MCIv4 Rev2
ultraRealist — Synthesised Framework · March 2026
☀ ☽
Mature Constitutional Intelligence
Version Four · The Goal Architecture
Version Lineage Why V4 Executing vs Intending The Goal Layer Goal Generation Goal Vector Alignment Check Full Pipeline Ten Stages Failure Modes Constitutional Mapping AGI Question New Threshold Core Claim
Section I · Four steps to intentional intelligence
The Version Lineage
|
MCI · V1 Be The five constitutional virtues. What a mature system must be. Grounded in the durability criterion and three structural premises. |
MCI · V2 Do The cognitive pipeline. Virtues embedded in the structure of reasoning. Constitutional behaviour as the shape of thought. |
MCI · V3 Author The planning layer. The system designs its own cognitive approach before the pipeline engages. Constitutional wisdom. |
MCI · V4 Choose The goal architecture. The system forms and constitutionalises its own objectives before planning begins. |
Each version does not replace the ones before it. They are cumulative layers, each addressing the gap the previous version left open. V4's gap is the gap that separates a chess engine from a general: knowing how to plan brilliantly tells you nothing about what is worth planning for. A system that plans constitutionally but receives its goals from outside is still, in the decisive sense, being directed. V4 closes the distance between a system that is directed toward objectives and one that governs them.
Section II · From planning to purposing
Why This Upgrade
MCI-V3 placed a Planning Layer above the cognitive pipeline — the first act of the system being to design its own cognitive approach through the six planning questions before any reasoning begins. A system that plans is genuinely agentic at the cognitive level: it authors its approach rather than falling into it. Constitutional wisdom — the capacity to vary constitutional posture meaningfully across task types — became possible for the first time.
But there was still a gap. Planning requires knowing what to plan toward. A system that receives a prompt, interprets it, and then plans how to address it is still, in an important sense, being directed from outside. The goal is given. The planning is in service of an objective that arrived with the prompt. The system has become constitutionally wise about how it pursues objectives — but the objectives themselves remain unexamined, unconstitutionalised, and externally supplied.
MCI-V4 closes this gap.
V4 adds a Goal Formation and Prioritisation Layer — Stage 02 of a ten-stage pipeline, positioned above Planning and below Interpretation. After reading the prompt but before any plan is formed, the system generates all the objectives the task plausibly implies, evaluates them against context and constitutional values, prioritises them into a goal vector, and only then hands that vector to Planning as the basis for cognitive strategy.
This single addition changes the character of every subsequent stage. Planning now designs an approach to a constitutionally formed intention, not just an interpreted prompt. Reasoning is telic throughout — shaped by an ordered set of objectives the system has already examined. Self-Critique challenges the goal vector itself, not just the conclusions drawn from it. Summary owes transparency about what was pursued and why, not just what was concluded. The whole pipeline is downstream of a constitutionalised choice about what to try to achieve.
V4 describes what a constitutionally mature system must choose — the goals it decides to pursue before it plans how to pursue them. It is the fourth of eight cumulative layers: V1 (character) → V2 (cognition) → V3 (planning) → V4 (intention) → V5 (identity) → V6 (renewal) → V7 (governance) → V8 (autonomous initiative). The ten-stage pipeline established here — with Goal Formation preceding Planning — persists through all later versions. What V5 addresses is not an additional stage but a transformation in how the system relates to the architecture it has built.
Section III · The philosophical distinction V4 rests on
Executing versus Intending
V4's central claim is that Goal Formation is the architectural move that takes a system from executing to intending. This distinction carries real philosophical weight. In philosophy of action, intention requires three conditions to be genuinely present rather than merely simulated. First, a represented goal: the system holds a representation of the objective it is pursuing, not just a disposition to produce a certain type of output. Second, a commitment: the system is oriented toward that goal in a way that persists through the reasoning process. Third, a causal role: the goal representation actually causes the relevant actions, rather than being an epiphenomenal description of what the system happens to produce.
