Solutions · Law Firms

First-pass covenant summary in minutes, ready for associate review

The first pass through a credit agreement is the least valuable hour of an associate's day and the one most likely to miss something. ExactCov produces the extract and the citations so the review starts at the review, and keeps client papers where the engagement letter says they must stay.

Credit agreements (LMA, LSTA)IndenturesIntercreditor agreementsAmendments and restatementsHouse forms and precedentsTerm sheets against drafts
The problem

Every covenant summary is built from scratch, clause by clause.

An associate opens the agreement and starts at the definitions. Precedent comparison means six documents in six windows. Amendments and A&Es are read in isolation, then reconciled by hand. And the tools that could help are ruled out because nobody has read their retention terms.

CitedEvery covenant, basket, definition and carve-out with a verbatim quote and a page, checked by the machine before a person sees it.
ResolvedEvery defined term followed to its leaf, including the hop that leaves the document for the intercreditor.
ComparedHouse styles, sponsor forms and market standards side by side, amendments read against the original.
Retained nowhereZero data retention in writing, self-hosted storage options, and nothing used to train any model.
The extract

What the first pass produces

From 892 pages to a register in 27 minutes, every figure with a quote and a page.

Covenants and tests

The operative provisions, as drafted.

  • Financial covenants with step-downs, cure rights and springing tests
  • Incurrence baskets with grower legs and ratio alternatives
  • Carve-outs, and the carve-outs to the carve-outs

Definitions

The number is six definitions away from the covenant.

  • Every chain resolved hop by hop, median four, up to nine
  • Cross-references and the hop that leaves the document
  • Restated sections read as superseding the original

Review queue

The fields a person should read, and only those.

  • Each quote verified against the page by code
  • The handful the validator cannot verify routed to a reviewer
  • Export to Word and Excel with the citations intact
The first pass

From 892 pages to a register in 27 minutes

892 pages narrow to the covenant and definition pages, then to a register of 47 covenants, 61 tests, 112 baskets, 203 carve-outs and 418 defined terms. Of 1,014 citations, 1,003 verify against the page and 11 go to a reviewer. Those eleven are the eleven an associate reads first.

Read the piece
892 pages, scanned A&R + 4 amendments + ICA + security Where the terms live pp. 187 to 341 + 62 pages of definitions Register covenants47 tests61 baskets112 carve-outs203 defined terms418 27 min1,014 citations 1,003 quote found on page kept, no human needed 11 could not be verified routed to a reviewer the check that matters: is the quoted passage on the cited page, word for word, and is the figure inside it? the 11: a superseded pricing grid, a basket sized by reference to the intercreditor, an add-back cap read the wrong way round
From 892 pages to a register in 27 minutes. Every figure carries a verbatim quote and a page; the machine checks each one, and the eleven it cannot verify are the eleven a person reads.
Where a first pass fails

Four fields that look like numbers and are structures

A threshold that is a step-down schedule with a springing condition. A grower basket with two legs that cross over time. A definition chain four links deep that leaves the document. A restated section that supersedes the original. A first pass returns the number on the page and misses the covenant; the extract has to carry the structure.

Read the piece
Threshold extracted: 5.00x T1T4T6T9 5.00x4.50x from test 6 springing: tested if revolver > 35% drawn · equity cure 2 in 4 Grower basket extracted: 50m closeyear 2year 4 fixed leg 50m 20% of EBITDA greater-of flips here Defined-term chain extracted: one paragraph EBITDA Cons. NI Transactions p. 2 Intercreditor, cl. 4.2 a link leaves the document: store the gap Restated section extracted: both versions s.7.01 originalsuperseded s.7.01 restatedAmendment 4, in force from a definition can change with the covenant text untouched Every field is a structure, not a scalar. Every element carries a quote and a page. What cannot be verified goes to a person.
Four places a first pass returns the number on the page and misses the covenant: a step-down schedule with a springing test, a grower whose legs cross, a definition chain that leaves the document, and a restated section.
From our research

How deep do defined terms go?

A leverage covenant is one line. Resolving it means following Consolidated EBITDA into Consolidated Net Income and onward, a median of four hops and up to nine across 300 agreements, with one branch ending in the intercreditor. Extraction that stops at hop one has read the label, not the test. The extract shows every hop.

Definition chains in 300 agreements
One covenant line, resolved Section 8.01(a): Total Net Leverage Ratio shall not exceed 4.50x hop 1hop 2hop 3hop 4hop 5hop 6 Total Net Leverage Ratio Consolidated Total Net Debt Consolidated EBITDA Consolidated Net Income permitted add-backs, capped Pro Forma Basis, Acquisitions Accounting Principles, frozen GAAP Permitted Priority Debt "as defined in the Intercreditor Agreement" Chain depth, leverage covenants 300 agreements, hops to a number or to GAAP 123456789 median 4, mean 3.4 weighted by term, max 9 38% have a hop that leaves the document
The covenant is one line. The number it tests is six definitions away, and one branch of the chain ends in another agreement entirely. Extraction that stops at hop one has read the label, not the test.
Go deeper

Reading for law firms

Short pieces from the blog and the measurements behind them from Research.

Blog · 3 min

What we found in an 892-page credit agreement

Forty-seven covenants, 61 tests, 112 baskets and 418 defined terms out of one scanned facility in 27 minutes, with 1,003 of 1,014 citations verified by machine. What the AI got wrong, and how we knew.

Read
Blog · 3 min

What document extraction gets wrong

Threshold schedules, basket stacking, defined-term chains, grower baskets, restated sections and cross-references that leave the document. The six failure modes we test against, and what to ask a vendor to show.

Read
Blog · 3 min

The clause is written once. The check happens every month.

A NAV decline trigger takes an hour to negotiate and then lives for a decade in a PDF nobody re-reads. The risk is not the term. It is that nothing connects the term to the calendar.

Read
Blog · 3 min

Where AI belongs in covenant work, and where it must not

Headroom, due dates, exposure and state changes are deterministic code, always. Reading, ranking, flagging and drafting are model-assisted, with a check before anything downstream depends on them. The table, and the reasons.

Read
Research · 3 min

How deep do defined terms go? Definition chains in 300 agreements

A leverage covenant is one line. Resolving it means following Consolidated EBITDA into Consolidated Net Income and onward, a median of four hops and up to nine. Where the chains go, how often they leave the document, and what that does to extraction.

Read
Research · 3 min

Right number, wrong clause: measuring citation faithfulness

Field accuracy is the metric everyone reports. It misses the quadrant that fails an audit: the value is correct and the clause it points to is not. That quadrant was 4.6 percent of our fields before the validator and is 1.1 today, inside a 97.4 percent first pass, and it is invisible to a spot check.

Read
Getting started

Run it in parallel. Don't run a pilot.

The documents are loaded once. What already arrives by email is copied to a mailbox we watch. Nothing changes for the team, and after one cycle the two outputs are compared, disagreement by disagreement.

Week 0A closed matter loaded. Extract compared with the summary the team produced at the time.
Weeks 1 to 4Live matters run on both sides. First-pass extracts and precedent comparisons produced from the register.
Week 4Compare: what the extract found that the summary missed, hours saved per matter, what the reviewer queue contained.
AfterPer matter or per practice group. Client papers stay in your region, or in your estate.

Try it on a closed matter

Bring a credit agreement from a matter that has closed and the summary your team wrote. We show the extract, the citations and the differences.