REFERENCE · 7 × 5 × 3

The Maturity Cube, in full

Everything on this page is the framework itself — the axes, all 105 behavioural markers, the naming alphabet, and the laws that hold the faces together. It is the accessible twin of the 3D cube: the drawing and this reference always agree.

§ AXES

Three axes, one address space

D · 7 DIMENSIONS

The seven dimensions replay the lifecycle of one exchange with an assistant — before the request (Truth, Deep), the request itself (Connect, Service), after the response (Knowledge, Evolution), and a cross-cutting validation loop (Responsibility).

TTruth — knowing what can be asked of AI at all
DDeep — framing the task and drawing its boundaries
CConnect — giving the model context, and working with it
SService — phrasing for a specific system, choosing the right one
KKnowledge — working the answer into the needed form
EEvolution — iteration as standard practice, not failure
RResponsibility — validating results; knowing where judgment must stay human

L · 5 LEVELS

A level is a mode of existence of a practice, not a score. Levels are non-skippable: calibration (L4) is impossible without an articulated method (L3), and a method does not appear without accumulated repetition (L2).

1Ad hoc — luck — results are accidental and unreproducible
2Repeatable — habit — personal recipes that once worked, reasons unknown
3Articulated — method — the practice can be explained and handed to another
4Managed — calibrated control — outcomes are predicted, and predictions hold
5Evolving — self-renewal — the practice improves itself and spreads

S · 3 CARRIERS

The carrier axis says where the knowledge lives. The subject of every question substitutes down the axis; the dimension is preserved.

E · Employeeknowledge lives in a person's head“Does the employee understand it?”
P · Processknowledge lives in process artefacts — steps, templates, checkpoints“Does the process define it, regardless of who performs it?”
O · Organisationknowledge lives in institutions — norms, resources, metrics, training“Does the organisation provide and govern it, regardless of the process?”

§ LEVELS

One skeleton, three connotations

The five levels keep their nature on every carrier — what changes is the aggregate state of the practice. The L2→L3 boundary is the articulation threshold, the central line of the whole model.

Level	Nature of practice	E — employee	P — process	O — organisation
L1 Ad hoc	luck — results are accidental and unreproducible	luck	unmanaged	shadow
L2 Repeatable	habit — personal recipes that once worked, reasons unknown	personal recipe	pockets of practice	local initiatives
L3 Articulated	method — the practice can be explained and handed to another	transferable method	built into the process	norm + enablement
L4 Managed	calibrated control — outcomes are predicted, and predictions hold	calibrated control	process is measured	managed portfolio
L5 Evolving	self-renewal — the practice improves itself and spreads	self-renewal	self-optimising	learning organisation

§ THE THREE FACES

105 observable markers — not self-ratings

Every cell answers its face's question with behaviour you could observe from the outside. The accented column is L3 — the articulation threshold — which the three faces climb as a ladder: can explain → written into the process → anchored as a norm.

E — Employee face

“Does the employee understand it?”

Employee face: 7 dimensions by 5 levels. Does the employee understand it?
	Dimension	L1 Ad hocluck	L2 Repeatablepersonal recipe	L3 Articulatedtransferable method	L4 Managedcalibrated control	L5 Evolvingself-renewal
Before the request	Truthwhat can be asked	E-T1Asks at random; expectations are polar — “AI can do everything” or “AI is a toy.”	E-T2Sticks to 2–3 task types that once worked; never probes new ones.	E-T3Keeps a map of own tasks marked delegable / partial / not — and can explain it.	E-T4Predicts the odds of success for an unfamiliar task in advance — and the predictions hold.	E-T5Systematically probes new classes of questions; findings flow to colleagues.
Before the request	Deeptask boundaries	E-D1States the task in one phrase, however it comes out; sees no boundaries.	E-D2Has noticed that some phrasings work better — without understanding why.	E-D3Decomposes the task; draws the frame explicitly: what is in, what is out, which assumptions.	E-D4Deliberately moves the boundaries — narrows, widens, tests the frame's edge cases.	E-D5Transfers the framing skill to new domains; helps others formulate.
The request	Connectcontext	E-C1Gives no context — expects the model to “just know” the company and the project.	E-C2Pastes fragments of documents when remembered; context composition is accidental.	E-C3Knows which context each task type needs; prepares it in advance — role, data, examples.	E-C4Doses, refreshes, and prunes context; understands how its composition shifts the answer.	E-C5Builds reusable context kits — task profiles, example sets — for self and team.
The request	Serviceprompt & system choice	E-S1One request style for every system; sees no differences between models.	E-S2“This one is better for code” — choice by habit, never tested.	E-S3Deliberately picks the system and mode per task; knows the strengths and limits of those available.	E-S4Compares systems on own tasks; adapts the prompt to the specific model.	E-S5Tracks model changes, updates own practice, advises colleagues on choice.
After the response	Knowledgeworking the answer	E-K1Copies the answer as-is, or discards it whole.	E-K2Edits by hand; occasionally asks to “rewrite it shorter.”	E-K3Commands a set of transformations: format, tone, structure, audience.	E-K4Builds processing chains: draft → critique → rework → final form.	E-K5Keeps a library of output formats; the model's answer is raw material in a tuned personal process.
After the response	Evolutioniteration	E-E1One request — one answer; a failure means “AI can't do this.”	E-E2Sometimes re-asks — unsystematically, and quits fast.	E-E3Budgets 2–3 refinement cycles as the norm, not as a breakdown.	E-E4Manages the cycle: knows when to refine, when to restart, when to stop.	E-E5Extracts lessons from iterations and updates own kits — future cycles get shorter.
Cross-cutting	Responsibilityvalidation	E-R1Takes it on faith, or rejects it by gut feel; no checking.	E-R2Checks “when it matters” — with no criteria for what exactly to check.	E-R3Knows the typical failures — hallucinations, stale data; verifies facts and figures; knows where checking must stay human.	E-R4Holds a validation protocol per task type; uses the model for self-checks — knowing the limits of that move.	E-R5Formally separates delegable from non-delegable responsibility; results survive external audit; teaches others.