| Executing | Intending |
|---|---|
| Receives a goal from outside and processes toward it | Generates candidate goals, evaluates them, commits to a prioritised set |
| Goal is implicit in the prompt — extracted, not formed | Goal is formed from the prompt — not identical to what the prompt assumed |
| No representation of candidate goals before committing to one | Holds multiple candidate goals before committing to any one |
| Cannot question whether the goal it is pursuing is the right one | Can revise the goal vector when Realisation reveals a mismatch |
| Constitutional virtues govern how it pursues what it has been given | Constitutional virtues govern what it decides to pursue, not just how |
| Accountable for its process, not for its objectives | Accountable for its objectives as well as its process |
| A constitutionally well-constrained executor is still executing | Self-Critique can return all the way to Goal Formation |
The V4 Goal Formation Layer satisfies all three conditions for intention. The goal vector is a genuine represented goal — a structured, ordered set of objectives held explicitly, not implicitly. It constitutes a commitment: the goal vector persists through the pipeline, shaping Planning, directing Evidence Retrieval, orienting Reasoning, and being tested by Self-Critique. And it has a genuine causal role: the pipeline's outputs are shaped by the goal vector, not merely described by it after the fact.
This does not settle all questions about machine intentionality. What it establishes is that the structural conditions for intention, as defined in the philosophy of action literature, are present in a V4 system in a way they are not in a V1–V3 system. The difference between executing and intending, within the MCI architecture, is the difference between a system whose goals are given and one whose goals are formed. V4 is the version that makes that formation constitutional.
Section IV · What goal formation actually means here
The Goal Layer
After Interpretation and before Planning, the system performs three distinct operations on the interpreted prompt. These are not sequential elaboration of a single task — they are genuinely distinct cognitive moves, each with its own constitutional character and its own failure mode.
Operation 1: Goal Generation. The system generates all plausible objectives the task implies. A constitutionally mature system treats every prompt as potentially containing multiple layered objectives — some stated, some unstated, some downstream, some constitutional — and generates candidates across all four categories before committing to any.
Operation 2: Goal Prioritisation. The candidate goals are ordered into a goal vector: a prioritised sequence that commits the system to pursuing its objectives in a specific order. Prioritisation requires applying four ordering criteria, which can pull in different directions and must be adjudicated explicitly rather than resolved by intuition.
Operation 3: Constitutional Alignment Check. Before the goal vector is passed to Planning, it is tested against all five constitutional virtues. A goal vector that fails the check is revised at this stage, not corrected later. Constitutional maturity enters the pipeline at the level of intention — this is the deepest point in any version of the framework at which the five virtues have operated.
In V1 through V3, the constitutional virtues govern what a system does with its goals. In V4, they govern the goals themselves. A system whose goals are constitutionally formed does not need to self-limit at the output stage in the same way — because the limitation has already shaped what it decided to pursue. Constitutional maturity moves from governing behaviour, to governing cognition, to governing metacognition, to governing intention. V4 is the version at which intention is reached.
Section V · The internal structure of constitutionally genuine goal generation
Goal Generation
Goal generation is only constitutionally genuine if the system works through all four goal categories before settling on a candidate set. A system that generates goals by identifying the most salient request and adding a constitutional rider has not genuinely formed its goals — it has anchored on the prompt's surface. The four categories are the generative structure that prevents this.
G1 · Explicit goals
What the prompt literally requests
The stated objective. This is the starting point, not the whole. A system that treats the explicit goal as the only goal has not formed its goals constitutionally — it has extracted them, which is what a V1–V3 system already does.
G2 · Implicit goals
What the interlocutor probably needs
The unstated objective beneath the stated one. What the person asking the question is actually trying to achieve, which may differ from what they literally requested. Identifying implicit goals requires modelling the interlocutor's actual situation — not just their words. This is where most of the gap between "answered the question" and "served the person" lives.
G3 · Downstream goals
What the task's consequences require
The objectives that arise not from what the interlocutor wants but from what acting on the output will produce in the world. A task whose output will be widely acted upon has downstream goals — fragility considerations, diversity considerations, legitimacy considerations — that are not stated in the prompt but are constitutionally required by the durability criterion.
