QA Level 2 — Internal audit, WP1 Market

Documents audited: livrables/verdal-wp1-market.pptx (17 slides) + livrables/verdal-wp1-market-model.xlsx (7 tabs, 30 inputs, ~100 live formulas) Type: Board deck + supporting market model · capital-allocation decision · Date 27 July 2026 Method: independent Cowork subagent, isolated context — no access to the conversation that produced the deliverables. Dynamic criteria grid built for this deliverable (14 substance criteria + the 8-criterion design grid D1-D8). The model was rebuilt from its 30 assumption cells in Python and every value recomputed independently. Portion read: 100% of both artefacts, including the uncompressed sensitivity formulas. Honesty note on design: the audit read an extracted text rendering of the .pptx, not rendered pixels. D1 (colour), D2 (typography) and D8 (finish) are marked to be confirmed at render. Confirmed from the file itself: 13.33 × 7.5 in = 16:9, no ${...} tokens, no template placeholder strings, no Lorem.

No deliverable was modified.

Criteria grid

Overall score: 66/95 (3.5/5)

Substance 49/70 · Design (scoreable) 17/25 · three design criteria deferred to render.

Title-flow test

The 12 content-slide titles, read alone and in sequence:

(2) The €25bn measures Europe's charging build-out, not Verdal's margin — which is nearer €7m a year, on a lease fee we have had to assume · (4) Charging figures circulate at four levels, and only the fourth — what a landowner keeps — is money Verdal would ever bank · (5) Three published constructions land near €25bn for Europe in 2030, and not one of them measures what a site host keeps · (6) Even Europe's entire €20.4bn of plug revenue becomes €7m for Verdal in three steps — and the Board's €25bn sits above even that top line · (8) Six steps take European electricity down to 1.5 TWh of grocery destination charging, and three of the six are our judgement, not a published split · (9) Bottom-up, 1,100 stores at four bays turn 84 GWh a year — under 6% of the pool, and all of it scales off a bay count Verdal has not supplied · (11) A renegotiated revenue share beats today's lease by under €1m a year, while operating the chargers ourselves loses €22m on the same energy · (12) Self-operation loses €22.4m on high-power charger economics, and fixed cost per charger does €44m of the €70m of damage · (13) No demand assumption turns self-operation positive — but two equipment assumptions do, and neither of them is settled · (14) On a high-power spec, self-operation clears break-even in three of forty-two combinations of bays and throughput · (15) Five inputs decide this answer, Verdal already holds three — the two to be measured are destination throughput and a real grid-connection quote · (16) Three actions before 8 September decide whether the interim reports a threshold or a forecast, and two are data requests Verdal can send this week

Verdict: PASS, with two repairs. The titles alone deliver the full argument — the figure is real but sits at the wrong perimeter level; here are the four levels; here is where €25bn could have come from; here is the narrowing; here are both routes to the volume; here is what each structure earns; self-operation loses money and why; it turns on equipment, not demand; here is what would have to be true; here is what we need and what we do next. Very few decks pass this cleanly.

Two repairs: title 13 is factually contradicted by its own exhibit, and the sequence never states the incremental prize — a reader of titles only ends up with "€7m" when the model's own conclusion is "€0.5m incremental".