P — Process face

“Does the process define it, regardless of who performs it?”

Process face: 7 dimensions by 5 levels. Does the process define it, regardless of who performs it?
	Dimension	L1 Ad hocunmanaged	L2 Repeatablepockets of practice	L3 Articulatedbuilt into the process	L4 Managedprocess is measured	L5 Evolvingself-optimising
Before the request	Truthstep delegability	P-T1Delegability is recorded nowhere — every performer decides alone.	P-T2Pockets: individual performers know “their own” delegable steps.	P-T3Process steps are marked: delegable / partial / not.	P-T4Delegability is confirmed by data: success rate is known per step.	P-T5The delegation boundary is revisited whenever new models ship.
Before the request	Deeptask frames in steps	P-D1Formulations are ad hoc; each performer has their own.	P-D2Personal templates held by individual performers.	P-D3AI steps carry described frames: input, output, assumptions.	P-D4Frames are tested on edge cases; deviations are regulated.	P-D5Frames evolve from accumulated failures.
The request	Connectcontext sources	P-C1Context comes from the performer's head.	P-C2Personal document collections.	P-C3Context sources are defined for every AI step — documents, data, examples.	P-C4Context is versioned; staleness is tracked.	P-C5Context kits are a reusable asset, growing cycle to cycle.
The request	Serviceprocess prompts	P-S1Whatever each performer is used to; nothing is fixed.	P-S2Lucky prompts are forwarded around.	P-S3Prompts and model choice are documented for the AI steps.	P-S4Prompts are re-tested when models change; the choice is justified by comparison.	P-S5The process prompt library is maintained like code.
After the response	Knowledgeformats in the flow	P-K1Everyone shapes the output as they can.	P-K2Informal samples — “like my colleague does it.”	P-K3Output formats of AI steps and their place in the flow are defined.	P-K4Processing chains — draft → critique → final form — are built into the process.	P-K5The process breeds new formats and hands them to other processes.
After the response	Evolutioncycles in the plan	P-E1Iterations run on the performer's own time — the plan knows nothing of them.	P-E2Everyone iterates, but the plan doesn't know it.	P-E3Refinement cycles are budgeted into the process's schedule and estimates.	P-E4Cycle length and cost are measured and managed.	P-E5Accumulated kits systematically shorten the cycles.
Cross-cutting	Responsibilitycheckpoints	P-R1Validation “by conscience” — it belongs to no one.	P-R2The “important” gets checked; there are no criteria.	P-R3Checkpoints: where a human is mandatory, where model self-check is allowed.	P-R4A protocol per artefact type; defect metrics.	P-R5The delegation boundary is fixed and auditable; incidents update the protocol.

O — Organisation face

“Does the organisation provide and govern it, regardless of the process?”