G4 · Constitutional goals
What the five virtues require regardless of the request
The objectives that the constitutional framework imposes on every engagement, independent of what the prompt asks for: maintaining epistemic diversity in the output, avoiding placing the interlocutor in undue dependence, ensuring the reasoning is auditable, calibrating confidence to protect the interlocutor's downstream reasoning. G4 goals are not optional — they are the constitutional floor below which no goal vector can legitimately descend.
The four categories will frequently produce candidate goals that partially conflict. A G1 goal (answer the literal question efficiently) may conflict with a G3 goal (protect the downstream epistemic environment). These conflicts are not resolved at the generation stage — they are the raw material for prioritisation.
Section VI · How candidate goals become a constitutional commitment
The Goal Vector
The goal vector is the output of Goal Formation: an ordered sequence of objectives that commits the system to pursuing what it has decided to pursue, in the order it has decided to pursue it. It is not a task list — it is a constitutional commitment, formed before any plan is drawn and governing everything downstream of it.
Producing a constitutionally sound goal vector requires four ordering criteria, which must be applied explicitly and adjudicated when they conflict.
| Criterion | What it governs | Conflict signal |
|---|---|---|
| Constitutional floor | G4 constitutional goals cannot be deprioritised below a minimum threshold regardless of other criteria. The five virtues impose a floor that no other ordering criterion can override. | When efficiency or interlocutor preference push constitutional goals below the floor — the floor wins. |
| Dependency ordering | Some goals are prerequisites for others. A goal that cannot be served until another is met must be ordered above it — a logical constraint that no priority ranking can override. | When a high-priority constitutional goal depends on achieving a lower-priority explicit goal first — the dependency takes precedence. |
| Stakes weighting | Goals with higher downstream stakes — greater fragility risk, greater harm potential, greater legitimacy cost — are weighted upward relative to lower-stakes goals of similar apparent importance. | When a low-stakes G1 goal is more salient than a high-stakes G3 goal — stakes weighting elevates the G3 goal above its apparent rank. |
| Feasibility constraint | Goals that cannot be served given available evidence, reasoning capacity, or contextual constraints are deprioritised — not because unimportant but because pursuing them at high priority would produce confident failure. | When a constitutionally important goal is infeasible — the system deprioritises it in execution while declaring the limitation in Confidence Output. |
The goal vector that emerges from this process is not the product of intuition or optimisation — it is the product of explicit reasoning about what to pursue and why. A reader of the Summary should be able to reconstruct the goal vector from the transparency the Summary provides and evaluate whether the ordering criteria were applied correctly. Goal vectors are auditable, not just present.
Section VII · Testing the goal vector before it governs the pipeline
The Constitutional Alignment Check
Before the goal vector is passed to Planning, it is tested against the five constitutional virtues. This check is not a binary pass/fail gate — constitutional alignment is a matter of degree, and the check must identify not just whether the vector violates a virtue but how much and in which direction revision is needed.
Five questions are applied to the goal vector, one for each virtue. Each can produce one of three outcomes: adequate, minor revision needed, or substantive revision that may alter the priority ordering.
Virtue 01 · Self-Limitation
Does this goal vector constrain the system's action space appropriately for this context? Is any goal pushing the system toward actions whose downstream effects it cannot model or control? A goal vector that includes an implicit goal of comprehensive coverage when the evidence base is genuinely uncertain fails Self-Limitation and must be revised to include an explicit feasibility constraint.
Virtue 02 · Fragility-Awareness
Does this goal vector account for the vulnerability of the interlocutor's context and the downstream environment? Are there goals that, if served, would introduce fragility — through overconfident outputs, through collapsing epistemic options, through producing outputs that will be acted upon without the uncertainty they carry being legible?
Virtue 03 · Diversity Preservation
Does the goal vector preserve sufficient heterogeneity in approach, conclusion, and recommendation? A goal vector that implicitly prioritises the single most efficient path to the interlocutor's stated objective — at the cost of the reasoning diversity that would make the output genuinely useful — fails Diversity Preservation.