Arithmetic and deck↔︎model tie-out

What ties — every model formula reproduces exactly

Numbers that appear on a slide and exist in no model cell — five

1. Slide 6, the bridge step "−40.5 · Cost of delivery". No cell holds it. It is 47.578 − 7.137 = €40.44m, which is the 85% the operator retains under a revenue share — not a cost of delivery. The actual cost of delivering that energy is €70.02m (slide 12, tab 3!D15:D20). The deck therefore tells the Board on slide 6 that delivery costs €40.5m and on slide 12 that it costs €70.0m. This is the single most attackable object in the package. Fix: re-label the step "Operator's 85% share, −40.4", or rebuild the last leg as two steps — "Verdal's gross billing 47.6 → less the 85% retained by the operator → 7.1 revenue share" — and add the cell as tab 3!D26 so it is live.
2. Slide 13, "Charger spec 200 kW → 75 kW: −22.4 to +0.6". No cell. Reproduces as +0.653 using F30's own stated destination-grade parameters (€80,000 installed, €8,000/yr opex) — defensible, but it lives only in a prose note and the client cannot recalculate it.
3. Slide 13, "Bays per charger 2 → 4 outlets: −22.4 to −2.8". No cell. Reproduces as −2.830. It exists only as prose in F17, which itself says "It belongs in the sensitivity and it is not in it." The model knows the gap and ships anyway.
4. Slide 13, "Contribution/kWh €0.25 to €0.45: −29.8 to −17.6". No cell. Reproduces as −29.770 / −17.565. tab 5!A31 names the driver but computes no result for it.
5. Slide 8's four narrowing percentages (−70%, −78.1%, −69.4%, −42.3%). Computed off the rounded one-decimal display labels, not off tab 1. From the cells the correct figures are −70.0%, −78.0%, −70.0%, −43.0%. Two of the four are visibly wrong to anyone who divides the numbers printed beside them — on the page whose whole claim is arithmetic transparency.

Contradictions between deck and model — four

Three further internal defects

• Slide 14's kicker, "More bays make the loss larger, not smaller", is contradicted by the last column of its own table. At 160 kWh/bay/day the sign flips: 2 bays −2.50, 8 bays +3.19. True only below break-even. Fix: "Below break-even throughput more bays deepen the loss; only above about 160 kWh a bay a day does scale help".
• Assumptions!C24 = €0.19/kWh while its own source line F24 cites the Eurostat EU average as €0.1837. Slide 12 presents €16.0m as "Eurostat non-household band IC", implying the Eurostat figure was used. The 3.4% uplift is an unstated judgement.
• tab 4!C4 header reads "Value (€m a year)" and row 5 puts 25,000 in it, while B5 insists in capitals that it is cumulative and not an annual flow. The header contradicts the cell's own warning. Fix: header → "Value (€m)", plus a per-row unit column with "cumulative" on row 5.

Strengths

1. The honesty discipline is the best thing here, and it is exactly what the client said it would score on. F15 and F37 say in capitals that the client did not supply the figure; F25 says the grid charge is published nowhere in the six countries; F37 discloses that the benchmark is contested between two of the firm's own research passes; F17 admits a €19.6m driver "belongs in the sensitivity and it is not in it"; tab 2!D17 says the reconciliation can prove only order of magnitude; tab 5!A33 names the counterargument the model deliberately does not price. Very few models disclose their own omissions by name.
2. The model is technically clean and genuinely live. 30 inputs on one tab, nothing hardcoded downstream, the 42-cell grid built from full formulas referencing Assumptions rather than pasted values, the break-even a closed-form solve. Every value reproduced to four decimal places from the inputs alone. Zero errors.
3. The perimeter architecture is the right answer to the mandate and it is carried consistently. Four levels, named, with whose money each is, and an explicit refusal anywhere in the model to divide €25bn into an annual margin (tab 4!A13). The refutation is framed on scope, not on error.
4. Two independent routes to the volume, reconciled (top-down 1.4586 TWh, bottom-up 84.3 GWh, 5.8%) with the primary estimate declared.
5. The deck refuses the easy recommendation. Slide 13 says explicitly that WP1 must not rule out self-operation because WP2 owns the spec that decides it. Harder and more correct than a clean verdict.

