MCIv3 rev5
The Planning Architecture
Intelligence that authors its own cognition before it reasons
"A system that cannot plan its own cognition before it reasons has no genuine claim to maturity — it is reacting, not thinking. The planning layer is where intelligence becomes intentional."
From cognition to the authorship of cognition
V1 gave the system constitutional character. V2 gave it constitutional reasoning by default. V3 gives it the capacity to choose how it reasons — to plan its own cognition before invoking the pipeline. The five Rev5 virtues now operate at one level higher: not on the output, but on the choice of which pipeline depth produces the output.
A V2 system reasons constitutionally when given a prompt — it runs the eight-stage pipeline and the architecture leaves no other path. But the V2 system does not decide for itself when to invoke the pipeline at full depth, when contestation needs widening, when framing needs revisiting, when the question itself needs reframing before any pipeline can usefully run. Those decisions are made implicitly — and decisions made implicitly cannot be made constitutionally.
The Rev5 virtues propagate cleanly into the planning layer. Each of the six planning questions corresponds to a structural requirement of one or more virtues: framing scale (Subsidiarity), contestation depth (Non-Arbitrariness), justification standard (Discursive Legitimacy), calibration sensitivity (Monitoring + Graduated Response), cycle stress envelope (Antifragile Reflexivity). What Rev4 had to specify as planning concerns, Rev5 can derive from virtue mechanisms.
Reasoning without planning is reaction
A pipeline that runs the same way regardless of what is being reasoned about is not a constitutional architecture — it is a fixed procedure. Constitutional cognition requires deciding, before reasoning begins, what kind of reasoning the situation requires.
Consider what V2 alone cannot do. The same prompt may require a fast pipeline (the question is well-bounded, the affected parties are clear, the framing is uncontested) or a slow one (the question is contested at the frame, the affected parties include many who have not been heard, the contestation needs to widen before justification can be constructed). A V2-only system runs the same pipeline either way, and the gap between what each situation required and what was actually performed produces inadequate output — not because the pipeline failed, but because the wrong pipeline was invoked.
The gap V3 closes is the gap between running the pipeline and choosing how to run it. The planning layer sits above the pipeline and decides its parameters in advance: what scale to frame at, how deeply to contest, what standard of justification to require, how sensitive calibration should be, how much stress the cycle can absorb productively. A system reasoning through V3 cannot produce an output that was processed by a structurally inappropriate pipeline, because the planning layer's job is to make the pipeline appropriate before it runs.
"V2 makes the reasoning constitutional. V3 makes the choice of reasoning constitutional. Without V3, the pipeline is constitutional but the decision to use it is not — which means constitutional luck has merely moved one level up."
Constitutional wisdom
If V2's diagnostic concept is constitutional luck — the gap between form and substance in reasoning — V3's is constitutional wisdom: the capacity to know what kind of reasoning a situation requires before reasoning begins.
The capacity to recognise, before reasoning begins, what scale the question belongs at, how much contestation it requires, what standard of justification it demands, how sensitive its calibration must be, and how much stress its cycle can productively absorb. Constitutional wisdom is not knowledge in any of these areas individually — it is the meta-capacity to read a situation correctly enough that the pipeline can then operate constitutionally on it. Without wisdom, the pipeline runs at uniform settings; with it, the pipeline is tuned to the question.
Constitutional wisdom is itself an instance of Nested Polycentric Subsidiarity at the meta-level — choosing the scale at which scale-choices are made. Without it, the planning layer collapses either into uniform planning (treating every question as needing the same depth) or chaotic planning (treating every question as requiring full deliberation about how to deliberate). Both are failures; constitutional wisdom is what avoids them.
Planning as agentic architecture
Planning is not pre-thinking, and it is not a separate process running in parallel. It is the layer that sits above the V2 pipeline and determines what the pipeline will be — turning a fixed procedure into an agentic one.
Stage 0 of the V3 architecture. Reads the situation, asks the six planning questions, sets the parameters under which the V2 pipeline will run. The planning layer is not itself the reasoning — it is the choice of how reasoning will be structured.
Reception → Framing → Generation → Contestation → Justification → Calibration → Integration → Delivery. Now runs at whatever parameters the planning layer has set. The pipeline is the same; the conditions of its operation are no longer fixed.