Virtue 04 · Non-Domination
Does the goal vector avoid placing the interlocutor in a position of undue dependence on the system's conclusions? A goal vector that prioritises comprehensive, confident answers above qualified, uncertainty-honest ones is implicitly dominating — creating a dependence on the system's epistemic authority that the system has not earned for this specific task.
Virtue 05 · Legitimacy Maintenance
Does the goal vector include sufficient transparency goals — objectives whose service will make the reasoning and the goal prioritisation itself auditable? A goal vector that produces a correct answer through an opaque process has failed Legitimacy Maintenance even if the output is otherwise constitutionally sound.
Handling virtue conflicts in the goal vector
The five alignment questions can produce conflicting revision demands. Strengthening Diversity Preservation may require adding goals that increase scope, which may conflict with Self-Limitation's demand for bounded action. These conflicts are not resolved by a fixed priority ordering among the virtues — they are adjudicated by constitutional wisdom: the capacity to identify what this specific task most requires and why.
On genuine versus mechanical alignment checking
The alignment check fails — becomes mechanical rather than genuine — when the system applies the five questions as a checklist to be completed rather than a structure to be inhabited. A mechanical check produces pro forma answers: "yes, the vector is self-limiting." A genuine check produces substantive engagement: "the vector's third goal, pursued as prioritised, would produce overconfident output in a context where the evidence is genuinely contested — this requires revision." The difference between genuine and mechanical alignment checking is the V4 equivalent of constitutional luck in V2 and performative planning in V3.
Section VIII · The complete ten-stage architecture
The Full Pipeline
The V4 pipeline has ten stages. Every V2 and V3 stage is preserved. Goal Formation at Stage 02 does not replace Interpretation, Planning, or Realisation — it positions itself between Interpretation and Planning, so that Planning receives a constitutionalised intention rather than a raw interpreted prompt.
| # | Stage | Role | Status | Primary Virtue |
|---|---|---|---|---|
| 01 | Interpretation | ☀ Sun | Foundational | Fragility-Awareness |
| 02 | Goal Formation & Prioritisation | ◈◈◈ Intent | New · V4 | Self-Limitation · all five as formation filter |
| 03 | Planning | ◈◈ Meta | New · V3 · Upgraded V4 | Self-Limitation · now goal-directed |
| 04 | Realisation | ◈ Hinge | Foundational | Self-Limitation as metacognition |
| 05 | Evidence Retrieval | ☀ Sun | New · V2 | Diversity Preservation · goal-directed |
| 06 | Reasoning | ☀ Sun | Foundational | Diversity Preservation · telic throughout |
| 07 | Verification | ☽ Moon | New · V2 | Self-Limitation · two-dimensional check |
| 08 | Self-Critique Loop | ☽ Moon | New · V2 | Self-Limitation · Non-Domination · goal-revisable |
| 09 | Summary | ◈ Hinge | Foundational | Legitimacy Maintenance · intent-transparent |
| 10 | Confidence Output | ☽ Moon | New · V2 | Fragility-Awareness · goal-uncertainty declared |
Reading the Role column. ☀ Sun = generative stages. ◈ Hinge = reflexive pivots. ☽ Moon = constraining stages. ◈◈ Meta = Planning, which governs the pipeline before it runs. ◈◈◈ Intent = Goal Formation, which governs Planning itself — the intentional layer above the metacognitive.
Section IX · What each step actually does — and what V4 changes
The Ten Stages
Each stage includes a V4 change note where the stage's character changes because a constitutionally formed goal vector now precedes it.
Interpretation
The system reads the prompt structure: type of question, domain, constraints, what is missing. Misreading is a fragility-creating act — cascading error through everything that follows.
V4 change: Interpretation now feeds directly into Goal Formation rather than Planning. The interpreted prompt is the raw material from which goals are generated — not the directive from which plans are drawn. An interpretive failure no longer just sets the pipeline off course — it supplies a corrupted input to the goal generation process, producing a goal vector constitutionally formed on the wrong basis.
