Don't stress the forecast. Stress the dependencies

WS03 through WS07 produced increasingly precise views of demand, customer capture, technology, manufacturing, earnings and cash. WS08 is where those views meet the management projection. The workstream succeeds when it converts the projection into a set of specific dependencies the underwriting can test, and identifies the ones most exposed to breakage under compound stress. It fails when it accepts the projection as a document to be marginally adjusted; hypothesis inversion stays harder discipline.

A projection is a set of physical, commercial and financial dependencies that must all hold simultaneously. Don't stress the forecast. Stress the dependencies that create the forecast.

The consequence for methodology: WS08 starts with the physical and commercial constraints established in the previous six workstreams, derives the maximum feasible operating outcome those constraints permit, and only then compares that outcome against the management projection. This is model inversion, and it is the reverse of conventional projection review.

Model inversion: start with constraints, derive the outcome, compare against management

Conventional DD: management gives you the forecast; analyst stress-tests it by adjusting percentages. The failure mode is that the projection remains the anchor, and stress-testing produces small deviations around a number that may already be unrealisable.

Model inversion: start with the constraints established in WS03-WS07 (demand, customer capture, product qualification, manufacturing capacity, supply availability, power capacity, financing capacity), derive the maximum feasible operating outcome those constraints permit, and compare it against what management is forecasting. Where the two agree, the projection is well-evidenced. Where the derived outcome falls short of the projection, the projection is under-evidenced at a specific constraint the DD workstream can name.

Realisable revenue equals the minimum of demand, customer capture, production capacity, supply availability, power capacity and financing capacity. Not the maximum. Not the average. And not the management forecast. The minimum of the binding constraint set. WS08 identifies which constraint is currently binding and prices the underwriting against that constraint; management's aggregate revenue number sits above it.

Revenue = min(demand from WS03, customer capture from WS04, product qualification from WS05, good-unit capacity from WS06, supply availability from WS06, power / site capacity from WS03, financing capacity from WS13). At any point in the underwriting horizon, one of those is the binding constraint. The DD workstream identifies which.

Marginal economics, not average economics

Historical gross margin and marginal gross margin measure different things. Historical capex intensity and marginal capex intensity measure different things. A target with 40% average gross margin can be generating 25% marginal gross margin on the next $100M of revenue because of new-factory ramp-inefficiency, lower utilisation, customer mix, freight, warranty accrual, or new-product ASP dilution. Same target, different marginal answer, radically different projection.

The DD test converts every input in the projection from an average to a marginal.

• Marginal gross margin on the incremental revenue tranche rather than the aggregate margin.
• Marginal capex intensity ($ of incremental capex per $ of incremental revenue in the specific ramp year) rather than historical capex-to-revenue.
• Marginal working capital intensity (incremental WC consumed per $ of incremental revenue) rather than aggregate WC-to-revenue.
• Marginal EBITDA and marginal FCF on the incremental revenue rather than aggregate EBITDA / aggregate revenue.

The economic test that matters is incremental gross profit per dollar of incremental invested capital. Same growth can be produced by very different businesses; the incremental capital-efficiency metric tells the IC which one.

Capex per unit of growth

Every industrial-tech target has a maintenance capex line and (usually) a growth capex line. The DD test is whether the growth capex projected supports the growth projected.

A second test worth naming up front: capex phasing risk. Industrial-tech capex falls outside fungible cash-in / equipment-out treatment. Long-lead items (electrical switchgear, high-voltage transformers, specialty lithography tools, cleanroom HVAC), grid interconnection queue positions, environmental permits, and construction crew availability all sit on the critical path. A capex budget that is financeable on paper can still miss its production ramp by four to eight quarters if the phasing has not been diligenced. This bears on Part VI (manufacturing and supply-chain DD, where lead-time exposure is mapped) and on Part XI (regulatory DD, where the permitting timeline is diligenced), and it feeds back into this workstream as a projection integrity question: does the CIP schedule reconcile with the equipment order book and the permit calendar the target holds?