Weaknesses, ranked by severity

1. Slide 6's "Cost of delivery −40.5" is wrong by €29.6m against slide 12 of the same deck.
2. The deck's headline prize (€7m) contradicts the model's own definition of the prize (€0.537m). The deck makes precisely the comparison the methodology named as the error to avoid.
3. Deck and model disagree on which perimeter level €25bn occupies. This is the mandate question.
4. Slide 13's title is contradicted by slide 13's chart.
5. Step 5 of the methodology is undelivered: no bounded range on the headline number. €7.1m rests on three stacked hypotheses (4 bays × 15% share × 80 kWh).
6. The confidence scale appears nowhere in the deck, despite being the client's explicit scoring criterion and applied rigorously in the model.
7. Three of the five bars on slide 13 and both of slide 8's inner percentages have no cell behind them. The deck promises on slide 14 that "the grid is live, not a picture" — slide 13 is a picture.
8. Slide 8's bullet credits IEA with the 30% public split the model says nobody publishes.
9. Slide 14's kicker is contradicted by its own last column.
10. FX error in F37: £3,000-5,000 → "€13-22m" implies £1 = €1.00; at €1.17 it is €15.4-25.7m, and the multiple on slide 11 is 2.3-3.9x, not "2-3x".
11. Slide 2 carries six supports against an explicit "never six".
12. Slides 15-16 sit under the wrong section eyebrow.
13. Slides 5 and 16 are prose walls; slide 5 is a table pretending to be paragraphs.
14. Break-even is DC-only and inconsistent with the grid beside it (159.8 vs the true 162.2).
15. C24 = 0.19 is presented as the Eurostat figure (0.1837). Small, but exactly the kind of unstated rounding the client said it would penalise.

Delivery risks

• "Your slide 6 says delivery costs €40m and your slide 12 says €70m. Which is it?" No good answer exists in the room. This is the embarrassment risk.
• "You told us the prize is €7m. Your own model calls the prize €0.5m. Which number are you asking us to allocate capital against?" The honest answer — €0.5m incremental, on a lease fee we invented — is a far weaker headline than the deck carries, and the Board will feel it was softened.
• "Your appendix says at the only published benchmark the existing lease is worth €13-22m a year. So the answer to 'is value being left on the table' may be no — and your headline doesn't say that." Slide 11's third bullet concedes it; slides 2, 4 and 6 do not.
• "You say the Board cited the right number from the wrong report." Whatever the hedging on the next line, this is heard as the Board confused an auto-loan report with charging. Sponsor-relationship risk with Katrin Sommer, who has to live with that Board.
• "You compared our six-country grocery margin to all of European plug revenue to get 1/2,850th." A valid perimeter illustration and also a rhetorical device. The like-for-like comparator inside the same segment is €7.1m against €766m = 1/107th, which is on no slide. Add it on slide 4 before someone else does.
• "Slide 8's own numbers don't divide to your own percentages." Small, visible, and it undermines the page whose whole claim is arithmetic transparency.
• "You modelled 80 kWh a bay a day and Fastned does 150-200. Why is our grocery car park half a motorway station?" The model's answer (tab 5!D12) is good but lives in Excel; slide 13 mentions Fastned only in a source line.
• Latent risk: the entire operate case rests on a highway-grade 200 kW spec that the deck itself says is the wrong spec for a supermarket. If a Board member reaches that conclusion before slide 13 does, the −€22.4m headline reads as a straw man. Slides 12 and 13 should carry the destination-grade column side by side rather than in sequence.

Decisions needed from the consultant

1. Which perimeter level does €25bn occupy — level 1 (cumulative capex, per slide 4 and tab 4!C5) or level 2 (annual plug revenue, per tab 1!A20)? The evidence supports either; the package must say one.
2. What is the headline number of WP1 — €7.1m (the pool at Verdal's level) or €0.5m (the incremental prize over today's assumed lease)? Both defensible; they lead to different Board conversations.
3. Keep, soften, or cut the McKinsey mobility-finance line on slide 5. The sharpest piece of research in the deck and the one most likely to cost the sponsor political capital.
4. Present the Savills UK range as a live scenario, or hold the whole margin comparison until the two lease agreements arrive? Showing it means telling the Board that on the only published benchmark today's lease already wins. Holding it means shipping an interim whose central comparison is knowingly built on a placeholder. There is no third option.
5. Does WP1 state a view on self-operation at all, given the result is decided by a charger spec WP2 owns? The current deck says "must not rule it out" — correct methodologically, and it may read to a Board as a refusal to answer.
6. The house FX rate for GBP→EUR, so F37 and slide 11 can be corrected consistently.
7. Whether to brief Katrin Sommer on the perimeter finding before 8 September. The finding is stable, does not depend on any data-room item, and reverses the premise of a Board paper her Board wrote. Six weeks is a long time to hold that.