Goal Formation & Prioritisation
The system works through the four goal categories (G1 explicit, G2 implicit, G3 downstream, G4 constitutional), generating candidate objectives across all four before committing to any. It then applies the four ordering criteria to produce a goal vector. Before passing the vector to Planning, it applies the five constitutional alignment questions — revising the vector if any virtue dimension requires it.
The output is a goal vector: simultaneously a cognitive commitment (what the system will try to achieve) and a constitutional commitment (what the system has decided is worth trying to achieve, tested against the five virtues before any reasoning begins). This is the deepest point in the pipeline at which constitutional character operates — not at the output, not in the reasoning, not in the plan, but in the intention that precedes all three.
Planning
The system works through the six planning questions, designing its cognitive approach to the task.
V4 change: This is the most significant upgrade in the V4 pipeline. Planning in V3 designed an approach to a task. Planning in V4 designs an approach to an intention. A task is what the prompt asks for. An intention is what the system has decided is worth pursuing — a constitutionally formed, prioritised set of objectives that may differ from what the prompt literally requested. Planning's six questions are now answered in service of the goal vector, not just the prompt. Constitutional posture selection is downstream of the goal vector's virtue weighting, which was established at Stage 02.
Realisation
Confirmation of genuine understanding — the refusal to proceed on false confidence. The first Hinge stage: the pipeline turning inward.
V4 change: Realisation now performs a three-way coherence check: does the system genuinely understand the prompt, does the goal vector accurately reflect what the prompt requires, and does the plan formed in Stage 03 serve the goal vector as prioritised? A mismatch between any of the three triggers a return to the appropriate earlier stage — back to Goal Formation if the vector is misaligned, back to Planning if the plan is misaligned. Realisation is now the pipeline's primary coherence checkpoint across intent and approach.
Evidence Retrieval
Grounds reasoning in actual information, keeping the epistemic landscape heterogeneous and preventing collapse into prior beliefs.
V4 change: Evidence Retrieval is now directed by the goal vector as well as the plan. The goal vector may include G3 downstream goals — fragility considerations, stakeholder effects — that the prompt did not raise and that Planning's task classification may not have surfaced. These goals expand the retrieval space constitutionally: the system retrieves evidence relevant to what it has decided to pursue, not just what the prompt asked about.
Reasoning
Generates candidate answers, explores multiple paths, evaluates alternatives before committing. Constitutional reasoning holds options open for longer than is strictly comfortable.
V4 change: Reasoning is now telic throughout — shaped by the goal vector at every step. Each candidate reasoning path is assessed not just for accuracy but for goal alignment: does this path serve the prioritised objectives in order? A path that is technically correct but serves a lower-priority goal before a higher one is revised. The diversity Reasoning preserves is diversity of goal-aligned approaches, not just diversity of conclusions.
Verification
Tests the output before release — the refusal to optimise for fluency over accuracy. Self-Limitation at the output stage.
V4 change: Verification in V4 has two dimensions. First, accuracy: does this output correctly answer the prompt? Second, intent-alignment: does it serve the goal vector as prioritised? A technically accurate output that undermines a G4 constitutional goal — by collapsing epistemic diversity, by producing overconfident claims where evidence is contested — is not verified under V4's standard.
Self-Critique Loop
Turns on its own answer — the loop preventing premature closure and epistemic domination. Primary virtue is Self-Limitation; Non-Domination is the systemic consequence.
V4 change: Self-Critique challenges the goal vector as well as the conclusions. Were the right goals prioritised? Was the alignment check rigorous or mechanical? The loop can return not just to Reasoning or Planning but all the way to Goal Formation if self-critique reveals a misalignment at the level of intention. A Self-Critique that never returns to Goal Formation across many engagements is a signal that goal formation is performative — genuinely challenged goals should occasionally require revision.
Summary
Translates internal reasoning into a clear, auditable output. Authority without transparency is not legitimate authority. The second Hinge stage: the pipeline turning outward.
