The receipt gap
The verification calculators in this category price the hours a reviewer spends checking an agent’s work before it goes out. That is a real cost and they price it well. This calculator prices a different one: the ask that arrives later, from a carrier, a regulator, opposing counsel or a client, to prove what the agent did and what governed it, and the share of your records that are still there to answer.
The models that price supervision assume every record produces on demand. Most have no field for it. This one does, and it defaults to 100% so the assumption is priced by you, not by us.
The hours a check saves stay saved only if the receipt outlives the review.
Your programme
Nothing is sent anywhere. The arithmetic runs in your browser.
Your estimateActions taken by an agent that fall under a control you would have to substantiate later.
Your estimateSet this to your firm's retention policy, or to the longest limitation period you are exposed to on this work.
Your estimateDiscovery, a carrier audit, a regulator, a client dispute, a sanctions motion. No published measurement exists for agent actions specifically.
Your estimateThe input most verification models leave out. Set it to 100% if your receipts can be re-verified by a third party, any number of times, without the vendor that issued them.
Your estimateReconstructing what the agent did and what governed it, without a record that authenticates itself: pulling logs, a custodian declaration, the partner review that follows.
PublishedFRE 902(14) lets a written certification of a hash-verified record stand in for an authenticating witness. This is the certification, not a reconstruction.
Your estimateThe remainder is priced at the non-lawyer rate.
PublishedClio 2025 Legal Trends, US average lawyer rate.
PublishedClio 2025 Legal Trends, US average non-lawyer rate.
What the assumption costs
The difference between a programme where every record produces on demand and yours.
At 100% availability there is no gap. That is the claim being made when a verification model has no field for this. Drag the availability slider to price it.
Exposure does not end with the fiscal year. The ask arrives when it arrives.
Every one of these is something you may be asked to substantiate.
The population your records actually have to serve. It is much smaller than the action count, and it arrives late.
Priced at a blended $252 an hour.
The figures above are labour. They do not include what happens when an unsubstantiated action becomes the subject of a motion. In Couvrette v. Wisnovsky, the largest AI-citation sanction recorded in a US court to date, two lawyers were assessed $110,204.38 in December 2025. Only $15,500 of that was the sanction itself. $94,704.38 was the other side’s fees, and the case was dismissed with prejudice a week later. That is a tail, not an expected cost, which is why it is stated here rather than added in.
Where every number comes from
Published inputs are cited. Inputs with no published measurement are marked as your estimate and left to you. Nothing here is a vendor figure, including ours.
| Input | Status | Source |
|---|---|---|
| Lawyer rate, $349/hr. Non-lawyer rate, $187/hr. | Published | Clio, 2025 Legal Trends Report, US averages. Source |
| Certification instead of a witness when the record produces. | Published | Federal Rules of Evidence 902(13) and 902(14), effective December 1, 2017. A written certification by a qualified person, meeting Rule 902(11)/(12), authenticates a record generated by an electronic process or copied by a process of digital identification. The Advisory Committee's stated purpose was to spare the expense of producing an authenticating witness. Source |
| The sanctions tail: $110,204.38, of which $94,704.38 was opposing fees. | Published | Couvrette v. Wisnovsky, D. Or., December 2025 (Magistrate Judge Mark Clarke). Fifteen fabricated citations and eight fabricated quotations across three briefs; sanctions calibrated at $500 per citation and $1,000 per quotation; fees awarded to the defence; action dismissed with prejudice. Reported as the largest AI-citation sanction in a US court to date. |
| Why the ask is getting more likely. | Published | Damien Charlotin's AI Hallucination Cases database recorded 1,598 court decisions as of June 9, 2026, up from roughly 200 in mid-2025. ComplexDiscovery counted over $145,000 in penalties in Q1 2026 alone. Stanford RegLab (Magesh et al.) measured the leading legal research tools fabricating citations 17 to 33 percent of the time. |
| Share of actions ever asked about. | Your estimate | No published measurement exists for agent actions specifically. Default 2%. Set it from your own discovery, audit and dispute history. |
| Share of records producible and re-verifiable when asked. | Your estimate | Default 100%, which is the assumption every model without this field is making. Lower it to the share you can actually re-verify, by a third party, any number of times, without the vendor that issued the record. |
| Hours to substantiate one action by hand. | Your estimate | Default 6. Pulling logs, establishing what governed the action at the time, a custodian declaration, the partner review that follows. |
| Hours when the record does produce. | Your estimate | Default 0.5. The Rule 902(14) certification itself. The rule is published; the time it takes is yours. |
What this does not claim
- That a record makes anything admissible. Rule 902(14) addresses authentication, which is one element of admissibility. The court decides admissibility.
- That a record prevents a bad action. It does not. Blocking is a control; a record is what you produce afterwards, whichever way the control went.
- That the defaults describe your firm. Four of them are marked as your estimate because nobody has measured them. Change them.
Price the assumption, then test it
Seal a record yourself, then verify it twice: today, and whenever you come back. The anchor sits in a public log we do not operate. If a receipt you hold from anyone, including us, cannot pass that test, the availability slider above is where it belongs.