The working is always shown.
The evidence is earned.

We publish how the model is built: what it reads, what it refuses, and the mark every number carries. Inside the diagnostic, every formula appears with your own numbers in it; nothing is computed that you cannot check by hand. What we keep is the thing no one can copy overnight: the discipline of where every number comes from, and a calibration ledger that only reality can write.

The equation

Revenue times margin is EBITDA, so the multiple is an EBITDA multiple and the equation is arithmetic, not metaphor. The model's whole job is to compute the margin and the multiple from observable structure, and to refuse to move for anything else.

The fourteen dimensions, three models

The diagnostic scores fourteen dimensions on behaviourally anchored 1–5 scales, grouped into three models: WHY (the business model: segments, pricing, revenue mix, unit economics, customer experience), HOW (the operating model: capability, efficiency, sourcing, governance), WHAT (the technology model: product, data velocity, architecture, AI, security and resilience). The three roll into one readiness index, and the index carries the model's causal spine: execution cannot outrun strategy. A platform resting on an undesigned business model is capped, and the diagnostic says so out loud. Every anchor traces to the research it rests on, and the citation appears on the screen where you score it.

Fig. I · The gap map, as every instance draws it

What is published, and where

The model is deterministic, and it is published in full to every client: the fourteen dimensions weighted into the readiness index with its causal caps, a multiple built from recurring-revenue share and readiness, a margin built from revenue mix and operating discipline. Inside the diagnostic you see every formula with your own numbers in it, every figure with its provenance mark, every anchor with its date. Nothing is computed that you cannot check by hand, and the anchors move only through the calibration ledger, never without a versioned entry.

The multiple, by sector

The band is where the market clears, sector by sector: it is set by what a company makes, and a company's place within the band is set by revenue quality, the exact variable Engine 2 moves.

Business model	EV / EBITDA	What moves you up the band
Project-based services, non-recurring	~5 to 6×	Nothing structural. This is the floor.
Testing & assurance (TIC)	~11 to 14×	Growth and recurring contracts
Fintech (blended)	~8 to 12×	Infrastructure and recurring over one-off
Cybersecurity & RegTech	~12 to 18×	Recurring revenue, mission-critical retention
SaaS / software, high recurring	~14 to 20×+	Net revenue retention and Rule-of-40 growth

Third-party comparables, 2025 to 2026, cited as published (Aventis Advisors, DealMatrix, Windsor Drake, market M&A references). Recurring revenue above 60% carries a premium over one-off work for the same EBITDA; that premium is what the model reads as your recurring share. The bands are context, marked ASSUMPTION until the ledger graduates them, never an appraisal of your company.

How the moves are ordered

Pricing a gap is half the work. The other half is saying which move to run first, and a ranking that cannot be interrogated is the same black box as a valuation nobody can check. So the ordering is published in principle, exactly as the equation is.

It weighs the euro that repeats, over the months of runway left to compound it: a move begun in month one of a five-year hold has sixty months to work, and one begun in month twelve has forty-eight. It adds what the move does to enterprise value through the multiple, taken from the same recompute the rest of the model uses. It discounts the whole of it by the confidence band on the move, because a wide band is a statement about evidence rather than about ambition. And it divides by what the move costs to put in place and to keep for its first year, taken at the top of the range: a plan that only survives at the bottom of a cost nobody has measured is not a plan.

Then three constraints are applied after the score, each one shown on the page with the reason written out:

• The causal cap. Business model before operating model before technology. A move that outruns the model beneath it is worth less than its own arithmetic says, and the room names the dimension that has to move first.
• Cash. A move whose phased cost crosses the line on the thirteen-week watch is flagged, not quietly ranked first. Cash is checked before ambition.
• Capacity. Two moves with the same owner cannot both start in month one. The clash is surfaced and a sequenced start is offered.

The client can move the weights inside stated bounds and watch the order change, which is more persuasive than a fixed answer and harder to argue with. And when a stated target cannot be reached with the moves available, the model quantifies the shortfall and names the class of action that would close it: an acquisition, a new segment, a change of capital structure. And it says plainly that it does not price that action. The refusal is the output, not a gap in it.

Provenance: five marks, no exceptions

Every figure the machine prints carries one of five marks: LAW LAW+ MARKET ASSUMPTION SIMULATION. Law is cited to the rule and dated. Market is benchmarked to third-party data. An assumption says so, and moves the moment you supply the real number. A simulation is a what-if on your own declared scores, computed by the same engine, and never becomes a headline.

The calibration protocol

The anchors change only through a pre-registered ledger: each engagement records the baseline scores, the scenario chosen and the model's prediction at entry, before any outcome is known; it closes once, with realised numbers and evidence. Anchors are refit by a human, in a versioned commit, and an automated test fails if an anchor ever moves without its ledger entry. Our test chain runs more than ninety invariants on every change, including: raising a readiness score can never lower a lawful euro, and eligibility belongs to the granting bodies alone, never to the model.

The architecture: six layers, one loop

An operating system that runs after we leave. Each layer does one job, and the last one feeds the first.

Fig. II · The loop

And then the loop closes. Every signed line in the Ledger writes back into the Drivers: which play, which sector, which realised euro. Five verified rows replace a borrowed coefficient with your own. The system gets more accurate the more it runs.

Capital is none of the six. It is the fuel the six burn. Sixteen Luxembourg and EU instruments are screened separately, with their own readiness score, and loans, guarantees and tax credits are shown beside the grants, never added to them.

What we do not do

• No fabricated case numbers, ever. Our proof line stays empty until a mandate reconciles.
• No revenue forecasting: the model moves structure, not sales.
• No score inference from documents: scores are declared and dated, by named seats, and disagreement between seats is reported as a finding, not averaged away silently.
• No macro assumptions: no interest rates, no market sizing, no narrative multiples.

The anonymous benchmark opens at twenty companies: sector and size band only, nothing else stored. Add yours from the diagnostic.