A ratio is only as good as its inputs
A leverage ratio, a DSCR, an interest coverage test — every covenant ratio a credit agreement defines is ultimately a formula applied to a small set of inputs pulled from a company's financial statements. Get the formula right but the inputs wrong, even slightly, and the ratio reported to a lender doesn't reflect reality, regardless of how carefully the math itself was done.
Most of the risk in a covenant calculation lives here, in the inputs, not in the arithmetic. A total that's off by a rounding difference, an add-back applied one quarter and forgotten the next, an interest figure that includes a fee it shouldn't — each looks small in isolation and each can shift a ratio enough to matter.
Why this isn't a simple lookup
Financial statements aren't laid out the way a covenant schedule needs them. EBITDA isn't a single line on an income statement — it's assembled from operating income, plus depreciation and amortization, plus whatever specific add-backs the credit agreement allows, each of which might sit on a different page or in a different footnote.
Doing that assembly by hand, correctly, every quarter, for every add-back the agreement permits, is exactly the kind of repetitive precision work that's easy to get right once and hard to get right identically twelve times in a row.
What consistent extraction actually means
Consistent extraction means the same line item is found and read the same way every period, even when a financial statement's exact layout shifts slightly from quarter to quarter — a new line added, a subtotal repositioned, a footnote renumbered. The specific figures matter less than whether this quarter's number was found using the same logic as last quarter's.
That consistency is what makes a covenant trend meaningful. A leverage ratio that looks like it's improving because an add-back quietly stopped being included isn't actually improving — it's a change in method disguised as a change in performance, and it's exactly the kind of inconsistency a lender's own review is built to catch.
Why add-backs are the riskiest input
Every credit agreement that defines EBITDA also defines what can be added back — non-cash charges, one-time restructuring costs, sometimes pro forma adjustments for a recent acquisition. These add-backs are where the largest swings in a calculated ratio usually originate, and where inconsistency is most likely to slip in unnoticed.
An add-back claimed once, for a genuine one-time event, is straightforward. An add-back claimed every quarter for something that looks less one-time each time it recurs is the kind of pattern a lender's credit team is trained to notice — which is exactly why each add-back, once identified in the source financials, is tracked period over period rather than treated as a one-off entry.
Why the same word means different things
“EBITDA” sounds like a fixed accounting term, and in a general sense it is — earnings before interest, tax, depreciation and amortization. In practice, every credit agreement layers its own permitted add-backs on top of that base definition, so “EBITDA” in one facility can be a meaningfully different number from “EBITDA” in another, calculated from the exact same underlying financial statements.
“Debt” carries the same ambiguity. Some agreements count only funded debt; others fold in capital lease obligations, letters of credit outstanding, or specific guarantees. Reading the raw figures a statement reports is the same task regardless of which definition applies — what changes is which figures get summed into which final number, and that mapping has to follow your specific agreement, not a generic template.
| Term | Where the variation usually shows up |
|---|---|
| EBITDA | Which add-backs are permitted — stock comp, restructuring, pro forma acquisition effects |
| Debt | Whether capital leases, guarantees and letters of credit are included |
| Fixed charges | Whether scheduled principal, taxes and capex are all counted, or only interest |
| Cash flow available for debt service | Whether working capital changes and maintenance capex are subtracted |
This is exactly why extraction and calculation are kept as two separate steps here. Getting the raw figures right, consistently, is a document-reading problem this feature is built for. Applying your agreement's specific definition to those figures is a legal-and-financial judgment call that belongs with your team — conflating the two is how a generic tool ends up quietly wrong for a specific facility.
What gets read
| Input | Typical source |
|---|---|
| Operating income, depreciation, amortization | Income statement |
| Identified add-backs and one-time items | Income statement notes or management schedule |
| Total funded debt | Balance sheet or debt schedule |
| Interest expense, scheduled principal | Income statement and debt schedule |
How it works
Upload the period's financial statements
Income statement, balance sheet, and any supporting debt schedule.
Each input is extracted
EBITDA components, debt balances and interest figures pulled with a confidence score.
Compared against the prior period
Add-backs and methods checked for consistency, not just presence.
Inconsistencies flagged
A figure that doesn't match the prior period's method is marked for review.
Exported
Excel, CSV or JSON, with every input traceable to its source statement.
An EBITDA figure, traced
A company reports operating income of $2.4M for the quarter. Depreciation and amortization add $310K. A one-time legal settlement, explicitly permitted as an add-back under the credit agreement, adds another $180K — the same add-back category claimed in the prior quarter for a different, also one-time, item.
| Component | Amount |
|---|---|
| Operating income | $2,400,000 |
| Depreciation and amortization | $310,000 |
| Permitted add-back (one-time legal settlement) | $180,000 |
| EBITDA (before agreement-specific adjustments) | $2,890,000 |
Because the add-back category was used in two consecutive quarters for two different one-time items, it's flagged — not as an error, but as a pattern worth a second look before the certificate goes out, since a lender reviewing two consecutive “one-time” add-backs is likely to ask the same question.
