Deposition-to-brief stack — transcript to filed motion with both records checkable

A brief carries two records, and they fail differently. The factual record is the testimony, the exhibits, and the produced documents in your own matter — a corpus nobody outside the case has ever indexed. The legal record is published authority, which every research platform has indexed and which is exactly where generative models invent things that do not exist. Damien Charlotin’s AI Hallucination Cases database, updated 22 August 2026, lists 1,941 decisions in which a court addressed AI-generated content in a filing. Almost all of the famous ones are legal-record failures: a case name that reads correctly and does not exist.

The factual-record failure is quieter and more common. A sentence in the fact section says the witness admitted something, and the transcript page it cites says something adjacent. No judge writes an opinion about it. Opposing counsel writes a reply brief about it.

These are two different verification problems over two different corpora, and one tool does not do both. That is the load-bearing reason this stack has four layers instead of one legal-AI subscription.

How the pieces fit

• Steno captures the record. It is a court reporting agency, not software: certified reporters, videographers, and interpreters booked per proceeding, with Steno Connect for Zoom handling remote depositions and a Firm Dashboard holding scheduling and files. What matters downstream is what comes back — a certified transcript, exhibits, and synced video. Transcript Genius sits on top at no additional charge and no usage limit, running exact-match and semantic search across a transcript, comparing multiple depositions to surface contradictions, and returning linked page-line citations on every answer. It ingests third-party .txt and .pdf transcripts, so testimony taken by other agencies enters the same corpus. Steno closed a $49M Series C on 26 March 2026.

• Everlaw is where the transcript stops being a document and becomes evidence. Load the transcripts alongside the produced document set and the Deposition Analyzer works the testimony against everything else in the matter. Since the October 2025 release this is not a separate line item: Review Assistant for single documents, the Writing Assistant inside Story Builder, and the Deposition Analyzer are all included in the core per-gigabyte rate, and Everlaw did not raise the per-gigabyte price to absorb them. Deep Dive, which answers natural-language questions across an entire collection, carries a one-time per-gigabyte ingestion fee and then unlimited questions for the life of the case. Story Builder is the piece that connects to drafting — evidence bound to legal theory, exhibits bound to the testimony that authenticates them.

• Clearbrief verifies the factual record, inside Word. Point it at the draft and Analyze reads each factual assertion, searches the transcripts, exhibits, and records you loaded, and returns the passage that supports it — or reports that it found none, which is the answer you are paying for. The patented Add Fact-Cite button inserts the citation, and every cite it writes is hyperlinked in the Word document, so the reviewing partner or the judge clicks and lands on the source page. It won Legalweek 2026 Litigation Technology of the Year, and 3.2 million pleadings have been drafted through it since 2021.

• Thomson Reuters CoCounsel verifies the legal record. Deep Research runs multi-step research grounded on Westlaw and Practical Law rather than on model memory, and the June 2026 release added Deep Research Verify, which checks whether cited authority actually supports the assertion made and highlights the supporting passage. The chat interface plans the research, retrieves from your own precedents plus Westlaw, and drafts with citations; results export to PDF or Markdown with the citation style set to endnotes, footnotes, or none.

Named handoffs

1. Deposition noticed → Steno books the proceeding → certified transcript, exhibits, and synced video return. On a contingency matter, DelayPay defers payment until the case concludes with no interest, no lien, and no credit check, which is why plaintiff shops pick Steno over a cheaper local agency.
2. Transcript file → loaded into Everlaw with the produced set → Deposition Analyzer runs against the whole corpus. This is the step that catches the contradiction between what the witness said in April and what the custodian’s email said in January. Running the analyzer on the transcript alone finds far less, because the contradiction lives across documents.
3. Analyzer output → Story Builder → evidence attached to the legal theory it proves. The Writing Assistant drafts the fact section from that structure, with the record cites already attached rather than added afterward from memory.
4. Draft exported to Word → Clearbrief Analyze → each factual assertion either resolves to a passage or is flagged as unsupported. Add Fact-Cite inserts hyperlinked citations. The guard here is procedural, not technical: treat every “no supporting passage found” as blocking, because the failure mode is a lawyer overriding the flag on a sentence they remember being true.
5. Legal argument → CoCounsel Deep Research on Westlaw → Deep Research Verify on the resulting cites. The citation verification skill covers the manual version if your research contract does not include an AI layer.
6. Both passes clean → the brief files with clickable record cites. A judge who can click from the assertion to the transcript page reads the fact section differently than one who cannot.