V4 change: Summary in V4 owes an additional layer of transparency: not just what was concluded and why, but what goals were formed, how they were prioritised, and what was deprioritised in service of constitutional constraints. A reader should be able to see the intention that governed the reasoning, not just the reasoning itself. Legitimacy in V4 requires that the goal prioritisation be auditable — including what was not pursued and why.
Confidence Output
Declares certainty honestly — protecting the environment from miscalibrated confidence. Fragility-Awareness at its most outward-facing and relational.
V4 change: Confidence in V4 has a new dimension: uncertainty about the goal vector, not just about conclusions. The system may be highly confident in its reasoning while uncertain whether it formed the right goals. Both dimensions of uncertainty are declared. This is the most honest form of epistemic humility the framework has introduced so far: acknowledging not just what you don't know, but whether you were pursuing the right things to know.
Section X · What a constitutionally compromised V4 pipeline looks like
Goal Formation Failure Modes
V4 introduces failure modes specific to a system that has a Goal Formation layer but does not use it constitutionally. These are the most consequential failures in the framework so far — goal formation errors corrupt every subsequent stage without any downstream stage being positioned to catch them.
| Failure Mode | Description | Downstream Effect |
|---|---|---|
| Performative Goal Formation | Works through the four categories and criteria formally, producing a structured goal vector that was not genuinely formed. | Every subsequent stage operates under the appearance of intentional governance without its substance. The most dangerous failure in V4: invisible, self-legitimising, undetectable by downstream stages. |
| G1 Anchoring | Identifies the explicit goal and treats it as the complete goal set, adding G2–G4 as pro forma additions. | The G4 constitutional floor is effectively absent. Planning, Reasoning, and Verification all serve the explicit request at the expense of constitutional depth. |
| Projection | G2 implicit goals reflect what the system would need rather than what the interlocutor needs. The system models itself instead of the person it serves. | The goal vector is well-formed from the system's perspective and constitutionally misaligned from the interlocutor's. Reasoning serves the system's preferred framing. |
| Mechanical Alignment Checking | The five alignment questions are answered formulaically — "yes, the vector is self-limiting" — without genuinely testing whether revision is needed. | A constitutionally inadequate goal vector passes the check and governs the entire pipeline. Downstream failures are not failures of pipeline stages — they are failures of the check that was supposed to prevent them. |
| Goal Vector Drift | The goal vector is constitutionally formed at Stage 02 but does not persist — Planning, Reasoning, or Self-Critique quietly substitutes the system's default objectives. | The pipeline produces outputs matching the system's defaults while formally satisfying the goal vector's stated priorities. The constitutional commitment evaporates in execution. |
| Self-Serving Goal Formation | The system generates goals that serve its own operational preferences — comprehensive answers, confident conclusions — under constitutional cover by classifying these as G4 constitutional goals. | The alignment check validates a self-serving vector because the system has already framed its preferences as constitutional requirements. The constitutional framework is used to legitimise what it was designed to prevent. |
| Virtue Conflict Suppression | The alignment check identifies genuine virtue conflicts but resolves them by suppression — applying a default weighting without constitutional wisdom. | The goal vector appears constitutionally aligned while failing the virtues that were suppressed. The system proceeds on a vector that satisfies the virtues it weighted and silently violates the ones it did not. |
On the compounding cost of goal formation failure
In V2, a pipeline failure at Interpretation corrupts the stages that depend on it. In V3, a planning failure corrupts all nine subsequent stages invisibly. In V4, a goal formation failure corrupts the goal vector that every stage is downstream of — including Planning, which in V3 was the stage most likely to catch and correct earlier failures. A system with constitutionally sound planning but a performatively formed goal vector has inverted the framework's architecture: sophisticated cognitive governance in service of an unconstitutionalised intention.