Trailing-twelve-month figures
Many covenants — particularly leverage and coverage ratios — are tested on a trailing-twelve-month basis rather than a single quarter, smoothing out seasonal swings that a single-quarter snapshot would exaggerate. Building this figure means extracting the same input from each of the last four quarters and summing them.
A single misextracted quarter anywhere in that trailing calculation throws off the whole twelve-month figure, which is exactly why each quarter feeding a trailing total is extracted with the same rigor as the current period, not treated as a lower-stakes historical lookup.
Manual vs. automatic
| Manual | Automatic |
|---|---|
| Add-backs tracked in memory or a loose note | Every add-back tracked and compared period over period |
| Trailing figures rebuilt by hand each quarter | Four quarters extracted and summed automatically |
| Inconsistency discovered at the lender's review | Inconsistency flagged before the certificate is submitted |
| Redone by hand for a second facility's definition | Each facility's specific inputs tracked independently |
From one facility to a full credit stack
A single term loan with one covenant is a manageable manual exercise. A company with several facilities, each defining EBITDA slightly differently, multiplies the number of input sets that need extracting and tracking every quarter — without multiplying the time available to do it.
Extracting inputs the same way regardless of how many facilities are in place keeps the per-input effort flat as the credit stack grows, with each facility's specific definitions still applied correctly to its own set of figures.
Who uses this
CFOs and finance teams
Covenant ratio inputs extracted the same way every quarter, without a manual reassembly each time.
Companies with multiple credit facilities
Each facility's specific inputs extracted and tracked independently.
Outside accountants preparing certificates
A verified, traceable input set received rather than raw statements to work through.
Finance teams managing add-back scrutiny
Every add-back tracked period over period, flagged if the pattern looks inconsistent.
Edge cases worth knowing
A pro forma adjustment for a recent acquisition — treating an acquired business as though it had been owned for the full trailing period — is one of the more complex add-back categories, since it requires figures from the acquired entity's own financials, not just the parent company's. These are flagged for manual confirmation rather than extracted automatically, given how much judgment they typically require.
A change in accounting method — a shift from cash to accrual recognition for a specific revenue stream, for instance — can shift reported figures in ways that look like a covenant-relevant change but actually reflect a bookkeeping change. Flagging a sudden shift in a line item that's normally stable helps surface this distinction before it's mistaken for a real trend.
Why consistency matters more than speed
A covenant calculation done quickly but inconsistently from quarter to quarter is worse than one done slowly but the same way every time — an inconsistent calculation produces a ratio trend that doesn't actually reflect the business, which is precisely the kind of gap a lender's own review exists to find.
Extracting the same inputs the same way every period, with every deviation flagged rather than silently absorbed, is what keeps the reported trend trustworthy — to your own finance team as much as to the lender reading it.
What a confidence score actually tells you
Every extracted input carries a confidence score, and it's worth being specific about what that score does and doesn't mean. A high score means the figure was read from a clear, well positioned printed number with no ambiguity about which line item it belongs to — not that the figure is necessarily the right one for your credit agreement's specific definition.
A lower score flags exactly where a human look is worth the time: a total obscured by a footnote marker, a figure split across a page break, a column header that doesn't match the expected pattern. Reviewing the handful of flagged fields each quarter, rather than re-checking every figure from scratch, is what makes the whole process fast without becoming careless.
What this doesn't do
Doesn't apply your credit agreement's EBITDA definition
It extracts the raw inputs and identifies add-backs; applying your agreement's specific formula stays your calculation.
Doesn't judge whether an add-back is permitted
It flags an add-back for review; whether your agreement actually allows it is a legal and financial question for your team.
Doesn't build pro forma acquisition adjustments automatically
These require judgment and figures from outside the immediate financials — flagged for manual handling.
Doesn't replace your accounting or legal counsel
It surfaces figures and inconsistencies; interpreting them stays with your team.
Getting your first quarter extracted
There's no configuration step that has to happen before the first extraction — no template to build, no chart of accounts to map, no add-back categories to pre-define. Upload the current period's financial statements and the extraction runs against them directly, surfacing every input a standard covenant calculation typically needs.
The first quarter is also the natural point to note the reasoning behind any add-backs your specific credit agreement permits — since there's no prior period yet to compare against, this is the baseline every future quarter's consistency check will measure against, and it's worth a few extra minutes to get that starting point right.
Most finance teams run their first extraction against a recent, already-familiar quarter before relying on it for the actual live certificate — a useful way to sanity-check that every figure lines up with what you'd expect from a statement you already know well, before trusting it on a period you haven't reviewed by hand yet.
That first-run check also tends to be the moment a finance team notices a line item their specific statement format handles slightly differently than expected — a subtotal labeled unusually, an add-back note buried in a footnote rather than the main body — worth confirming once, early, rather than discovering it the quarter it actually matters.
Security and privacy
Uploads are encrypted with TLS from end to end.
Processing runs on infrastructure with SOC 2-aligned controls.
Original documents are deleted shortly after processing.
Nothing you upload is ever used to train AI models.
For financial data that reveals the full shape of a company's performance and debt, that matters — details are on the security page.