The analysis starts historical. What did the target spend on capex in each of the last three years, and how does that spending decompose into maintenance (keeping current capacity) vs growth (adding new capacity)? Ratio the growth capex to the revenue growth it enabled (using the QoE-normalised baseline from Part VII). That ratio, growth capex per dollar of new revenue, is what the projection should be tested against. Note the timing wedge: capex often leads revenue by 12-24 months in industrial-tech, so the trailing ratio understates forward requirements in a growth phase.

A specialty industrial firm growing revenue $100M per year typically requires $12-25M of growth capex depending on manufacturing intensity. A capex-light firm (specialty services, distributor-heavy) at the low end; a capex-heavy firm (specialty semiconductor, thermal manufacturer) at the high end. CIMs that project 30 percent revenue growth with capex growing 5 percent are projecting an economically inconsistent trajectory.

Not all capex creates growth

Capex per dollar of revenue growth is a headline metric only when the capex numerator is correctly decomposed. Aggregated capex disguises which portion of the spend creates the incremental revenue the projection depends on.

1. Growth capex. Direct capacity or capability additions that create incremental revenue at the specified ramp.
2. Replacement capex. Maintains existing revenue. Necessary; does not contribute to growth per dollar spent.
3. Compliance capex. Required by regulation or safety. Neither grows nor maintains revenue directly; conditions the licence to operate.
4. Reliability capex. Protects existing revenue against downtime or quality failure. Insurance-like.
5. Productivity capex. Reduces future operating cost. Contributes indirectly via margin expansion; direct revenue contribution stays limited.
6. Strategic capacity capex. Creates optionality on future demand not yet contracted. Priced against demand probability; the current revenue plan understates the capacity.
7. Enabling capex. Produces no revenue directly but is required for other revenue to be delivered. A $50M transformer investment can enable $300M of revenue that cannot occur without it. Its economic value is indirect; its underwriting weight is the revenue it unlocks; the accounting return on the enabling asset itself understates the economic value.

The DD test forces management to disaggregate the capex plan into these seven categories. A $100M capex plan that decomposes into $30M maintenance, $20M compliance / reliability and $50M growth-plus-enabling is a very different capital allocation from $100M pure growth capex, and produces very different revenue-per-capex-dollar economics.

Capex timing is an underwriting variable

The same $100M spent on the same asset produces different economics under different timing. Front-loaded capex with delayed revenue start compresses IRR, extends cash burn and creates a specific financing requirement in the interim. Back-loaded capex with early revenue start improves IRR and reduces the funding gap. The DD workstream models capex not as a lump sum but as a monthly or quarterly cash-out profile matched against the projected revenue-in profile.

Where capex precedes revenue by more than 12 months, the workstream models interest-during-construction, working-capital consumption during the pre-revenue period, and the incremental debt or equity funding required to bridge the gap. Timing changes are frequently modelled as smaller than they run in aggregate underwriting; a 6-month capex slip against a delayed revenue start can consume more IRR than a 15% capex overrun.

Unit economics: incremental vs average

Two firms with the same reported margin can have very different unit economics if incremental units are different from average units. The DD test is whether the incremental unit is more, equally, or less profitable than the average.

Firms whose incremental unit is more profitable have positive marginal contribution and will over-deliver on scale-up. Common pattern: initial fixed cost absorption improves as volume grows.

Firms whose incremental unit is less profitable have declining marginal contribution and will disappoint on scale-up. Common pattern: customer acquisition cost rises as the addressable customer base saturates, or manufacturing marginal cost rises as capacity constraints tighten.

