A diligence report tells you what was read. Its silence about what was not read is the part you are actually buying — and the obligations inside those contracts outlive the memo that failed to mention them.
Every contract diligence exercise produces two documents. The first is the report: a description of what counsel read and what they found. The second is written in invisible ink and never circulated — the list of agreements nobody opened, and the representations the buyer is therefore relying on instead. The second document is the one that prices the deal.
This paper makes four claims. (1) Sampling is a coverage assertion, and it is the only material number in a diligence process that is never stated. A report that says the material contracts were reviewed is making a claim about a denominator it does not disclose. (2) Coverage was a budget constraint and has become a choice. When reading the whole room costs a fraction of reading a fifth of it, declining to state coverage stops being a limitation and starts being a disclosure decision. (3) A diligence review is a statement about a moment, not a room; data rooms move, and a review that cannot be cheaply re-run describes a data room that no longer exists. (4) The obligations discovered in diligence do not expire at completion — they become the operating company's day-one problem, and the artefact that carries them across is almost never the report.
We argue throughout from the same five properties that govern audit-grade evidence in regulated work: traceability, independence, reproducibility, accountability and freshness. A buy-side finding earns its authority the same way a regulatory one does, and for the same reason — someone will eventually dispute it.
Open any legal due diligence report on a mid-market transaction and read the scope paragraph. It will tell you which categories of agreement were reviewed, and it will usually tell you the threshold — contracts above a certain annual value, or the counterparties representing the top decile of revenue. What it will not tell you is the fraction. Of the four thousand documents in the room, how many were opened?
The number is never stated because for the whole history of the practice it could not be improved. An associate reads perhaps forty agreements properly in the time available. A room holds thousands. The ratio is not a failure of diligence; it is the arithmetic of a three-week exclusivity window against a corpus assembled over fifteen years of trading. Everyone in the process knows this, which is precisely why nobody writes it down.
The consequence is that the unread portion of the estate is transferred silently into the warranty package. A representation that there are no change-of-control provisions requiring consent is, in substance, a statement that none were found in the part that was examined. That is a materially different assertion, and its price is set as though it were the stronger one.
This is not an argument that diligence is done badly. It is an argument that one of its central quantities has been unmeasurable, and that unmeasurable quantities do not appear in reports. When the constraint moves, the reporting convention should move with it.
In our white paper on audit-grade AI we drew a distinction between testimony and evidence. Testimony is an assertion that rests on the authority of whoever makes it. Evidence is an assertion that carries its own basis, so that a reader can check it without trusting the speaker. Regulated institutions stopped accepting testimony some years ago. Transaction practice has not yet had to.
The distinction transfers exactly. We reviewed the material customer contracts and found no assignment restrictions is testimony: it is true or false depending on facts the reader cannot reach. Clause 14.2 of the CloudCore MSA, on page 27, states that the Customer shall not be entitled to audit the Supplier's premises is evidence: the reader can open the page and disagree.
| As testimony | As evidence |
|---|---|
| The material contracts contain no unusual change-of-control provisions. | Of 412 agreements, term extracted in 407; 9 carry consent-on-control provisions, each quoted with document and page; 5 could not be read and are listed. |
| Liability is capped at market norms. | Caps extracted per agreement into a sortable column; 3 uncapped indemnities cited verbatim. |
| We are comfortable with the customer base. | Auto-renewal and notice provisions per contract, with the passage each was read from. |
The right-hand column is not a better opinion. It is a different kind of object: one that survives a challenge without its author in the room, and one that a buyer's counsel, a seller's counsel and an arbitrator three years later can all inspect and reach the same reading of.
Note the second half of the first row. The evidential form states what it could not read. This is not a weakness in the output — it is the property that makes the rest of it worth anything, because a report that never says unread gives you no way to tell a clean estate from an incomplete review.
The economics that produced sampling were real. Reading a contract properly — not skimming it, but extracting a defined set of provisions and recording where each one was found — took a trained person the better part of an hour. Four thousand documents was not a budget line; it was an impossibility.
That constraint has moved, and it has moved by more than an order of magnitude. A machine that reads every agreement against a fixed question set, records the passage behind each answer, and reports what it could not read, changes the shape of the problem: the expensive resource is no longer reading, it is judgement. Counsel stops spending the window locating the paragraph and starts spending it deciding what the paragraph means for the deal.
