State Machine Coverage

The state machine,
layer by layer.

Twenty-three stops, four gates and one exit, in canonical order — scored against the six layers a stop needs before it is real. Click any cell for its definition, deliverables, acceptance and failure mode.

LC is now its own column. The conversion lever — the specific thing that delivers what the customer wants and moves them to the next stop — was previously buried and therefore skipped. As a column, an undefined lever is visible across the whole chain at once.

The discipline that keeps L1 and LC distinct: L1 is a belief or need and is never an action. LC is always an observable event. If you cannot watch LC happen, it is not an LC — which is also what makes L4’s completion rate computable.

B — Built — added 2026-08-19. The five thinking layers can all read green while a store still has nothing to use. B scores whether an operating surface actually exists for the seat that performs the stop. It is the only column that answers “is this in production,” and it is deliberately the last one — a surface built before its row is defined is the most expensive artefact in the programme.

Conversion levers exist where the event is transactional, and vanish where it is behavioural

15 / 28

All of Segment 4 has a defined LC — submitted, approved, signed, walked, funded, closed. Those stops are events, so the lever names itself.

Every gap sits in Segments 1–3 and 5, where the conversion is a behaviour rather than a transaction. Six are missing outright: A1 · 04 · A2 · A4 · 21 · 23.

Two of those six are the most valuable conversions in the business. A2 — showing up: a booked appointment becoming a physical arrival, the largest leak in Segment 2, and the journey map itself flags show rate as not surfaced. And 21 — first trip completed, which is the North Star.

So the chain can describe everything about closing a deal, and almost nothing about getting someone through the door or getting them camping.

Nothing is in production

0 / 24

No stop scores defined on Built. Eleven have a prototype — the portal app, the Floor Workspace check-in wizard, six competing MCP surfaces, a desking prototype, a walkthrough capture tool, and the internet-lead experiment harness. Thirteen have no surface at all.

The inversion is the finding. Segment 4 is the only fully measured block in the chain and it contains zero surfaces — 13 through 20 are entirely unbuilt. Segment 3, the least finished on measurement, holds almost every prototype that exists. So the chain measures most precisely exactly where nobody can do anything, and is built most heavily where the numbers are thinnest.

Stop 17 remains the sharpest single cell: every metric exists, the CIT Tracker covers it, it carries a universal requirement — and it has no screen. Stop 21, the North Star, has no surface and no capture event.

The mechanism works: answering L1 fills L2 for free

4 / 8

Three L1 answers assigned four of the eight agents without anyone doing an L2 pass. Signal Detection and Clarity fell out of 01; Information Collection and Decision Support fell out of 02.

This is why L1 goes first even though L2 is the larger gap. L2 cannot be specified before L1 says what needs understanding — an agent with nothing to present is just a name.

L2 is still empty across every stop past 03

21 / 24

The roster is fully specified — goal, bounded authority, tools, feedback loops. Twenty-one stops have no agent mapped, so every “the system handles this” in the fulfillment column remains an intention.

L1 is a boundary everywhere it is not yet answered

Every stop has exit criteria. Eighteen still do not define what the customer needs to understand. The hesitation ledger is the only pre-existing content and it clusters on 08–11 and E1.

Measurement and screens remain inversely distributed

Segment 4 is fully measured and barely built. Stop 17 is the sharpest case: the CIT Tracker is its primary dashboard, every metric exists — and it has no screen, no chunk, while being a universal requirement.

Segment 2 is the reverse — with two corrections as of 2026-08-19. The dashboards reaching 04 and 05 count leads rather than qualifying, pricing or timing them \u2014 which is why 04’s L4 was rescored partial on 2026-08-19. And these stops are not surfaceless: the internet-lead-to-appointment experiment harness runs both, emitting server-side events already. It is a harness, not a product — but “no surface at all” was wrong.

Both ends of the chain are still the thinnest

Gates A1 and A4 have no metric and no dashboard. E1 has zero measurement. Segment 5 has one partial L4 across three stops. Every one is a place a customer enters, leaves, or stops being counted.

Layer	What it takes	Remaining
L1	Define what the customer needs to understand at each stop, and in what form. Three answered; the intake file carries the open question for each of the rest.	18
LC	Name the observable event that moves the customer out of each stop. Must be watchable — if it cannot be observed it cannot be a lever, and L4 cannot rate it. Priority: A2 show rate and 21 first-trip completion.	13
L2	Assign the eight agents to stops, then define what the system decides, presents, and hands to a human. Four agents are already in play from the L1 work.	21
L3	Seat stops 21–23, answer “who owns Finalized,” then add the justification field to the 17 that already carry a role and verb.	3 seats
L4	Instrument A1, A2, A4, E1, 21, 23, and consume `vw_survey` at 22. Stop 21 needs a first-trip completion event that exists nowhere.	6 + 1 event
B	Ship something. For the eleven prototypes: pick a candidate of record, close prototype-to-store-ready, benchmark first per R12, then roll out with fallback and rollback criteria. For the thirteen with nothing: name the seat and the device before drawing anything. Priority is where B and L4 disagree most — 17 and Segment 4, fully measured and entirely unbuilt.	24

Layer

What it takes

Remaining

Define what the customer needs to understand at each stop, and in what form. Three answered; the intake file carries the open question for each of the rest.

Name the observable event that moves the customer out of each stop. Must be watchable — if it cannot be observed it cannot be a lever, and L4 cannot rate it. Priority: A2 show rate and 21 first-trip completion.

Assign the eight agents to stops, then define what the system decides, presents, and hands to a human. Four agents are already in play from the L1 work.

Seat stops 21–23, answer “who owns Finalized,” then add the justification field to the 17 that already carry a role and verb.

3 seats

Instrument A1, A2, A4, E1, 21, 23, and consume vw_survey at 22. Stop 21 needs a first-trip completion event that exists nowhere.

6 + 1 event

Ship something. For the eleven prototypes: pick a candidate of record, close prototype-to-store-ready, benchmark first per R12, then roll out with fallback and rollback criteria. For the thirteen with nothing: name the seat and the device before drawing anything. Priority is where B and L4 disagree most — 17 and Segment 4, fully measured and entirely unbuilt.