Caps, Sub-Caps and Carve-Outs: What Actually Survives When Several Things Go Wrong

There are three ceilings on what you can recover from a contractor, and only two of them are in the contract. The first is the layered cap structure — daily rates, sub-caps, an aggregate LD cap, an overall limit. The second is the exclusion clause, which removes entire categories of loss before any cap is reached. The third is the contractor's balance sheet, which does not appear in the document at all and is frequently the binding constraint.

A 100% liability cap on a $90m contract is a $90m promise from an entity that may be worth $25m. The cap describes what is owed. It says nothing about what arrives.

This is the eighth post in the series and the last of three on risk allocation. You will get the full layered structure, why carve-outs routinely fail to do what both parties intended, why a standard consequential loss exclusion probably does not exclude lost profits, how the layers apply when a project is late and short and defective at the same time, and how to model all three ceilings together.

ℹ️ Note: This describes how these mechanisms work in practice. It is not legal advice — the classification of losses and the enforceability of exclusion clauses are jurisdiction-specific and turn heavily on the precise drafting.

What Does a Liability Cap Actually Cap?

Several different things, at several different levels, and the levels do not necessarily interact the way the headline number implies. A contract typically contains four tiers plus a set of exceptions, and a claim travels up through them.

Tier	What it limits	Typical level	Covered in
Daily rate	Exposure per day of delay	Derived from carrying cost	Post 2
Delay LD sub-cap	Total delay damages	~10–15% of contract price	Post 2
Combined LD cap	Delay plus performance LDs together	~20–25% of contract price	Posts 2 and 4
Overall liability cap	All contractor liability under the contract	Often up to 100% of contract price	This post
Carve-outs	Liabilities excluded from the cap entirely	Unlimited, or separately limited	This post

Clifford Chance makes the point that ought to govern how the overall figure is set: caps should be "project specific" rather than driven by "market practice." A number borrowed from a comparable deal is a number that was calibrated to somebody else's risk profile, contractor and capital structure.

The tiers matter because a claim can be defeated at any one of them while passing every other. A delay claim within the daily rate and within the delay sub-cap can still be blocked by the combined cap if a performance claim got there first — the point made in post 2 and worth re-testing every time a second claim arises.

Why Can a 100% Cap Be Illusory?

Because the cap is a contractual ceiling and recovery is a credit question. Clifford Chance states it plainly: "a 100% cap can be illusory on a very large project," depending on the contractor's financial capability.

This is the same structural issue post 2 raised about security, arriving from the other end. There, the question was what instrument pays — an on-demand bank guarantee versus a surety bond requiring proof of default. Here it is more basic: is there anything behind the number at all?

Three things determine the answer, and none of them is the percentage:

The contracting entity's net worth. A project-specific subsidiary with nominal capitalisation can sign a $90m cap without any capacity to honour it. The cap binds the entity that signed, not the group.

The parent guarantee's scope. Where a guarantee exists, check whether it guarantees the full capped liability or only defined obligations. A guarantee limited to performance obligations may not reach a damages claim.

What is posted rather than promised. Retainage, letters of credit and bonds are collectable. The rest of the cap is an unsecured claim that ranks alongside every other creditor if the contractor fails — and a contractor facing a nine-figure liability is a contractor with an incentive to fail.

The practical consequence is that the cap should be modelled as three numbers, not one: what is claimable, what is capped, and what is collectable. The third is usually the smallest and is the only one that funds anything.

ℹ️ Note: A larger cap against a weaker counterparty is often worse than a smaller cap against a stronger one, because it substitutes a headline number for the credit analysis nobody then performs.

Which Liabilities Sit Outside the Cap?

The ones an owner cannot sensibly accept as capped, plus the ones the law will not permit to be limited. Clifford Chance identifies eight categories that are typically excluded:

• Third-party liabilities — death, injury, property damage and statutory fines
• Rework costs — the "cost of carrying out the works" to rectify defects
• Insurance recoveries — construction and erection all-risks proceeds
• IP infringement — third-party intellectual property claims
• Advance payment repayment — where structured as a loan
• Fraud and wilful misconduct — "often legally non-limitable anyway"
• Liquidated damages — "typically sub-capped within the overall cap"
• Enforcement costs — the expense of pursuing contractual remedies

Two of these deserve attention beyond the list.