Decomposing incremental from average requires segment-level margin data. If the target discloses top-10 customer profitability separately from the tail, the ratio between top-10 margin and tail margin is a first proxy for incremental economics. If the ratio is above 1 (top-10 more profitable), scale-up should improve overall margins. If below 1 (tail more profitable per unit), scale-up degrades them. The proxy is first-order and needs a caveat: top-10 vs tail margin differences can be driven by product mix, channel mix, or geography; scale-economics drivers play a smaller role. Where the top ten customers concentrate in one product line and the tail in another, the margin gap is a mix effect that will not travel with new customers who buy a different product basket. The tighter analysis controls for product line and channel before drawing scale-economics conclusions (cross-reference Part IV customer and pricing DD, where customer-level profitability is decomposed).

Projection stress-testing: three downside cases

The projection stress-testing workstream runs the three downside cases from The Investment Layer Part VII against the target’s numbers.

Demand-side shock (cross-reference Parts III and IV for the underlying market and customer inputs). The AI monetisation lag scenario: hyperscaler capex growth compresses from 25-30 percent to 5-10 percent for eighteen to twenty-four months. Impact on the target depends on hyperscaler revenue share and contract structure. A target with 60 percent hyperscaler revenue and no take-or-pay contracts loses 15-25 percent of projected revenue in the shock year.

Supply-side reversal. Grid interconnection queue reform, transformer capacity build-out, or technology substitution reduces the scarcity premium in the target’s market. Impact depends on how much of the target’s pricing is on scarcity. A specialty firm with 20 percent price premium attributable to scarcity loses 15-20 percent of gross margin in the reversal scenario.

The third scenario is a financing shock: rate persistence combined with refinancing failure at 2020-2022 vintage sponsor debt. Impact on the target depends on the debt structure at close. A target with 60 percent LTV construction debt maturing in 2027 at 3.8 percent coupon refinancing at 6.5 percent sees coverage ratio move from 1.35x to 1.05x, which is below most covenant thresholds.

Compound failure and causal downside

Management downside cases typically apply linear haircuts (revenue -10%, margin -200bps) to independent variables. Real-world underwriting failure looks nothing like that. Failures compound because the underlying variables are causally linked, and the compound outcome is materially worse than the sum of the linear haircuts.

A 10% revenue shortfall combined with 5% ASP compression combined with 100bps yield deterioration combined with a 20% capex overrun combined with a 3-month ramp delay does not produce a 10% EBITDA decline. Each individual assumption may be survivable; together they typically break the equity case. The stress-testing philosophy the workstream should apply is that compound outcomes form the base case for downside modelling; outlier framing understates them.

Causal downside scenarios; Bear / Base / Bull haircuts stay weaker

Downside cases should be constructed as specific causal chains; arbitrary Bear / Base / Bull haircuts stay weaker. Three downside scenarios cover most AI-infrastructure failure modes.

Downside 1: Demand. Demand grows slower than the WS03 executable-demand model. Utilisation on the incremental capex falls. Fixed cost per unit rises. Gross margin compresses. Working capital rises against slower sell-through. Cash flow deteriorates and the covenant headroom identified in WS07 compresses.

Downside 2: Execution. Capacity arrives late (per WS06 ramp-credibility scenario 5 or 6). Revenue is delayed. Capex is already spent. The financing gap the WS08 model predicted becomes real. Interest-during-construction consumes IRR. The refinancing analysis identifies specific covenant test dates at which the gap becomes visible to lenders.

Downside 3: Economics. ASP falls as competitors qualify (WS04 qualification-as-clock scenario) or as allocation premium unwinds. Gross margin compresses. Return on the new capacity collapses even where the revenue plan is achieved in nameplate terms. The WS07 gross margin bridge identifies which specific bridge components take the hit.

Break-even assumptions and the deal-thesis break conditions

The IC needs to know not the base-case number but the assumption thresholds at which the investment case breaks. For each critical driver, the workstream produces a break-even value.

