Evidencing Beginning of Construction: The File You Assemble Years Before Anyone Asks
A beginning-of-construction position is not established on the day construction begins. It is established years later, in a data room, by whatever records happen to exist. The physical work either was or was not of a significant nature — but the question is decided on documents, and the documents are created by people who are not thinking about a tax controversy four years away.
The gap between having a position and being able to prove it is where most of the cost sits. A tax equity investor will not fund against a position it cannot diligence. An insurer will not cover one it cannot underwrite. And a $75m credit, once you add interest, penalties, fees and the gross-up, can require a policy limit approaching $147m.
This is the third and last post on beginning of construction. You will get what diligence actually examines, what makes a contract binding for these purposes and why that interacts with your EPC liquidated damages cap, what evidence supports each pathway, what tax insurance covers and excludes, and how to size a policy against the real exposure rather than the credit.
ℹ️ Note: This describes how these mechanisms work in practice. It is not tax, legal or insurance advice, and the underlying guidance is unsettled following the June 2026 vacatur discussed in the earlier posts.
What Is Actually Being Diligenced?
Two things: the date, and whether progress continued from it. Everything else is evidence supporting one or the other.
That sounds narrow and is not. Establishing the date requires proving either that physical work of a significant nature occurred — a qualitative judgement — or that 5% of total facility cost was paid or incurred, which requires proving the spending, the delivery, and the eventual total cost. Establishing continuity requires a construction record spanning up to four years.
The mindset that produces a defensible file is adversarial rather than administrative. The question is not "did we begin construction?" It is "if a reader with no context and an incentive to disagree examines this in 2031, what will they conclude?" Records that satisfy the first question routinely fail the second, because they were written by people who already knew the answer.
What Makes a Contract Binding for These Purposes?
Two requirements, and the second one has a trap in it that connects directly to the contracts series.
A written contract is binding only if it is enforceable under local law against the taxpayer or a predecessor, and does not limit damages to a specified amount — for example through a liquidated damages provision.
That second requirement would exclude almost every construction contract ever written, since nearly all of them cap damages. Hence the exception: a contractual provision that limits damages to an amount equal to at least five percent of the total contract price will not be treated as limiting damages to a specified amount.
Read that against post 2 of the contracts series. Market practice puts delay liquidated damages sub-caps at roughly 10–15% of contract price and combined LD caps around 20–25%, all comfortably above the five percent threshold. So the standard package passes.
The risk is in the non-standard package. A heavily negotiated contract where a sponsor traded a lower LD cap for a better price, a small equipment supply agreement with a modest liability limit, or an aggregate cap that a first claim has already eroded — any of those can put a contract below five percent of its price, and a contract that is not binding does not support a beginning-of-construction position built on work performed under it.
This is a cross-check almost nobody runs, because the two questions live in different teams. The commercial team negotiates the cap. The tax team relies on the contract. The five percent floor is the point where those two conversations should have met.
ℹ️ Note: When safe harbour equipment is ordered under a supply agreement, check the damages limitation in that agreement specifically — not the EPC. The supply contract is the one the beginning-of-construction position actually rests on, and it is usually the one with the thinnest liability provisions.
Whose Work Counts?
Work performed by the taxpayer, and work performed for the taxpayer by other persons under a binding written contract — provided the contract was entered into prior to the manufacture, construction or production of the property or its components.
The sequencing requirement is absolute and is the most common documentary failure. A contract signed after manufacturing began does not pull that manufacturing into the position, however genuine the work was. Purchase orders issued against a master agreement, verbal instructions later papered, and amendments that expand scope after work started all create the same problem: work that happened outside a binding contract that predated it.
The evidence needed is therefore dated, not merely existent. Executed signature pages with dates, delivery records, and manufacturing progress reports that post-date the contract. A file containing a contract and a photograph of a completed transformer proves less than one containing the contract, the production schedule, and dated progress photographs showing the sequence.
What Does "Paid or Incurred" Actually Require?
More than a wire transfer. The 5% safe harbour is satisfied when a taxpayer pays or incurs five percent of total project cost, and for an accrual-method taxpayer — which most project companies are — "incurred" is a three-part test rather than a payment date.
