Know what the package allows before you pitch it, and what the book is doing after you close it
Syndicate and DCM desks need the full package from an OM or credit agreement in minutes, and a defensible view of what is market. Credit-risk teams need every test on every name recomputed the moment data lands. One register serves both, with the same citations.
The document is read hard once. Every question afterwards goes back to the PDF.
A capacity analysis is rebuilt by hand for each pitch from a precedent that was itself built by hand. Investor diligence is answered from memory while the deal is live. After close, the compliance certificate is filed and read weeks later, and a months-old ratio is quietly reused because nothing flagged it as stale.
What the desk gets from each document
Extracted at origination, kept for the life of the facility, shared between the desk that structured it and the team that monitors it.
Capacity and flexibility
The incurrence package as a table, not a memo.
- Debt and lien baskets with used and remaining portions
- Restricted payments: builder basket, carve-outs, ratio tests
- Portability, change of control, MFN and margin ratchets
Market position
Every term placed against what the bank has already closed.
- General debt basket as a share of EBITDA across your own deals
- Median and interquartile range for each term
- A new draft's ask flagged outside the range, comparables one click away
The whole book
For credit risk: thousands of names, most of them quiet.
- Every reporting obligation on one calendar, chased when late
- Every test recomputed on arrival, headroom and trend shown
- Stale figures shown as stale, never silently reused
"What's market" is sitting in documents you have already read
Market terms are a distribution, not a recollection. Every closed deal's general debt basket, RP capacity and MFN threshold on one axis, the range and median marked, and the new draft's ask placed against them. Pitch books, market updates and investor diligence answered from the bank's own precedent, cited.
Read the pieceStale data is a control failure, even without a breach
A quarterly review slips from April to July. Trading continues against a five-month-old figure. The July review finds no breach, and the book records nothing. The consequence was three months of exposure taken against a number the control was meant to refresh. The register makes stale a state, not an age.
Read the pieceOpen-weights models against the frontier
Banks that need extraction to run inside their own estate ask whether an open model is good enough. Model alone, the frontier leads by five points. Give every model the same section tree, definitions index and candidate spans first, and the two larger open models land within a point and a half of the frontier. The architecture matters more than the model.
Open-weights vs frontier on covenant extractionReading for banks
Short pieces from the blog and the measurements behind them from Research.
"What's market" is sitting in documents you have already read
Every desk reconstructs market terms from memory and a few recent deals. The distribution of every basket, cure and blocker across your own closed deals is a query once the corpus is in a register.
Read Blog · 3 minHow much can they actually raise?
Incurrence capacity is the question that keeps returning and never has a stored answer. What a capacity table looks like, why it needs a history, and why it is a credit signal months ahead of the maintenance tests.
Read Blog · 3 minStale data is a control failure, even without a breach
A review that slipped is not harmless because nothing broke. The desk was pricing exposure off old information. Why staleness needs its own population, its own metrics and its own owner in the committee pack.
Read Blog · 3 minThe five things a credit exec should see before 9am
Due soon, overdue, blocked and by whom, near covenant, deteriorating fast. Five lists in a fixed order that turn the whole thesis into one screen, each row traceable to the documents behind it.
Read Research · 4 minOpen-weights models against the frontier on covenant extraction
We ran four classes of model over 60 agreements and 1,840 cited fields. The frontier still leads on raw reading, but the gap sits almost entirely in definition chains and citations. With the deterministic pre-pass the production first pass reaches 97.4 percent, and the second read 99.7.
Read Research · 3 minRight 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.
ReadThe parts banks use most
Everything on this page runs on the same register as the rest of ExactCov. These are the parts this team lives in.
Agreements Intelligence
Debt and lien capacity, restricted payments, portability, MFN and value-leakage flags on the day the documents arrive, every figure cited to its clause and every definition resolved to its leaf.
Agreements Intelligence PlatformCommand Center
The whole book on one screen: what has arrived, what is stale, what is in breach and what is drifting, with headroom trends, scenario shocks and committee packs built from the register.
Command Center PlatformAgents
Chase requests what is due on the calendar the documents set, Extract reads what lands, Test recomputes every affected covenant within nine seconds, Brief writes the morning list and Escalate treats silence as an event.
AgentsRun 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.
See it against a live deal
Bring an OM or credit agreement from a deal in market and three closed precedents. We show the capacity table, the market position and what the register would have flagged.