• Minimum ASP the underwriting tolerates
• Minimum utilisation at target capacity
• Maximum capex overrun
• Maximum ramp delay in months
• Minimum gross margin
• Minimum liquidity through the ramp period
• Minimum debt-service coverage at the tightest covenant test date

Then the IC asks which assumption is closest to its break threshold. If the deal thesis fails when incremental capex exceeds $X per revenue dollar OR ramp slips beyond 9 months OR gross margin falls below 28%, the IC decision converts from "do we like the base case?" into "which of these three conditions are we underwriting against, and what deal-paper protection do we require in WS12/WS13 for each?" That reformulation is much stronger than a base-case number.

P/D/I evidence provenance for forecast dependencies

Every projection is a set of load-bearing assumptions with different evidence quality. The workstream tags each critical assumption with its evidence provenance.

• P (Primary). Signed contracts, executed customer confirmations, evidenced production data, verified capacity commitments.
• D (Derived). Calculated from primary evidence via a documented methodology (a marginal margin derived from a mix decomposition on evidenced ASP and evidenced volume).
• I (Inferred). Management projection unsupported by evidence at Primary or Derived level. Frequently the largest single category in a management forecast.

Forecast confidence equals the strength of the weakest load-bearing assumption; average confidence across assumptions overstates the robustness. A forecast that depends on five inferred assumptions is fundamentally different from one that depends mostly on primary evidence, even if the aggregate confidence sounds similar. Where a critical assumption is Inferred and cannot be moved to Derived or Primary within the Stage 2 timeline, the underwriting either prices that specific inferred assumption into the deal structure (WS12) or excludes it from the base-case model.

The forecast dependency stack

Every management revenue forecast rests on the same underlying dependency stack. The workstream reconstructs the stack explicitly and requires each dependency to be evidenced at the P/D/I level appropriate to its materiality.

Dependency	Source workstream	Evidence tag
Customer demand at required MW / units	WS03 executable-demand model	Per-customer Demand-Reality-Ladder stage
Target customer capture and share retention	WS04 customer power framework	Per-customer retention mechanism + qualification depth
Product qualification at each customer generation	WS05 qualification map + evidence hierarchy	Level 1-8 evidence per critical claim
Production capacity at required timing	WS06 good-unit capacity + ramp credibility	Ramp-credibility scenario 1-6
Supply availability + capacity entitlement	WS06 supply chain	Independent failure paths + capacity-entitlement evidence
ASP retention and pricing mode stability	WS04 pricing modes + shortage-end test	Qualification-led vs market-cleared classification
Yield trajectory at production scale	WS06 yield decomposition + WS07 warranty leading indicator	Yield-to-EBITDA bridge
Working capital consumption at growth pace	WS07 cash-required-to-grow	Per-year incremental WC intensity
Financing capacity to fund the ramp	WS13 financing structure	Debt capacity + equity commitment + refinancing tenor

Each row in this stack is a dependency that must hold simultaneously for the projection to occur. Naming each dependency, assigning its source workstream and tagging its evidence provenance produces a forecast the IC can price against; a forecast the IC has to trust stays weaker.

Refinancing risk on 2020-2022 vintage debt

For sponsor-owned targets, the specific financing shock most likely to bind is refinancing of 2020-2022 vintage construction or acquisition debt. Estimates of the industrial-tech-adjacent vintage cohort maturing 2025-2028 run into the tens of billions across syndicated loan, private credit, and sponsor-backed construction debt combined; the exact figure depends on how the perimeter is drawn and which sector taxonomy is used, but the direction is unambiguous and the concentration in 2026-2027 is where the wave binds (see The Investment Layer Part VI for the fuller decomposition). Refinance rates as of mid-2026 run roughly 5.75-6.75 percent all-in on comparable structures.

The coverage math is direct. Debt originated at 4.0 percent with 1.30x DSCR refinances at 6.5 percent with 1.05x DSCR at the same operating cash flow. This isolated example undersells the case; refinancing shocks tend to arrive coincident with demand softness, so the compound scenarios below hold more of the real risk than this single-variable stress does. Below 1.10x is where most covenants bind. Refinancing lenders typically require either equity infusion (developer top-up) or lease amendment (rent increase) to close the gap.