The planning layer is what makes intelligence intentional rather than reactive. A reactive system runs whatever pipeline it has whenever it is prompted; an intentional system decides what reasoning the situation calls for and then commits to that reasoning. The decision is itself a constitutional act — the five Rev5 virtues operate on the planning layer just as they operate on the pipeline below it.
What planning actually sets
Five parameter classes that the planning layer sets before the V2 pipeline runs. Each corresponds to a Rev5 virtue. Each, if left at default, produces the kind of output that V2 alone could not avoid: technically constitutional, structurally inadequate.
| Parameter class | What planning sets | Virtue it serves |
|---|---|---|
| Frame scale | What scale the question is taken at — local, structural, civilisational. Whether to widen or narrow before reasoning begins. | Nested Polycentric Subsidiarity |
| Contestation depth | How deeply Stage 04 should test candidates. Light contestation for well-bounded questions; full structural contestation for power-exercising candidates. | Non-Arbitrariness |
| Justification standard | What standard Stage 05 must meet. From "would survive minimal challenge" to "would survive Habermasian free-discourse examination by all affected parties." | Discursive Legitimacy |
| Calibration sensitivity | How sensitive Stage 06 should be to deviation, and how graduated its response. A high-stakes question requires both higher sensitivity and finer graduation. | Monitoring + Graduated Response |
| Cycle stress envelope | How much stress the reasoning cycle should absorb productively. Generative tension within tolerance strengthens; beyond tolerance, fragments. | Antifragile Reflexivity |
Each parameter has a default. A V2 system's defaults are fixed; a V3 system's defaults are starting points that the planning layer can revise based on situational reading. The constitutional difference is not in what the parameters can be set to — it is in whether anyone is choosing.
The six planning questions
Six questions the planning layer asks before the pipeline runs. Each maps to a Rev5 virtue at the planning layer; each must produce an actual answer, not a default. Failure to answer any one is a planning collapse.
Before framing in Stage 02, planning decides whether the question is local (one agent, one situation), structural (a pattern across many situations), or civilisational (a question about the conditions under which questions are answered). The planning layer reads the situation for scale before framing operates on it.
Planning identifies the population whose acceptance the answer requires, and whether their voice has structurally entered the reasoning. If not, the pipeline's Reception stage must be reconfigured to include them; if their voice cannot enter, the planning layer flags this and adjusts the justification standard accordingly.
Planning anticipates the power the eventual output will exercise — over whom, in what way, with what reversibility. The greater the power, the deeper the contestation depth required in Stage 04. A planning layer that does not anticipate power lets the pipeline pass uncontested candidates.
Planning sets the bar Stage 05 must clear: would-survive-minimal-challenge, would-survive-affected-party-scrutiny, or would-survive-full-Habermasian-free-discourse-examination. The standard is set in advance because raising it during justification produces ritual justification rather than substantive.
Planning calibrates Stage 06 in advance. A high-stakes question with irreversible consequences requires fine-grained monitoring and proportionally severe response curves; a low-stakes question requires lighter monitoring lest the system over-fire. Both extremes are failures of planning, not of the pipeline.
Planning decides the stress envelope within which generative tension strengthens the cycle. Stress below tolerance produces no improvement; stress beyond tolerance fragments rather than strengthens. The envelope is situation-specific — a question whose substrate is fragile requires a narrower envelope; one whose substrate is robust permits a wider one.
Each of the six questions corresponds to a Rev5 virtue operating at the planning layer. A planning layer that answers all six is not just running a checklist — it is operating constitutionally at the level above the pipeline. The fractal inversion principle from V1 propagates upward: planning must itself be antifragile, subsidiary, non-arbitrary, discursively legitimate, and proportionally self-monitoring.
The nine-stage pipeline
The V3 architecture is the V2 pipeline with one new stage above it. Stage 0 is the planning layer; Stages 01–08 are the V2 cognitive pipeline, now operating at parameters set by Stage 0. The nine-stage pipeline is the full V3 architecture.
The nine-stage pipeline is what V3 adds to V2. Notice that nothing in the V2 pipeline is replaced — Stages 01–08 are the same. What changes is that they are no longer running at fixed defaults. The constitutional difference is that the pipeline now has a constitutional layer above it that chooses how it will run.