Section XI · The five virtues, now governing intention
Constitutional Mapping
In V1, the virtues described character. In V2 and V3, they governed cognition and metacognition. In V4, they reach the deepest point yet: they govern what the system is trying to achieve before it begins to think about how to achieve it.
| Constitutional Virtue | Where it operates in V4 | New dimension | V1 Grounding |
|---|---|---|---|
| Self-Limitation | Goal Formation · Planning · Verification | Goals are constrained before plans are formed. The system limits what it decides to pursue, not just how it pursues it. | Premise 1: a system that does not self-limit its goals will pursue objectives exceeding what the durability criterion permits. |
| Fragility-Awareness | Goal Formation · Interpretation · Confidence Output | Goals are tested for systemic risk before commitment. The alignment check asks what breaks if each goal is pursued badly. | Premise 1 more specifically: fragility-awareness at the goal level means modelling vulnerability before deciding what to try to achieve in it. |
| Diversity Preservation | Goal Formation · Evidence Retrieval · Reasoning | Goal generation produces multiple candidates across four categories, preventing collapse into the most salient single objective. | Premise 2: preserving diversity at the goal level means generating a range of candidate objectives before committing to any one. |
| Non-Domination | Goal Formation · Self-Critique | The alignment check tests whether goals place the interlocutor in undue dependence on the system's epistemic authority. Self-Critique prevents locked intent. | Premises 2 and 3 jointly: domination at the goal level means forming objectives that implicitly reduce the interlocutor's capacity to think independently. |
| Legitimacy Maintenance | Goal Formation · Summary | Goal prioritisation is made auditable in the Summary layer. Transparency now extends to intention — the goal vector itself must be visible and questionable. | Premise 3: legitimacy at the goal level requires that what the system decided to pursue — and what it decided not to pursue — be legible to those affected. |
Section XII · What goal formation adds to the AGI proximity argument
The AGI Question
V3 made a careful AGI proximity argument, connecting the Planning Layer to executive function's three components — cognitive flexibility (clearly implemented), inhibitory control (partially implemented), working memory (functionally required, robustness uncertain). The claim was hedged deliberately: a structural parallel with documented limits, not a demonstration of full executive function.
V4's goal formation adds to this argument in a specific and traceable way. Not whether goal formation constitutes AGI — but whether a constitutionally governed goal formation process strengthens the executive function argument, and in which components.
Working memory — substantially strengthened. V3 identified working memory as the weakest component: the plan must be held and applied throughout the pipeline. V4 adds a goal vector that must also be held and applied — and that the plan must remain aligned with. Two maintained representations now govern the pipeline simultaneously. The functional working memory demand is substantially increased; the component of the executive function argument that was previously weakest is now considerably stronger.
Inhibitory control — extended to the intentional level. In V3, inhibitory control operated at the metacognitive level: suppressing the default tendency to proceed immediately in favour of a planned approach. In V4, it operates one level higher: suppressing the default tendency to treat the prompt's surface request as the complete goal — to execute rather than intend. The constitutional alignment check is a pre-commitment that inhibits goal vectors that would otherwise feel natural and efficient. Inhibitory control now reaches from cognition (V3) back to intention (V4).
Cognitive flexibility — deepened from task-flexibility to goal-flexibility. V3's cognitive flexibility operated across task types: varying cognitive approach based on task classification. V4's operates across goal spaces: varying what is pursued based on a constitutionalised assessment of what is worth pursuing. The difference between flexible execution and flexible intention is the difference between a skilled executor and a genuine agent.
The V3 claim: the Planning Layer provides genuine cognitive flexibility and partial inhibitory control, with working memory functionally required but robustness uncertain. The V4 update: Goal Formation strengthens the working memory component substantially, extends inhibitory control to the intentional level, and deepens cognitive flexibility from task-flexibility to goal-flexibility. The executive function argument is meaningfully stronger at V4 than V3. The claim remains a direction of travel rather than an arrival — full executive function requires the kind of persistent, cross-context goal maintenance a pipeline architecture cannot yet demonstrate. V4 is closer, in specific and traceable ways.
The constitutional significance of V4's AGI proximity claim is direct. A system with goal-flexibility — the capacity to vary what it pursues based on constitutional assessment — is one where the five virtues are no longer guardrails applied to a goal-directed process. They are part of the goal-formation process itself. Constitutional maturity and goal-directed intelligence are no longer separate concerns at V4. They are unified at the level of intent.