Rework is the big one commercially. If the cost of making defective work good counted toward the cap, a contractor could exhaust its liability by building badly and then decline to fix it. Carving rework out means the obligation to complete the works properly survives independently of damages — which is why it is standard and why its drafting is worth checking rather than assuming.

Liquidated damages sit inside, not outside. Clifford Chance's framing is precise: LDs are "typically sub-capped within the overall cap." They consume headroom. A project that draws heavily on delay and performance LDs has less of the overall cap left for everything else, which is exactly the scenario the next section models.

Do Carve-Outs Count Toward the Cap or Sit Above It?

This is the drafting failure that turns a negotiated position into a dispute, and Clifford Chance flags it directly. A carve-out can either "accrue towards" the cap or sit "completely outside the capping regime" — and, in their words, "the latter is usually intended, but then not properly reflected in drafting."

Read that twice, because it describes an error that is both extremely common and entirely invisible until a claim is made.

The two constructions produce very different outcomes. If an uncapped indemnity accrues toward the overall cap, then a large third-party claim consumes the headroom that would otherwise have been available for delay and performance damages — the owner wins the indemnity and loses the LDs. If it sits outside, the two are independent and both are available in full.

The fix is mechanical and costs nothing at drafting stage: state explicitly, for each carve-out, whether amounts recovered count toward the aggregate cap. A clause that lists exceptions without saying which side of the ledger they fall on has deferred the question to litigation.

The same ambiguity affects the relationship between sub-caps and the overall cap. If delay LDs are sub-capped at 10% and the overall cap is 100%, does exhausting the delay sub-cap leave 90% for other claims, or 100%? Usually the former is intended. Frequently the drafting does not say.

Does a Consequential Loss Exclusion Exclude Lost Profits?

Probably not, and this is the single most consequential misunderstanding in the whole liability structure. Both parties typically believe lost profits have been excluded. In many contracts they have not been.

The classification runs through Hadley v Baxendale. As Charles Russell Speechlys sets out, the first limb covers "damages that arise naturally from the breach, in the ordinary course of things," and the second covers "damages that may reasonably be supposed to have been in the contemplation of both parties, at the time they made the contract." A clause excluding "indirect and consequential loss" operates on the second limb. It does not touch the first.

Where does lost profit fall? Not automatically in the second. Charles Russell Speechlys is explicit: "Financial losses, including loss of profit, which one would normally expect to flow from the breach, are likely to be classified as direct loss."

For a power project that is close to decisive. Revenue from selling electricity is not an exotic head of loss arising from special circumstances — it is the entire purpose of the asset and was plainly in contemplation when the EPC contract was signed. Lost generation revenue flowing from a late or underperforming plant looks very much like a first-limb loss.

The case law reinforces how far this reaches. In Croudace Construction, costs that began to "clock up at once" — workforce, plant, equipment — were held to be direct losses notwithstanding a consequential loss exclusion clause.

The drafting answer is itemisation. A generic exclusion is insufficient; the losses must be named. Charles Russell Speechlys points to the Transocean contract, which defined consequential losses to include "loss of business and business interruption, loss of revenue … loss of profit or anticipated profit." Only an enumerated clause of that kind reliably removes the head of loss.

ℹ️ Note: This cuts both ways, and which way depends on which side of the table you sit. A sponsor assuming its lost revenue claim is excluded may have more than it thinks. A contractor assuming it capped its exposure at the LD sub-cap may have considerably less protection than it priced. Neither party benefits from finding out during a dispute.

Why Does the Exclusion Matter More Than the Cap Percentage?

Because an exclusion removes a category of loss entirely, before any cap is applied. A cap limits the amount of a recoverable claim. An exclusion determines whether the claim exists at all.

Consider the order of operations when a project runs 120 days late:

1. Is the loss recoverable in principle? Exclusion clause and exclusive remedy provisions (post 2) are tested first.
2. Is it within the relevant sub-cap? Delay LDs against the delay sub-cap.
3. Is it within the combined LD cap? Competing with any performance claim.
4. Is it within the overall cap? Net of anything carved out.
5. Is it collectable? Security, then covenant strength.

A negotiation that spends its energy on step 3 and none on step 1 has optimised the wrong variable. Moving a sub-cap from 10% to 12% is worth $1.8m on a $90m contract. Establishing whether lost revenue is recoverable at all is worth the difference between a claim and no claim.

How Do the Layers Apply When Several Things Go Wrong at Once?