The implication for practice is uncomfortable and worth stating plainly. Once full coverage is affordable, a report that does not state its coverage is making a choice. The number was previously absent because it could not be improved; it is now absent because nobody has asked for it. We expect buy-side counsel to start asking, in the same way that regulators started asking institutions to show the evidence behind an attestation rather than the attestation itself.
We are not arguing that a machine should form the view. We are arguing that the denominator should appear in the report.
Diligence is conducted against a room that is still being assembled. Documents arrive after the first tranche; amendments surface in response to questions; a seller's counsel uploads the side letters on a Thursday afternoon in week three. A review completed on Tuesday is an accurate description of a data room that no longer exists on Friday.
The practical response has been to re-read selectively and hope the additions were immaterial — which is a second, compounding sampling decision layered on the first, and one made under more time pressure than the original.
Content addressing removes the dilemma. If every document carries a hash of its contents, the set that changed between Tuesday and Friday is computable rather than estimated, and only that set needs re-reading. The marginal cost of keeping a review current collapses toward the cost of the delta, which for a late-arriving tranche of side letters is small. Freshness — the fifth of the five properties — stops being a matter of diligence hygiene and becomes a property of the system.
This matters beyond convenience. A finding that cannot be cheaply re-derived is a finding whose currency degrades silently between the day it was made and the day it is relied upon at completion.
Consider what happens to a diligence finding after completion. A notice provision requiring ninety days' written warning before a renegotiation, discovered in week two and noted at paragraph 4.7 of the report, becomes the operating company's problem on day one. The report is filed. The obligation is not.
This is the structural failure that the industry has been least willing to look at, because it falls in the gap between two teams. The diligence team's mandate ends at signing. The portfolio operations team's mandate begins at completion, and begins with a PDF nobody reads because it was written to support a pricing decision that has already been taken.
The obligations discovered during diligence have all the properties of trackable work: a counterparty, a trigger, a deadline, and a consequence for missing it. What they lack is a destination. Carrying them from the diligence artefact into an operational register — with an owner and a date attached to each — is not a technically difficult thing to do. It is simply not what a report is for.
| Discovered in diligence | Becomes, after close |
|---|---|
| Consent required on change of control | A consent obtained, or a tracked exception with an owner |
| Auto-renewal with 90-day notice | A dated item on the calendar of whoever holds the relationship |
| Audit and inspection right absent | A renegotiation item, or a recorded acceptance of the risk |
| Reporting obligation to a counterparty | A recurring duty with a named owner |
| Uncapped indemnity | An escalation with a decision recorded against it |
The right-hand column is the 100-day plan, and most of it is already written by the time the deal closes — it is sitting in the diligence findings, in prose, in an appendix. The work is transport, not discovery.
A buy-side finding will be disputed — by the seller during negotiation, by a counterparty after completion, or by an investment committee asking why a risk was not flagged. The properties that let it survive are not rhetorical. They are the same five that govern evidence in regulated work.
The platform records only what it can observe. A side letter is marked prepared when the platform prepares it, and sent only when a person asserts that it was. We could log the send and be more impressive; the record would then contain an event nobody witnessed.
An evidence product that records unverifiable actions is worth less than one that records fewer. This is the same instinct that makes the coverage number worth publishing: a system that will tell you what it did not do is the only kind whose positive claims mean anything.
Data residency belongs in this list for a European buyer, not as a compliance checkbox but as a property of the evidence chain. A target's contract estate is among the most commercially sensitive material a firm handles. Deployment is on the customer's choice of sovereign EU hosting or AWS, with data held in the EU, and scanned pages are read locally rather than shipped to a third-party service for transcription.
We state these because a buy-side audience discounts a document that has no limits section, and because the discount is deserved.
The argument of this paper is narrow and, we think, hard to disagree with once stated. Contract diligence has always contained an undisclosed quantity — the proportion of the estate that was actually read. That quantity was undisclosed because it was not improvable. It is now improvable, and its continued absence from reports is a convention rather than a constraint.
The second argument is smaller and more practical. The obligations found during diligence survive the transaction that discovered them, and the artefact that records them does not. Moving those obligations into something with owners and dates is not innovation; it is finishing the job.
We would rather be judged on one data room than on this paper. Point the platform at a live or closed transaction, run your own diligence questions across the full document set, and compare what returns against what your team found — including, in particular, the list of what neither of you had read.