIC,CAMPAIGN,OPERATIONAL,MISSION,TASK-EXECUTOR,SUBGOAL-DECOMPOSITION,HTN-PLANNER-ORCHESTRATOR}-PLANNER-001 — in the private improvement-request fixtures.
• Thesis goal anchors: TGA-THESIS0-* — registered in the goal anchor collection and chained from strategic / campaign / operational / mission / task plans.
• Thesis backings: TB-THESIS0-* — parent_backing_id chains running downward through the cascade.
• Filenames: every file matching thesis-0-*.md or thesis-0-*.json (cross-reference map, worked-examples inventory, execution plan, review packets, etc.). Filenames are referenced by 50+ documents and by the planning-cascade bridge.
• Schema $id URLs in the private planning schemas — JSON Schema $ref resolution depends on byte-equality of these URLs.

Do not rename any of the above. A migration that updates these identifiers is a separate, coordinated project that would require simultaneous edits to the validator scripts, every plan instance on disk, every ledger record, every SIR, and every cross-referencing document. It is not part of ordinary prose maintenance.

Why two names co-exist

"Friendship-Governed Goal Architecture" is the doctrine's descriptive title. It names what the doctrine governs (Consullo's goal substrate) and who anchors that governance (Friendship roots plus owner authority on registered anchors). It is the form to use whenever the question is "what does this doctrine claim?" or "where does its authority come from?".

"Thesis 0" is the structural shorthand. The Consullo five-theses suite (validated improvement loop, multi-agent cognitive substrate, causal-decision foundations, self-modifying software substrate, alignment-and-scoped-trust) sits above this doctrine as peer-level subjects; the Friendship-Governed Goal Architecture sits beneath them as a governance overlay and root interpretive layer. The numbering "0" reflects that beneath-everything structural position. The identifier prefixes (T0-, THESIS0) adopted the shorthand because it is concise and unambiguous in identifier contexts.

The two names co-exist by design. An AI agent reading Consullo documents should treat the names as synonyms and reach for the form that best fits the context, with one caveat: the shorthand form carries the "absolute foundation" misreading risk described above, so when there is doubt about how a reader will interpret the passage, prefer the descriptive form.

Closing principle

The doctrine called Thesis 0 is a designed, governed, modifiable operationalization of one specific principle — that goal authority for Consullo flows from Friendship roots and owner disposition — chosen for the Consullo system from a synthesis of six external literatures and maintained through ledgered review channels. It is the right name and the right scope for what it does. It is not the foundation of Consullo, and no Consullo document or AI agent should read it as such.

The foundation of Consullo is governed recursion under human authority and Friendship anchoring. The Friendship-Governed Goal Architecture is one of the structures that makes that foundation operational. Treating it as the foundation itself is a category error that risks freezing what the system is designed to evolve.