Rarely cleanly, because the layers were negotiated as if each failure happened in isolation. Projects fail in combination — late, then underperforming, then defective — and the interactions surface only when the second claim arrives.

The sequence that causes the most argument is a project that is simultaneously late and short. Delay LDs accrue during construction. Performance LDs crystallise at testing. If they share a combined cap, the first claim to be quantified consumes headroom the second one needed, and the order in which they happen to be assessed changes the recovery.

Add a defect requiring rework and the picture gets more complex, because rework is usually carved out. Now some claims sit inside the cap, one sits outside it, and whether the outside one accrues toward the ceiling depends on drafting that, per Clifford Chance, is "usually intended" one way and often written the other.

Where Does Insurance Sit in the Stack?

Alongside the cap rather than under it, which makes it the one tier that can reach losses the contractual structure cannot. Clifford Chance lists insurance recoveries — construction and erection all-risks proceeds — among the liabilities typically carved out, meaning amounts recovered under the policies do not consume the contractor's cap.

That separation is the useful part. Every post in this series has ended at the same place: a residual carrying cost that no counterparty owes. Insurance is the only mechanism in the stack that can pay a loss without there being a breach at all.

The question is therefore not whether the project is insured but whether the events that actually cause delay are insured perils. Construction all-risks responds to physical damage. A supplier insolvency, a permitting delay, a queue position lost to a missed milestone, or a contractor that is simply slow are not physical damage, and no policy in the standard package responds to them. The gap between "insured" and "insured against the thing that went wrong" is where sponsors are most often surprised.

Where a delay does arise from an insured physical peril, delay-in-start-up cover can respond to the revenue loss and debt service the project incurs during reinstatement — which is precisely the uncovered carrying cost from post 2. Whether the project carries it, what its indemnity period is, and how its deductible compares to the LD daily rate are three questions worth answering at the same time as the cap negotiation, because they price the same risk from the other side.

The practical check: line the insurance programme up against the delay causes the schedule risk register already identifies, and mark which ones have a policy behind them. Usually fewer than expected.

How Do You Model the Liability Stack in Excel?

Build it as a cascade with three ceilings, and report the binding one. The output that matters is not the total claim — it is which constraint stopped you.

The inputs

Assumptions, labelled as such:

EPC contract price                $90,000,000
Delay: 120 days at $45,000/day    $5,400,000   claim
Performance buy-down NPV          $3,700,000   claim
Defect rectification cost         $4,000,000   claim (carved out)
Delay LD sub-cap                  10%          of contract price
Combined LD cap                   20%          of contract price
Overall liability cap             100%         of contract price
Contractor net worth              $25,000,000
Security posted (LC + retainage)  $11,000,000

Ceiling one — the contractual cascade

Delay_Recoverable   = MIN(5,400,000, 90,000,000 × 10%)         = $5,400,000
Perf_Recoverable    = MIN(3,700,000, 90,000,000 × 10%)         = $3,700,000

Combined_LD_Claim   = 5,400,000 + 3,700,000                     = $9,100,000
Combined_LD_Cap     = 90,000,000 × 20%                          = $18,000,000
Combined_Allowed    = MIN(9,100,000, 18,000,000)                = $9,100,000

Rework              = $4,000,000        (carved out of the cap)
Total_Contractual   = 9,100,000 + 4,000,000                     = $13,100,000
Overall_Cap         = $90,000,000  → not binding

Comfortable at every tier. This is the point at which most analyses stop, and it is the point at which the real constraints begin.

Ceiling two — the exclusion test

Lost_Revenue_Beyond_LDs = 120 days × (65,000 − 45,000)          = $2,400,000

Recoverable_If_Direct    = $2,400,000
Recoverable_If_Excluded  = $0

The uncovered carrying cost from post 2 reappears here as a claim rather than a loss. Whether it is recoverable turns entirely on the exclusion clause's drafting and on whether LDs were expressed as the sole remedy. Model it as a flagged scenario, not a number — the honest output is a range with the drafting question named.

Ceiling three — collectability

Claim               = $13,100,000
Contractual_Cap     = $90,000,000
Collectable_Secured = $11,000,000
Contractor_NW       = $25,000,000

Effective_Recovery  = MIN(Claim, Cap, Security + realistic unsecured)
                    ≈ MIN(13,100,000, 90,000,000, 11,000,000 + ?)

