Technical Due Diligence TDD

What the review actually produces

The output is a haircut with a document attached rather than a verdict. Every optimistic assumption the reviewer strikes reduces the cash flow available for debt service, and through the coverage ratio it reduces the debt the project can raise — so a weak technical package costs gearing, and gearing costs equity return.

Findings sort into three bins: things that must be fixed and evidenced before funding, which become conditions precedent; things that cannot be fixed and are instead reflected in a lower base case or a larger reserve; and things the reviewer flags for the operating years, which reappear as covenants and reporting obligations.

It is not a one-time gate either. The same file reopens at refinancing, at sale, and during construction, where the lender's independent engineer certifies drawdowns against milestones and checks that reserve accounts stay funded. It also reopens whenever the configuration moves: a cell revision, an enclosure change or a different converter takes the equipment off the file its safety evidence and its degradation data were generated on, which is a diligence event even when nobody planned one.

The degradation and augmentation model, input by input

The reviewer does not accept the retention curve; it traces the curve's inputs, and the first of them is the duty. The three written forms of a duty profile are the checklist: the reference duty the vendor's cycle-life data was measured at, the contracted duty in the warranty envelope, and the duty the revenue model actually dispatches.

Where those three disagree, the base case is wrong in a direction nobody has priced — a dispatch plan that runs deeper, hotter, faster or parked at higher resting SOC than the envelope permits is a warranty claim that will fail its envelope check years after the model was signed off.

Then the boundary and the year. A retention curve written against DC nameplate is a different promise from guaranteed energy at the point of interconnection, so the reviewer restates the curve at the bus the offtake pays at and checks the plant still clears in the worst contracted year.

Then the augmentation plan: whether tranche timing is read off the warranted curve rather than the calendar, whether the reserve is funded as a distinct line rather than folded into O&M, and whether the physical provisions exist on the drawings — reserved foundations, feeder positions, spare conduit and duct-bank ways, DC bus and conversion headroom. Reserved bare land with no conduit and no feeder position is a provision on paper only, and cumulative augmentation commonly totals 15-40% of original DC capacity over a 15-20 year term, which is not a line to leave undrawn.

The recurring finding is a linear fit. Two years of state-of-health data extrapolated straight to year 20 flatters the augmentation budget, because degradation is nonlinear with a knee late in life, and the same optimism reappears wherever the first-year step — commonly around 2-4%, before roughly 1-2% per year — has been smoothed into an average. Reviewers re-forecast against measured state of health and against the supplier's own cycle-test data rather than the marketing curve.

The guarantee stack, its definitions, and who stands behind it

Performance promises reduce to three families — retained capacity, round-trip efficiency, availability — occasionally joined by a standalone capacity-maintenance agreement alongside the supply agreement and the LTSA.

The reviewer reads which contract carries each promise, because that decides whose liability caps and termination rights apply, and reads the operating envelope behind each one, because a guarantee is only as wide as the box it is conditional on. Dispatch rights wider than the envelope guarantee a breach of one contract or the other, and a toll permitting more annual cycles than the warranty allows leaves the owner silently carrying the gap.

Availability is where the definition matters more than the percentage. Time-based and capacity-weighted availability are different measurements of the same plant, and excluded hours can either be removed from the denominator or deemed available in the numerator — two treatments that give different answers on identical downtime, in vocabulary borrowed from IEEE Std 762.

Contractual guarantees at utility scale commonly land at 95-98% measured annually, with mature LFP sites targeting 97% plus, and the number only becomes real once converted to hours: about 175 h/yr of unexcused downtime at 98%, about 263 h at 97%, about 438 h at 95%. A reviewer comparing a 97% guarantee under one definition against a 95% guarantee under another may be looking at the same machine. The reliability run that gates commercial operation is a separate instrument again, and passing it says nothing about the operating-year guarantee.

Then the credit, and then the record. A twenty-year guarantee from a thinly capitalised subsidiary is worth what that subsidiary is worth, so the reviewer identifies the signing entity, any parent guarantee or warranty insurance, escrowed spares and escrowed firmware, BMS configuration and access credentials, and the assignment and step-in rights that decide whether the guarantee survives a refinancing or a sale.

On an operating asset the record replaces the promise: capacity-test results trended against the guaranteed curve test over test rather than read as pass or fail, how availability was actually computed and which hours were excluded, BMS and EMS logs checked against the warranty envelope axis by axis, alarm and incident history, spares consumed, open claims, and the firmware campaign history.

Interconnection and curtailment exposure

Lenders generally want the interconnection agreement executed and the system-impact and facilities studies complete before financial close, and treat interconnection timing as one of the largest schedule risks in the model.

The reviewer reads the facilities-and-ownership exhibit against the interconnection one-line as two documents that have to agree, checks that the point of interconnection, the ownership boundary and the metering point are each defined rather than assumed to coincide, and reads network-upgrade cost and repayment treatment against the tariff governing this queue position rather than against the pro forma, since individual tariffs carry approved deviations.

Then the obligations and the revenue exposure that follow from them. FERC Order No. 827 revised the pro forma LGIA and SGIA, so its reactive-power requirement binds newly interconnecting non-synchronous resources rather than the existing fleet — which makes the governing agreement version, not the order's date, the thing to establish for a given project.

Curtailment gets its own line: whether the interconnection is capacity-limited, whether dispatch rights leave the plant energy-limited, what the revenue model assumed about hours it can neither charge nor discharge, and whether the offtake pays on availability regardless of dispatch. And a question that outlives close — who validates the plant model against the as-built for grid-code compliance testing, and who reruns it after a firmware change.

The fire-code file, the AHJ position and insurability

This workstream is almost entirely evidence-matching, and it is where a confident-looking package most often comes apart. UL 9540 is the safety certification of the assembled ESS product. UL 9540A is the separate fire-propagation test method, run at up to four levels — cell, module, unit, installation — that produces characterisation data rather than a pass.

The reviewer checks that the report in the file is the full report, at the level that supports the spacing actually built, on the cell part number and revision the factory is shipping, since a report on one cell does not carry to its successor in the same container. The hazard mitigation analysis is where that data meets this site's layout and exposures, and an AHJ-accepted HMA is a diligence gate alongside the UL 9540 listing and the capacity warranty.

Insurers underwrite from largely the same file and can require more than the code minimum — FM Global Data Sheet 5-33 is the visible example — so a fully NFPA 855-compliant, AHJ-approved site can still meet limited capacity, large deductibles or exclusions.

The reviewer therefore checks the probable-maximum-loss story the layout supports, whether the explosion-control strategy on the drawings is the one the analysis claimed, and whether the business-interruption waiting period lines up with availability obligations under the offtake, since a project can owe availability damages well before the policy responds. Insurance is a condition precedent in most financings, which puts this workstream in a different category from the rest: a gap here does not get priced into the base case, it stops the close.