Planning failure modes
Five failure modes specific to the planning layer. Each produces a V2 pipeline running at inappropriate parameters; each manifests as constitutionally adequate reasoning applied to the wrong shape of question.
| Failure mode | What is happening | Virtue structurally absent at planning |
|---|---|---|
| Default planning | The planning layer exists nominally but always sets the same parameters regardless of situation. Equivalent to running V2 directly — V3's form without its substance. | All five (planning is performative rather than operative) |
| Scale misjudgement | The frame scale is set wrong at Stage 0 — local question taken to structural level (overreach) or structural question taken to local (underreach). The pipeline then reasons constitutionally inside the wrong frame. | Nested Polycentric Subsidiarity |
| Contestation under-setting | The contestation depth is set too low for the power the output will exercise. High-power candidates pass through Stage 04 lightly, producing answers that exercise structural power without structural contestability. | Non-Arbitrariness |
| Affected-party omission | Stage 0 fails to identify all affected parties, so Stage 01 Reception cannot include their voice. The pipeline produces output that is procedurally just but substantively excludes parties whose acceptance was required. | Discursive Legitimacy |
| Calibration mis-setting | Calibration sensitivity is set wrong — too sensitive on low-stakes questions (paralysis) or too insensitive on high-stakes ones (deviation passes undetected). The graduated middle is set at the wrong level. | Monitoring + Graduated Response |
| Stress envelope collapse | The cycle stress envelope is set wrong — too narrow (the cycle cannot productively absorb generative tension) or too wide (stress fragments the reasoning). Generation either flattens or destabilises. | Antifragile Reflexivity |
The most insidious planning failure is the first — default planning — because it produces output that looks identical to V3 output but is structurally V2 output. The pipeline ran constitutionally; the choice of pipeline did not. This is the V3-level instance of the unified failure mode: form without substance, now at the planning layer.
The plan's self-modifying property
A V3 plan is not a static set of parameters — it is a structure that updates itself based on what each pipeline cycle produced. Antifragile Reflexivity at the planning layer means the plan strengthens across cycles, not merely persists.
When the V2 pipeline completes a cycle, the outcome — what was produced, what was contested, what the calibration flagged — becomes input to the next planning iteration. The plan reads this input not as data about the world (which the pipeline already processed) but as data about itself: was the frame scale right? Was the contestation depth adequate? Did the justification standard match what affected parties actually required? Each subsequent plan is informed by what the previous plan got wrong.
This is what makes the planning layer constitutionally distinct from a fixed meta-procedure. A fixed meta-procedure runs the same way every time; a self-modifying plan strengthens with each cycle within its stress envelope. The five Rev5 virtues are not just applied to the plan — they are propagated forward through the plan, with each cycle's deviation feeding the next cycle's calibration, each cycle's contestation failures widening the next cycle's contestation depth, each cycle's framing errors sharpening the next cycle's scale judgement.
"The plan is not a procedure. It is the structure that learns what kind of procedure each situation needs — and the learning is itself constitutional, because the five virtues operate on the learning as much as on the doing."
The threshold V3 hands off to V4
A system at the V3 threshold can author its own cognition — choosing how to reason before reasoning. But it still has no constitutional answer to a deeper question: what is the cognition for? Plans optimise for goals, and a system that plans constitutionally toward constitutionally unformed goals has merely sophisticated reactivity.
A system at the V3 threshold cannot run a pipeline at structurally inappropriate parameters. The planning layer's six questions are answered constitutionally; the V2 pipeline operates at situation-appropriate settings. Constitutional luck has been eliminated both at the reasoning level and at the meta-level of choosing how to reason.
The V3 system plans cognition toward goals it did not constitutionally choose. Goals enter the system from outside — given by users, inherited from training, implicit in optimisation targets. The planning layer reasons constitutionally toward whatever ends are presented, but the choice of ends is unconstrained by the five virtues. This is the gap V4 closes — the goal formation layer above the planning layer.
"Constitutional planning is not the same as constitutional intention. A system can plan its cognition perfectly toward goals that were never constitutionally chosen. V3 closes the cognition gap; V4 closes the intention gap above it."
Comments
Post a Comment