The $90m cap is doing no work at all. The binding constraint is $11m of posted security plus whatever an unsecured claim against a $25m entity is actually worth — which, if the failure that generated the claim also impaired the contractor, may be very little.

The output to report

Binding_Constraint  = "security and covenant strength, not the cap"
Headroom_Illusion   = 90,000,000 − 11,000,000                   = $79,000,000

That last line is the number to put in front of a negotiation. Seventy-nine million dollars of the agreed cap is decorative.

ℹ️ Note: Re-run the cascade after every claim, not once at signature. Sub-caps and the combined cap are consumed cumulatively, so the headroom available to a second claim depends on what the first one took.

To run the full version — the cascade applied to your own sub-cap structure, carve-outs flagged as accruing or non-accruing, and recovery tested against posted security rather than the contractual ceiling — prompt Dezzmond with your liability provisions and claim history.

What Do Lenders Actually Check?

Lenders treat the cap as the least interesting number in the section and go straight to what pays.

• Who signed, and what is behind them? The contracting entity's balance sheet and the parent guarantee's scope.
• How much is posted rather than promised? Security is the only tier that converts to cash without litigation.
• Do carve-outs accrue toward the cap? Explicitly stated, or left to be argued.
• Is the consequential loss exclusion itemised? A generic clause probably does not exclude lost revenue, in either direction.
• Are LDs the sole and exclusive remedy? If so, the exclusion question is largely academic — post 2 covers why.
• How much combined LD headroom remains after a delay claim? The performance claim competes for it.
• Is rework genuinely carved out? Otherwise a contractor can exhaust its cap and decline to fix the plant.

Frequently Asked Questions

What is a typical overall liability cap in an EPC contract?

Often up to 100% of the contract price, though Clifford Chance's position is that caps should be project-specific rather than set by market practice, and that a 100% cap can be illusory on a very large project depending on the contractor's financial capability.

Does a consequential loss exclusion stop me claiming lost revenue?

Not reliably. Loss of profit is not inherently indirect — financial losses that would normally be expected to flow from the breach are likely to be classified as direct, and a generic exclusion operates only on the second Hadley limb. Excluding lost profits requires naming them expressly.

Do liquidated damages count toward the overall cap?

Typically yes — LDs are usually sub-capped within the overall cap rather than sitting outside it. Drawing heavily on delay and performance LDs therefore reduces the headroom available for other claims.

What is usually carved out of the cap?

Third-party liabilities including death and injury, rework costs, insurance proceeds, IP infringement, advance payment repayment, fraud and wilful misconduct, and enforcement costs. Whether recoveries under those heads also count toward the cap is a drafting question that is frequently left unanswered.

Why does the contractor's balance sheet matter more than the cap?

Because the cap sets what is owed and the balance sheet sets what arrives. Beyond posted security, a capped liability is an unsecured claim ranking with other creditors — and the event generating the claim is often the event impairing the contractor.

Closing: Three Ceilings, and the Contract Contains Two

Liability provisions are read as a single number and function as a cascade. A claim has to be recoverable in principle, then within its sub-cap, then within the combined cap, then within the overall cap, then collectable. Each stage can stop it, and the stages most likely to stop it are the first and the last — the two that get the least attention.

The middle tiers are where negotiations concentrate because they are quantified and therefore negotiable. Moving a sub-cap by two percentage points is a visible win. Establishing that lost revenue is a first-limb loss, or that a carve-out does not accrue toward the ceiling, or that the counterparty cannot fund the number it just agreed to — none of those produce a number to report, and all of them are worth more.

This closes the risk allocation trio and, with it, the contracts arc of the series. Across eight posts the shape has been identical every time. The completion dates sit in two documents and drift. The delay LD rate is negotiated rather than derived. The cap exhausts before the longstop arrives. Capacity and PR are guaranteed while annual energy is not. Availability is remedied at the size of a fee. A tariff may not be a change in law. Force majeure stops the clock that was paying you. And here: the cap you agreed is bounded by an exclusion you may have misread and a balance sheet nobody checked.

Every document does what it says. The exposure lives between them, and only the model sees all of them at once.

The series turns next to the financing itself — debt sizing, coverage ratios and the cash waterfall — where the question changes from who owes what to how much the structure can absorb before equity funds it.

Sources: Clifford Chance — Caps and Carve-Outs · Charles Russell Speechlys — Indirect and Consequential Loss Exclusions: Is It Time for Change? · SgurrEnergy — Preparing a Bankable EPC Contract