A liability is incurred when all three are true:
- all the events have occurred that establish the fact of the liability
- the amount of the liability can be determined with reasonable accuracy
- economic performance has occurred with respect to the liability
The third is where safe harbour equipment orders live or die. Where property is provided to a taxpayer — as under a purchase agreement for modules, inverters or transformers — economic performance occurs as the property is provided, and property is provided when it is delivered or accepted, or when title passes.
So paying for equipment does not, by itself, incur the cost. Delivery or the passing of title does.
The 3½ month rule
There is a well-established exception, and it is the mechanism most safe harbour orders rely on. A taxpayer may treat property as provided when it pays the provider, so long as it can reasonably expect the provider to deliver within 3½ months — 105 calendar days — after the date of payment.
Mintz covers the guidance extending relief around this rule, and the detail that catches people is in the measurement: the 105 days run from the date of payment, not from year end. A December payment expecting a February delivery is comfortable. A December payment against a supplier quoting "Q2" is not.
Where it fails
The failure mode is a delivery that slips. A developer pays in late December to establish a position before a deadline, reasonably expecting delivery in February. The supplier slips to May. The payment was real, the intention was genuine, and the cost was not incurred in the year claimed — which can move the beginning-of-construction date into the following year, with all the calendar-year consequences the previous post described.
Three practical points follow.
"Reasonably expect" is judged at payment, not in hindsight. A documented delivery schedule from the supplier at the time of payment is the evidence that supports the expectation. A later email explaining the delay does not.
Title terms are worth reading. Where title passes on shipment rather than on delivery, economic performance may occur earlier than the physical arrival suggests — which can help. Where title passes only on final acceptance at site, it occurs later, which can hurt.
Partial deliveries complicate the arithmetic. If half an order arrives inside 105 days and half does not, the incurred amount is not the invoice total. The 5% test is applied to what was actually incurred, which means the cushion sizing from the first post in this series should be run on delivered value, not on ordered value.
ℹ️ Note: Track safe harbour equipment on a delivery register rather than a payment register. The two diverge exactly when it matters, and a position built from accounts payable records will overstate incurred cost by whatever is still sitting on a ship.
What Evidence Supports Each Pathway?
They fail differently, so they need different files.
| Physical work position | 5% safe harbour position | |
|---|---|---|
| Core evidence | Dated work records, progress photographs, manufacturing reports | Invoices, proof of payment, delivery and title documentation |
| Contract evidence | Binding contract predating the work | Binding contract predating the payment |
| The weak point | Whether the work was "significant in nature" | Whether total facility cost stayed low enough |
| Third-party support | Independent engineer confirmation | Cost accountant reconciliation |
| Fails because | Records were not kept contemporaneously | Costs overran and the denominator moved |
The recurring diligence finding on the physical work side is reconstruction. A sponsor certain that qualifying work occurred assembles the proof afterwards — from memory, from invoices that show payment rather than work, from photographs with no metadata. The position may well be correct and it is materially harder to insure.
On the safe harbour side the recurring finding is the cost question covered in the first post of this series: the spending is documented perfectly and the denominator grew.
What Does the Tax Opinion Do?
It converts a legal judgement into a condition precedent. Tax equity investors engage outside counsel to opine that the tax benefits "should" or "will" be respected, and those opinions sit as conditions in the tax equity documents alongside, increasingly, tax credit insurance.
The distinction between opinion levels is commercial rather than academic. A "will" opinion is close to certainty and is rare on judgemental questions. "Should" indicates a high level of confidence and is the working standard for most positions. Anything weaker generally means the risk has to be allocated elsewhere — through indemnity, insurance, or a price adjustment — rather than absorbed.
Which is why the evidentiary file matters more than the underlying merits. Counsel opines on facts it can verify. A strong position with a thin file produces a weaker opinion than a moderate position with a complete one.
What Can Actually Be Insured?