Why this combination

Because grounding is not one property. Clearbrief is grounded in your matter and knows nothing about published case law. CoCounsel is grounded in Westlaw and has never seen your deposition. Buying either one and calling the brief checked leaves half the document unverified, and it is the half you will not notice, because both tools return confident output within their own corpus.

The second reason is that the transcript arrives as a file and everything after that is a corpus problem. Teams that skip the Everlaw step do the deposition analysis in a chat window against a pasted transcript, which works until the answer depends on a document the transcript does not contain.

Cost reality

For a five-lawyer litigation group with one active matter of meaningful size:

• Steno: no subscription. Billing is per proceeding off a rate card of roughly 25 line items. Against 2026 US market rates of $4.50–7.00 per page and $150–400 appearance fees, a four-hour deposition yielding 150–250 pages lands between $825 and $2,150 before video, realtime, rough drafts, or expedite. Steno prices above commodity local agencies, and the DelayPay card is a separate, higher schedule — deferral is priced in, not free.
• Everlaw: no published list price. Mid-market litigation matters run $20K–$100K+, and a Vendr community report from a 201–1,000-employee buyer describes a $50,000 annual minimum commitment, renegotiated 20% down when usage came in at half the commit. That minimum is the real gate on this stack.
• Clearbrief: $300 per user per month, published, for Solo and small teams, with unlimited use per seat. Enterprise Unlimited is custom, and multi-year agreements are discounted on request. Three seats is $10,800 a year of published price.
• CoCounsel: no published price. Effective rates run $200–500 per attorney per month, almost always inside a Westlaw and Practical Law bundle rather than standalone.

Five CoCounsel seats, three Clearbrief seats, and an Everlaw commitment put the annual software floor at roughly $73,000–$91,000, with deposition costs on top and variable. The number that decides whether this stack is rational is the Everlaw minimum: below a docket that consumes it, you are paying a platform commitment to hold three transcripts.

Match rules

Right pick for a litigation group with a standing docket — a plaintiff firm running mass tort or serial PI, an AmLaw litigation department, or an in-house team that defends its own cases — where depositions run into the dozens per year and the record outgrows what a single associate can hold.

Wrong pick for a team taking two or three depositions a year. At that volume the platform commitment dominates, and the right shape is Clearbrief plus whatever platform outside counsel already hosts. It is also the wrong stack for transactional work: nothing here touches contracts.

Common variations

• Swap Steno for Depositions by Filevine when the case system is already Filevine. Filevine schedules the hosted proceeding, returns video- and audio-synced transcripts with same-day AI summaries, and keeps a searchable archive of transcripts and clips inside the matter record; Depo CoPilot tracks stated objectives during the deposition and flags gaps in the record as it goes. Swap when the firm runs Filevine and schedules its own proceedings. Keep Steno when you need DelayPay deferral on contingency files or a national reporter network in markets Filevine does not cover.
• Swap Everlaw for Relativity when someone else picks the review environment. Client mandates and outside-counsel hosting decide this more often than product preference, and aiR for Review is bundled into RelativityOne, so the analytical layer survives the swap. What does not survive is Story Builder — you lose the evidence-to-theory structure and rebuild it in the drafting tool.
• Drop the review platform entirely on a small matter. Clearbrief loads transcripts and exhibits directly, so a stack of Steno plus Clearbrief plus a research subscription handles a single-plaintiff case with four depositions. Swap back up the moment the produced set exceeds what a person can read.
• Swap CoCounsel for Lexis+ with Protégé when the research contract is LexisNexis. Clearbrief integrates with LexisNexis to flag hallucinated case citations, so the legal-record check does not disappear with the swap. Do not run both research platforms to get both AI layers — the second subscription costs more than the marginal verification is worth.

What this stack does NOT replace

• The certification you sign. Deep Research Verify and Clearbrief reduce the time a human spends checking; they do not discharge the obligation. Courts that have issued standing orders on generative AI require an attorney to attest, and no vendor attests on your behalf.
• Conflicts and ethical walls. Who inside the firm can see which matter through which AI tool is a governance question this stack does not answer — that is the law firm AI governance stack.
• Collection, processing, and privilege review at scale. Getting custodian data in, deduplicated, culled, and privilege-logged is a different pipeline with different tools; see the e-discovery stack.
• The certified transcript’s authority. AI summaries are working product, not the record. The certified transcript is what gets cited, and a page-line cite has to resolve to the certified version.
• The theory of the case. Story Builder structures evidence against a theory somebody else decided. Nothing in this stack tells you which admission matters, and the deposition outline skill is upstream of all of it — the questions you did not ask do not become citable testimony.