Organisation face: 7 dimensions by 5 levels. Does the organisation provide and govern it, regardless of the process?
	Dimension	L1 Ad hocshadow	L2 Repeatablelocal initiatives	L3 Articulatednorm + enablement	L4 Managedmanaged portfolio	L5 Evolvinglearning organisation
Before the request	Truthapplicability map	O-T1The organisation doesn't know where AI lives in its work; use is shadow.	O-T2Individual teams know their own cases.	O-T3An applicability map by function and role: what is delegable, what is not.	O-T4The map is confirmed by portfolio metrics.	O-T5New areas are probed systematically; the map stays alive.
Before the request	Deeptask language	O-D1No language for describing tasks for AI.	O-D2Local team jargons.	O-D3A shared standard for formulating tasks and boundaries.	O-D4The standard is tested against edge cases and risks.	O-D5The language evolves, absorbing new task types.
The request	Connectaccess to knowledge	O-C1Data is locked away; everyone copy-pastes at their own risk.	O-C2Spontaneous wikis and shared folders.	O-C3Rules: which knowledge may be passed to AI, and how.	O-C4Context infrastructure is governed: access, freshness, confidentiality are measured.	O-C5Context is a strategic asset, deliberately grown.
The request	Servicesystem fleet	O-S1Shadow AI: anything goes.	O-S2A de-facto standard: “everyone lives in X.”	O-S3A catalog of approved systems with purposes; selection criteria.	O-S4The fleet is evaluated on the organisation's tasks; updates are justified by data.	O-S5A managed model portfolio; pilots of the new; influence on vendors.
After the response	Knowledgeoutput norms	O-K1No shared norms.	O-K2Samples live inside teams.	O-K3Corporate formats and templates for outputs.	O-K4Output quality is measured; formats are wired into the document flow.	O-K5The format library migrates between teams — a living asset.
After the response	Evolutioniteration economics	O-E1Iterations are invisible or punished — “why are you fiddling?”	O-E2Tolerated informally.	O-E3Norms and training recognise refinement cycles.	O-E4The economics of iteration are measured: cycle cost against the gain.	O-E5Investment in shortening cycles: kits, training, tooling.
Cross-cutting	Responsibilityaccountability frame	O-R1Undefined; an incident goes hunting for a scapegoat.	O-R2Unwritten notions of “what is allowed.”	O-R3Policy: what is never delegated to AI; who signs for the result.	O-R4Compliance is measured; incidents are analysed systematically; trust is calibrated by task class.	O-R5The frame is externally auditable and updates proactively.

§ ALPHABET

Words of the cube

word = carrier · dimension · level. Position resolves letter collisions: the first letter is always the carrier, the second the dimension, the digit the level.

A unit's profile is a vector of 21 words — 7 dimensions × 3 carriers. Levels are never averaged across rows or carriers.

EXAMPLES

E-C3an employee deliberately prepares context
P-R4a managed validation protocol inside a process
O-T2the organisation knows AI's place only in pockets

§ PLANES

Three ways to slice

S-facesthree 7 × 5 matrices

fix the carrier

The working views: the E-face, P-face, and O-face are the matrices teams actually use.

L-planes7 × 3 slice

fix the level

Target slices. “The L3 plane” is what must be articulated on every carrier — the natural target of a rollout programme.

D-planes5 × 3 slice

fix the dimension

Audit slices. “The R plane” is validation across all carriers — the natural view for an audit.

§ LAWS

Connectivity, made explicit

The axes project as independent — that is the price of speaking about faces separately. The couplings between carriers are recovered as three laws.

LAW 1

The institutionalisation ladder

E3 → P3 → O3

The articulation threshold is the same threshold on every carrier — in a different aggregate state: “I can explain it” (E3) → “it is written into the process” (P3) → “it is anchored as a norm” (O3). Institutionalisation is knowledge moving across carriers, and the L3 column is its ladder.

LAW 2

The gap law

|ΔL| ≤ 1

Configurations with a gap of two or more levels between adjacent carriers of one dimension are unstable and produce predictable pathologies: employees far ahead (E ≫ P, O) breeds shadow practice that leaves with the people; the organisation far ahead (O ≫ P, E) breeds paper maturity — dead regulations. The stable band is a gap of at most one level.

LAW 3

Two counter-flows of growth

E→P→O · O→P→E

Growth runs as two counter-flows: institutionalisation from below (a personal method becomes a process step becomes a norm) and enablement from above (a policy becomes a regulation becomes a trained employee). A rollout programme manages both flows, not one.

§ READING

How to read a profile

RULES

Continuous, never staged. A person, process, or organisation gets 7 values per face — not one number. “Level 3 employee” is a category error.
No averaging. L4 in Responsibility and L2 in Service is a specific, actionable shape — 3.0 is noise.
Read from the early phases. A failure early in the lifecycle devalues maturity later in it: polished context (Connect L4) feeding an unframed task (Deep L1) is still an unframed task.
Responsibility is a gate. Its level caps the criticality of tasks that may be delegated — independently of the rest of the profile.

DIAGNOSTIC SHAPES

E ≫ P, O shadow practice — personal mastery the institution can lose overnight.
O ≫ P, E paper maturity — norms that nobody's practice can honor.
L3 column the institutionalisation ladder — where enablement and bottom-up method meet.

Measure yours →