Blank
Nobody owns it. Intention counts as blank.
The canvas
Twelve cells, four bands, one surface.
Intent sets what the system is for and who may act. Composition decides what work exists and who holds it. Motion is how work moves and improves. Proof is what survives contact with an auditor. Fill it for one workflow, never for a whole company.
Fill it forOne workflow, never a whole company
Time in the roomNinety minutes, five moves
ScaleBlank · Thin · Solid, anchors set locally
FormatPaper first, structured data second
Rests onSix first principles about work
Score a cell to try it. Click a cell name for its reference entry.Run the ninety-minute protocol
Work Composition Canvas · v1.0Twelve cells · four bands · one workflow
Workflow
Scored by
Date
Version
Intent
Read the bandwhat this is for, and who may act
01Purpose
What orients a person and an agent when neither can see the whole system?
Which decisions may an agent make alone, and who is able to stop it?
What is the smallest unit that can own a customer outcome end to end?
Composition
Read the bandwhat work exists, and who holds it
For each workflow, what is the human to agent ratio, and who signs the quality baseline?
05Strategy
Where on the evolution curve does each capability sit, and does that dictate build, buy, or let go?
What is owned, what is rented, and what is summoned only when needed?
Motion
Read the bandhow work moves, and how it gets better
07Flow
What limits throughput, and what is supposed to happen when the line stops?
08Learning
How does tacit judgment become reusable, and who owns the loop that has to close?
What synchronizes the work when there is no room and half the team is not a person?
Proof
Read the bandwhat survives contact with an auditor
10Evidence
What would you show an auditor, and can the system produce it without being asked?
11Mastery
Who is qualified to overrule the machine, and how would you prove it?
What are people paid for once output is no longer the scarce thing?
0of 12 scored0blank0thin0solid
A blank cell is a finding, not an embarrassment. Intention counts as blank.workcomposition.org · v1.0 · CC BY 4.0
The scale
Blank, thin, solid, and why we do not tell you where the line is
Three points per cell. Blank means nobody owns it. Thin means it was decided in a meeting and never written down. Solid means it is written, staffed, and producing evidence without being asked. One anchoring rule holds everywhere: intention counts as blank.
Thin
Decided in a meeting and never written down.
Solid
Written, staffed, and producing evidence without being asked.
The anchors are yours to set
A maturity model scores every organization against one ladder, which is false precision wearing a number. Solid for a supplier onboarding workflow at a defense supplier and solid for internal expense triage are not the same standard, and pretending otherwise produces scores that travel badly and mean nothing to the person who has to act on them.
So the scale ships uncalibrated on purpose. The three points tell you where you fall against anchors you set, for this workflow, in this regime. Calibration is local work: the organization does it, usually with a consultant, a tool, or both. Training exists to make a handful of the harder placement calls, not to hand you a threshold.
The one thing fixed everywhere is the anchoring rule. Intention counts as blank. Without that, every scale drifts upward until solid means somebody meant to.
Where the software analogy stops
The method borrows one move from software engineering: describing a recurring arrangement by its context, its forces, and its resolution, so that two strangers can argue about it without joining anything.
The borrowing stops at the people. Software components do not have careers, resentment, or a mortgage, and they do not read the pattern that reassigns them.
So a shape on this site describes a structure, never the people in it, and its preconditions are mostly about belief and tenure, which is where the structure stops being enough. The objection to the whole approach is published alongside it.
This boundary is named here, next to the calibration box, for the same reason that box refuses false precision. A method that pretends people are components will be told so by the first person it reassigns.
The twelve cells
What gets certified, cell by cell
The poster is the working surface. This is the one line per cell that the credential would assess. Open any row for the question, the tradition it inherits from, the organization that solved it first, and the documented failure.
| Cell | Name | What gets certified |
|---|---|---|
| Band 1 · IntentWhat this is for, and who may act. | ||
| 01 | Purpose | Writing intent specific enough that an autonomous unit can act on it, general enough to survive being wrong. |
| 02 | Authority | Decision-rights mapping and escalation trigger design, including resistance to automation bias. |
| 03 | Structure | Sizing the autonomous unit, and knowing the headcount at which it needs governance rather than good intentions. |
| Band 2 · CompositionWhat work exists, and who holds it. | ||
| 04 | Composition | Task deconstruction and automation-type matching, assessed on the candidate's own workflows.Test one workflow's fit |
| 05 | Strategy | Capability-evolution mapping, and knowing which parts of the map an agent may be trusted to operate. |
| 06 | Resources | Distinguishing leverage from lock-in when the resource is a model you do not control. |
| Band 3 · MotionHow work moves, and how it gets better. | ||
| 07 | Flow | Constraint identification in a mixed line, where the constraint is usually the review queue. |
| 08 | Learning | Designing the tacit-to-explicit conversion without flattening the variance invention needs. |
| 09 | Coordination | Accounting for coordination cost honestly, including the cost that reappears elsewhere after a cut. |
| Band 4 · ProofWhat survives contact with an auditor. | ||
| 10 | Evidence | Auditable agent operations: immutable trails, reasoning logs, a named accountable human per loop. |
| 11 | Mastery | Oversight competence: understand the system, intervene in it, halt it. Assessed by simulation, proctored, not by quiz. |
| 12 | Incentives | Designing reward for judgment and exception handling rather than for throughput a machine now supplies. |
The calibration record
Prove you calibrated, do not just claim a score
A score with no record behind it is an opinion with a number attached. Five fields, kept per cell, make a score comparable across units and defensible a year later when nobody remembers the conversation.
The anchor you set
What solid means for this cell, on this workflow, written in a sentence somebody outside the room could apply. Not a definition of the cell, a threshold for it.
Without this, a score is an opinion with a number attached.
Who set it, and in what role
Anchors set by the team that owns the workflow drift optimistic. Anchors set only by risk drift pessimistic. Record who was in the room, because the answer explains the score.
Anonymous anchors are unarguable, which sounds like a strength and is not.
What evidence moved it from thin to solid
Name the artifact. The written decision-rights list, the escalation trigger that fired in a test, the rostered person. If nothing can be named, the cell is thin.
This is the field that makes the score defensible a year later.
When it was last revisited
Anchors go stale faster than scores. A regime changes, a workflow gets an agent, a person leaves. Date the anchor, not just the rating.
An undated anchor is why maturity assessments feel dishonest by year two.
Where you differed from a sibling unit
Two units in the same company will anchor differently, and that is legitimate. Recording the difference is what lets them compare at all, and it is usually the most useful conversation the exercise produces.
Forcing identical anchors across units destroys the information.