Establish what is actually known
Contracts, tariffs, resource data, cost base, the term sheet as it stands today. We separate what is signed from what is forecast, and price the gap between them.
Method
A model you can follow.
Identified assumptions, traceable calculations and visible checks. The DECOMP standard gives the file your team receives a shared structure.
Illustrative solar asset · 10 contracted years, 5 merchant years
Invented data · Simplified modelWhen the contract ends.
Minimum debt service coverage1.23x
All 15 years pass.
Year 8 of 15
Maintenance in year 8 leaves the least coverage headroom.
Maintenance in year 8 leaves the least coverage headroom.
Threshold 1.20x
| Scenario | Minimum debt service coverage | Critical year |
|---|---|---|
| Base case | 1.23x | 8 |
| Price −15% | 1.16x | 15 |
| Price −25% | 0.90x | 15 |
Debt: €64.6m. Repayments stay fixed when the price changes. Tax, working capital and reserves are excluded, among other items.
Open the full modelHow an engagement is shaped
Assumptions, calculation, output. In that order, every time.
It is how a good model is built, so it is how we run the work. You always know which of the three we are in, and what leaves the room at the end of it.
01 — Assumptions
02 — Calculation
Build the thing that answers the question
A model sized to the decision in front of you. Debt sizing, coverage ratios, sensitivities, and the scenarios a credit committee will run whether or not you ran them first.
03 — Output
Hand over something you can operate
The model, the documentation and a working session with the people who will use it. If you need us to open the file every time a number changes, we have failed.
The cost of everyone inventing their own conventions
There is no shortage of talented modellers. There is a shortage of models two talented modellers can hand between them without a week of orientation. Every analyst who invents a personal layout adds a private language to a file that other people will have to read under time pressure, usually while a transaction is live.
A standard removes that cost by making the work predictable. The assumptions live here, the calculation runs this way, the checks say whether it worked. Everything genuinely difficult about the transaction is still difficult. Finding your way around the file is not.
DECOMP is the standard we build to. Decomp Finance maintains it and Decomp Academy teaches it. You can read what it requires, learn it without us, and hold our work against it.
The rules
Four rules that hold under deadline
There are more. These are the ones abandoned first under pressure, and the ones that cost the most when they are.
R.01
Simple formulas
A formula that needs explaining is a formula somebody will break. Six nested functions almost always means a missing intermediate row.
R.02
One input, one place
Every assumption entered once, in one location, with its source beside it. Numbers repeated across a workbook are how a file ends up with two versions of the truth.
R.03
Calculation is not presentation
Working blocks calculate. Output sheets present. When the two mix, changing a chart starts changing a result.
R.04
Visible checks
A model that cannot tell you it is broken is worse than one that is obviously broken. Every check row resolves to a visible pass or fail, including a staleness check on any solver block.
Proof
How you know it worked
Principles are easy to publish and hard to verify. These are the tests we apply before anything leaves, and the ones you can run yourself on the delivered file.
- Open the file cold and find any assumption within thirty seconds
- Change one input and trace its effect to the outputs without help
- Every check row visible on one screen, all reading pass
The design target is that a single-asset model built to this standard is reviewable in a day. A multi-tranche, multi-currency structure with hedging takes longer, and anyone who tells you otherwise has not reviewed one.
- No formula in the workbook longer than one readable line
- No assumption hard-coded inside a calculation formula
- Every row internally consistent across its full period range
- Sensitivities that produce a different answer, and only through the switch
Row-consistency testshow the working
For each calculation row r spanning periods t = 1..n:
the formula in (r, t) must be identical to (r, t−1) shifted by one
column, for every t > 1.
No exceptions. Where behaviour genuinely changes — end of construction,
end of the PPA, a refinancing date, tax loss expiry — the change belongs
in a timing flag row that the single consistent formula multiplies by.
A row with an exception is a row someone will copy across later.
A row with a flag is a row that stays correct when they do.
The test that finds the expensive defect
One inconsistent cell in a twenty-five-year row is the defect least likely to be found by reading the model and most likely to be material. Row-consistency testing is mechanical, takes minutes, and is the first thing we run on any file, including our own.
Handover
The file, documentation and handover
Every model ships with three things: the workbook, documentation covering structure, assumptions and sources, and a working session with the people who will operate it.
That last part is not a formality. It is where you find out whether the model is genuinely usable by someone who did not build it. If it turns out not to be, that is our problem to fix.
Notes
N.01 · 2026
Two structures, one minimum DSCR, different risk
The minimum tells you almost nothing on its own. What matters is how much of the profile sits near the covenant, and for how long.
N.02 · 2026
The last five years decide the case
Contracted years flatter every structure. The argument begins the day the offtake ends, and most models are thinnest exactly there.
N.03 · 2026
Overwritten cells and model review
Not elegant errors of financial logic. Overwritten periods, inconsistent definitions, and a rounding plug that has become material.
Next step
The method, applied to your project.
Tell us what the model needs to resolve and who will use it.