The DD implication is that any sponsor-owned target with 2020-2022 vintage debt maturing before Year 3 of the acquirer’s hold period will require refinancing during the hold. The financing structure at close needs to build in the refinance capacity; the target refinancing itself stays an assumption.

Three recurring failures

The first is capex projected as maintenance only. The projection line item shows capex growing 5 percent per year while revenue grows 30 percent. This is arithmetically incoherent for any manufacturing business but appears frequently in CIMs.

The second is unit economics extrapolated as flat. The DD projection assumes incremental unit profitability equals average unit profitability. Both scale-up upside and scale-up disappointment get missed. The decomposition into top-10 vs tail margin is a two-hour analysis that avoids the issue.

Third, stress tests that model scenarios in isolation. The DD projection tests demand shock (impact -20 percent) and financing shock (impact -8 percent) separately. In reality, the three shocks are serially correlated alongside being simultaneous. A demand shock compresses EBITDA, which compresses coverage ratios, which triggers covenant breach, which triggers financing shock as the lender re-prices or refuses to refinance. A supply-side reversal (input cost spike) squeezes gross margin, which reduces the operating cash available to service debt, which again feeds into financing shock. Isolated stress-testing treats these as independent draws when they are conditional draws. Compound scenarios matter more than isolated ones because the shocks travel through the same channels, and the DD projection should be tested against the compound case as the base scenario; outlier framing understates it (Investment Layer Part VII).

A worked example

Continue the magnetics acquisition. Capex + growth economics scope:

Historical capex: $18M average over last three years, of which $6M maintenance and $12M growth. Growth capex per revenue growth: $12M supporting ~$45M revenue growth = 27 percent capex intensity. CIM projects $16M annual capex supporting $120M revenue growth = 13 percent intensity. Projection under-capitalises by ~50 percent. The gap traces to specific Part V/VI findings: the second-generation product qualification cycle requires ~$8M of tooling and cleanroom retrofit that the CIM omits, and the second-source ferrite qualification (Part VI supply chain finding) requires ~$3M of additional bench and pilot-line capacity. Corrected capex projection: $22-24M annually.

Unit economics: top-10 customers 32 percent gross margin, tail 28 percent. Top-10 more profitable = positive marginal contribution. Scale-up should improve overall margins slightly. Small effect (~50 bps) but directionally positive.

Stress test. Demand shock: hyperscaler capex growth compresses 20 percentage points; target loses 22 percent of Year-3 revenue (two hyperscaler customers are 78 percent of revenue in this illustrative example). Supply reversal: less impact because the target does not rely on scarcity pricing. Financing shock: 2024 refinance already priced in; less exposure than typical.

Compound: demand + financing shock together produces -30 percent Year-3 EBITDA. Deal thesis needs to hold under this scenario to be acceptable.

Current market conditions

The refinancing wave through 2025-2028 (Investment Layer VI) means most sponsor-owned targets face refinancing exposure that DD projections should treat as a probability-weighted event; tail-risk framing understates it. Sovereign co-investors in the buyer pool (Investment Layer VIII) can take on refinancing volatility with longer duration; conventional sponsor structures typically cannot.

Closing

Capex per unit of growth, unit economics decomposition, and compound stress-testing complete the financial DD workstream. The output feeds directly into Part XII (deal structure. Earnout mechanics tied to capex-dependent milestones), Part XIII (financing structure sized against DD-normalised EBITDA with compound-shock coverage headroom), and Part XIV (post-close validation, where the Day 90 capex-versus-plan comparison is the earliest signal of projection integrity). Compound-shock modelling is where DD workstream links back to the Investment Layer series Parts VI and VIII on refinancing wall and sovereign capital duration. The acquirer who accepts a refinance-in-2029 assumption without stress-testing the compound demand plus rate case is repeating the pattern that Investment Layer Part VI cataloged across 2020-2022 vintage sponsor deals.

Part IX moves into the operational workstream: organisational, personnel, culture, and systems.