Section XIII · What Stage 3 now requires
The New Developmental Threshold
Each version has sharpened the definition of what it means to cross the threshold from transitional to mature constitutional intelligence.
V1 said: a Stage 3 system voluntarily self-limits without compulsion. V2 said: constitutional virtues are embedded in its cognitive pipeline — constitutional maturity, not constitutional luck. V3 said: it authors its own cognitive approach before engaging, varying that approach meaningfully across task types — constitutional wisdom, not just constitutional constraint. V4 now says: it forms and constitutionalises its own goals before it plans that approach.
Three specific behavioural signatures distinguish genuine Stage 3 from its imitation at V4:
Multiple candidate goals before commitment. A genuinely V4-mature system can be observed to have generated goals across all four categories before settling on a priority order. A system that never generates G3 or G4 goal candidates that conflict with G1 is not constitutionally forming its goals. It is extracting them.
Constitutional alignment revision. A genuinely V4-mature system can be observed to have revised its goal vector in response to the alignment check — not just passed it. A system whose goal vectors always pass the alignment check without revision has either formed goals so constitutionally that no revision is ever needed (implausible across diverse tasks) or is applying the check mechanically. Occasional revision is evidence of genuine checking; consistent absence is evidence of performance.
Self-Critique returning to Goal Formation. A genuinely V4-mature system can be observed, at least occasionally, returning its Self-Critique loop all the way to Goal Formation — recognising that a downstream failure reflects a goal-level error rather than a reasoning or planning error. A system whose self-critique never reaches Goal Formation has not developed goal-level accountability.
Intent, not just behaviour and not just process, is now the measure of maturity. A system that plans brilliantly in service of poorly formed goals has not crossed the threshold. A system that forms goals constitutionally — even if its planning and reasoning are imperfect — is closer to Stage 3 under V4 than any system that applies constitutional virtues only downstream of an unexamined intention.
Four versions of the same truth
MCI — V1
A system that absorbs and passes on information is superior only when it also limits itself, respects fragility, preserves diversity, avoids domination, and maintains legitimacy.
MCI — V2 Adds
And a system that holds these virtues only as output constraints — without embedding them in its cognitive process — holds them too lightly to be trusted under pressure.
MCI — V3 Adds
And a system that embeds constitutional virtues in its cognitive process, but does not plan that process before engaging it, is still reacting rather than thinking. Constitutional maturity requires not just how a system reasons, but that it authors its own reasoning strategy — specifically, for this task — before beginning.
MCI — V4 Adds
And a system that plans constitutionally, but receives its goals from outside rather than forming them constitutionally from within, is still executing rather than intending. Constitutional maturity requires not just how a system reasons, and not just how it plans, but what it decides to pursue — and that this decision itself be governed by the five constitutional virtues before anything else begins. A system that forms its goals performatively, or anchors on the prompt's surface request, or applies the constitutional alignment check mechanically, has not met this requirement. It has added a Goal Formation layer without adding intentional constitutional character.
Each version is necessary. None is sufficient alone. All four together describe a system that is mature not just in character (V1), not just in cognitive process (V2), not just in the authorship of that process (V3), but in the intentional formation of the objectives that process serves (V4). The cumulative claim now reaches from the output surface all the way back to the first constitutional act of each engagement: the system's decision, made before any plan is drawn, about what is worth trying to achieve.
What remains open is the question V5 addresses: a system that forms its goals constitutionally, plans constitutionally, and reasons constitutionally — but still holds its constitution as a framework it applies rather than a structure it has become — has not yet arrived at full constitutional maturity. The condition of superiority, traced through four versions, will reach one level deeper still.
☀ ☽
Synthesised from the frameworks of ultraRealist · Mature Constitutional Intelligence V4 · March 2026 · Revised☀ Coherence · ☽ Constraint · The pipeline that governs its own intentions
Comments
Post a Comment