More than most sponsors expect, including the two positions this series has been describing. Marsh lists the insurable positions, and two are directly on point:
- Begun construction — confirming that "onsite or offsite physical works are of a significant nature"
- Continuity and delay — providing "assurance that the facts and circumstances satisfy the continuous efforts requirement for tax credit eligibility notwithstanding COD is beyond the continuity safe harbor period"
That second one is the answer to the problem the previous post ended on. A project that will miss the four-year safe harbour is not left arguing facts and circumstances unaided; the argument itself can be insured.
The list extends further: structural coverage backstopping "that the allocation of tax and cash attributes between parties will be respected as anticipated in the financial model," qualified basis coverage confirming "the fair market value used to compute ITCs will be respected by the IRS," and repowering coverage underwriting "both the (re)qualification for tax credits and the 80/20 appraisal."
Crucially, coverage extends past the credit itself to "penalties and interest levied by taxing authorities," defence costs, and "a gross-up where the receipt of payment under the policy is taxable." Marsh describes a typical three-week timeline from initial discussion to binding.
What Does Insurance Not Cover?
The things a sponsor most wants covered, in some cases. Norton Rose Fulbright sets out the standard exclusions:
Legislative and regulatory change. Insurers "typically do not insure changes in the Internal Revenue Code or the Treasury regulations." That is a significant carve-out in a market where the guidance has changed twice in a year — the vacatur risk described in the first post of this series is precisely the kind of thing a policy will not carry.
Seller covenant breaches. If owners "abscond to Brazil and never file the election with the IRS to transfer the credits, the insurer would not cover that risk due to the breach of covenant."
Voluntary transfers by the seller, where the seller procured the policy.
Gaps around representations. Where a seller makes representations in the policy itself, "there could be a gap in coverage" if those prove inaccurate.
Norton Rose's broader point is that insurance does not remove the diligence obligation: buyers must analyse "the representations the insurer is requiring and what events are excluded from coverage and the level of coverage versus the potential exposure," and "still need to do some diligence and be sure they are transacting with trustworthy sellers."
Insurance also tends to substitute for indemnities rather than supplement them — some sellers "may prefer to not expose their balance sheets to tax credit indemnities," so the policy replaces the covenant rather than backing it up. That matters when sizing: if the policy is the only recourse, its limit is the recovery ceiling.
How Do You Size the Policy in Excel?
Against the total exposure, not against the credit. Norton Rose is explicit that policies "have caps on the payout for a claim," requiring right-sizing "to account for interest, penalties, and professional fees beyond the credit amount itself." The arithmetic is not intuitive.
The inputs
Assumptions, labelled as such:
Eligible basis $250,000,000
ITC rate 30%
Credit at risk $75,000,000
Years to expected resolution 4
IRS underpayment interest rate 7.0% (assumption)
Accuracy-related penalty rate 20%
Professional and defence costs $3,000,000 (assumption)
Marginal tax rate on policy proceeds 21%
Premium rate 2.5% of limit
Building the exposure
Credit_At_Risk = $75,000,000
Interest = 75,000,000 × ((1 + 0.07)^4 − 1) = $23,308,000
Penalty = 75,000,000 × 20% = $15,000,000
Costs = $3,000,000
Gross_Exposure = 75.0 + 23.3 + 15.0 + 3.0 = $116,308,000
The credit is only 64% of the exposure. Interest alone adds nearly a third again, and it compounds for as long as the controversy runs — which makes the assumed resolution period one of the most sensitive inputs in the calculation.
The gross-up
Required_Limit = Gross_Exposure / (1 − Tax_Rate)
= 116,308,000 / 0.79 = $147,225,000
Because a payment under the policy is itself taxable, the limit has to be larger than the loss it covers. A $75m credit needs roughly $147m of limit — just under double.
The premium
Premium = 147,225,000 × 2.5% = $3,681,000
As_%_of_Credit = 3,681,000 / 75,000,000 = 4.9%
Norton Rose puts premiums at "two to three percent of the maximum insurance payout," rising "if the facts are hairy" — which, for a position resting on contested guidance or a reconstructed evidentiary file, is exactly the situation.
The sensitivity that matters most
Run years-to-resolution down the rows and interest rate across the columns, with required limit in the cells:
Resolution in 2 years → Interest $10.9m → Limit $131.5m → Premium $3.29m
Resolution in 4 years → Interest $23.3m → Limit $147.2m → Premium $3.68m
Resolution in 7 years → Interest $45.5m → Limit $175.3m → Premium $4.38m
Nothing about the tax position changed across those rows. The only variable is how long a dispute takes, and it moves the required limit by $44m. Under-sizing a policy on an optimistic resolution assumption is the most common way a covered exposure turns out to be partially uncovered.
ℹ️ Note: Model the policy limit as a cap on recovery, not as a certainty. Where insurance has replaced a sponsor indemnity rather than supplementing it, anything above the limit has no counterparty at all — the same structure as the third ceiling in the liability caps post.
To run the full version — exposure built from your own basis and credit rate, interest compounded to a probabilistic resolution date, and the limit tested against both an insurance-only and an indemnity-plus-insurance structure — prompt Dezzmond with your credit position and policy terms.
What Do Lenders and Tax Equity Actually Check?
- Is the contract that supports the position binding? Enforceable under local law, with any damages limitation at or above five percent of contract price.
- Did the contract predate the work or the payment? Dated signature pages, not just executed ones.
- Are the physical work records contemporaneous? Reconstruction is the most common finding and the hardest to insure.
- What opinion level did counsel reach — "will" or "should"? And what is allocated elsewhere if it is weaker.
- Is there a policy, and what does it exclude? Code and regulation changes are typically outside coverage.
- Is the limit sized for interest, penalties, costs and the gross-up? The credit is usually around two-thirds of the exposure.
- Does the policy replace or supplement the sponsor indemnity? If it replaces it, the limit is the ceiling.
Frequently Asked Questions
What makes a written contract binding for beginning-of-construction purposes?
It must be enforceable under local law against the taxpayer and must not limit damages to a specified amount. A provision limiting damages to at least five percent of the total contract price is not treated as a limitation.
Does work by a supplier count toward beginning of construction?
Yes, where it is performed under a binding written contract entered into before the manufacture or production began. A contract signed after work started does not bring that work into the position.
Can a beginning-of-construction position be insured?
Yes. Insurers underwrite "begun construction" positions confirming that onsite or offsite physical work was of a significant nature, and separately insure continuity where commercial operation falls beyond the safe harbour period.
What does tax credit insurance typically exclude?
Changes to the Internal Revenue Code or Treasury regulations, seller covenant breaches, and — for seller-procured policies — voluntary transfers by the seller. Coverage gaps can also arise around representations made in the policy itself.
Why does a policy need to be larger than the credit?
Because the exposure includes IRS interest, penalties, and professional fees, and because a payment under the policy is itself taxable. Grossed up, a $75m credit can require a limit close to $147m.
Closing: The Position Is the File
Across three posts this series has treated beginning of construction as a legal question. It is not, in practice. The law sets the test; the file decides the outcome.
That reframing changes what a development team should do. The physical work test rewards contemporaneous records — dated photographs, work logs, manufacturing schedules — created by people who will never see the tax opinion. The 5% safe harbour rewards a cushion sized against a cost distribution rather than a budget. Both reward a binding contract signed before the work, with a damages provision above five percent of its price.
None of that is expensive at the time. All of it is impossible to create retrospectively, which is why the cost of a thin file shows up as a weaker opinion, a higher premium, or a credit the model assumed and the structure could not fund.
The next post opens the second sub-series in policy and regulation: the bonus adders, starting with domestic content — the percentage thresholds, the safe harbour cost tables, and where projects fail the test.
Sources: Marsh — Powering Renewable Energy Projects Via Tax Insurance · Norton Rose Fulbright — Tax Credit Insurance Mitigates Risk but Due Diligence Is Still Necessary · IRS Notice 2018-59 — Beginning of Construction for the Investment Tax Credit under Section 48 · Mintz — Treasury Extends Continuity Safe Harbor and Provides Safe Harbor for the 3.5 Month Rule