The Volume Problem: Why Discovery Review Broke

The numbers behind discovery review have stopped making sense. Audio and video evidence at the Denver District Attorney's Office grew 600% over five years, and mid-grade felonies that produced 3 or 4 recordings a decade ago now produce 50 to 60, according to Colorado Public Radio. That's the environment any discovery review software has to work in.

The math gets worse fast. A felony attorney handling roughly 100 cases, with 4 to 6 hours of bodycam per case, faces about 400 hours of video within 21 days of filing. Nobody watches 400 hours in three weeks while also doing everything else the job requires. Even minor misdemeanors can generate a terabyte of digital evidence.

What makes this a defense problem specifically is where you sit in the pipeline. Prosecutors and law enforcement generate the evidence and choose the systems that store it. Defenders receive it last, in whatever format and volume the producing agency decides, with no control over the vendors involved. James Karbach of the Colorado State Public Defender's Office put it plainly: something a prosecutor might prepare in two hours can take the defense 40, because "when you're downstream, you can't really control the vendors."

The practical consequences show up in the work itself:

• Bodycam, 911 audio, and jail calls arrive as raw media with no transcript, so every hour of footage is an hour of listening before you can even search it.
• The case often lives in the gap between sources: a CAD log at 21:14, bodycam starting at 21:31, a report that skips the 17 minutes in between. As one defense discovery analysis put it, a gap like that can decide a suppression motion.
• Video and call audio can be overwritten under an agency's retention schedule, so preservation has to be demanded up front, not after you've had time to review.
• Review is a comparison task, not a collection task. Collecting every file means nothing if the report and the footage are never read side by side.

Tom Raynes of the Colorado DA's Council acknowledged the scale honestly: terabytes of information "just growing exponentially every year," and his side's storage systems were built for a smaller era. The difference is that prosecutors built those systems. Defense teams inherit the output.

This is why generic tools fail here. A chatbot that summarizes a PDF doesn't handle 90 hours of footage, can't show you the second of video behind a claim, and raises confidentiality problems under Model Rule 1.6 when you upload client material. The volume problem broke discovery review, and fixing it requires tooling built for the media itself, with a grounded discovery review workflow that starts at intake, when the dump first lands in your inbox.

What Discovery Review Software Actually Has to Do

Before you compare a single product page, you need a test. Most tools marketed to defenders solve the wrong problem: they help you collect discovery faster, when the real work is comparing it. As one discovery methodology guide puts it, the case is usually in the gap between two sources that were never read side by side.

Think about how that gap actually looks. The CAD log shows dispatch at 21:14. The bodycam starts at 21:31. The police report narrates the encounter as one continuous event and never mentions the 17 minutes in between. None of those three documents is wrong on its own. Only when you read them against each other does the question appear, and a gap like that can decide a suppression motion. If your review tool can't put those sources on one screen, it's just an expensive filing cabinet.

The sharpest test from the research: if a tool cannot show you the page or the second behind each fact, it is not ready for a suppression motion. Every claim an AI makes about your evidence needs a clickable path back to the exact timestamp or PDF page. A summary you can't verify is a summary you can't cite, and a summary you can't cite is worse than nothing in front of a judge.

That's also why general AI chatbots fail here, and in two distinct ways:

• Fabricated citations. General chatbots can produce facts and citations that do not exist, which is disqualifying when the output feeds a motion.
• Model Rule 1.6 exposure. Uploading client material to a consumer chatbot raises confidentiality questions you don't want to answer in a bar complaint.
• No comparison layer. A chatbot can't sync a CAD log against bodycam timestamps or flag what the report skipped.

The volume math makes this urgent. Per Colorado Public Radio's reporting, felony attorneys carry roughly 100 cases each, with 4–6 hours of bodycam footage per case, all of it due for review within 21 days of filing. That's about 400 hours of video on a clock that's already running, and audio and video evidence in Denver alone has grown 600% in five years.

So the bar for any discovery review software is straightforward: it should ingest the dump, make every source searchable, and answer questions with citations to the exact moment. Evidence Copilot's discovery review workflow is built around exactly that standard, and the same discipline should apply to discovery intake: get the dump into a case file, then compare, don't just collect. AI is useful for the first pass. It does not replace watching the video yourself or deciding strategy.

The 2026 Tool Landscape, Honestly Assessed

Most of the numbers you'll see on vendor sites are self-reported, and none of these tools have been independently tested head-to-head. That's worth knowing before you commit a client's discovery to any of them.

The most established defense-side player is JusticeText, which has grown from roughly 40 public defender offices in 2023 to 70+ public defense agencies and 300+ private firms. Its vendor-reported claim is that users free up nearly 50% of their time, including one Nevada investigator who reviewed about 19 hours of bodycam in 2 hours. Note the discrepancy, though: Reduct's blog claims JusticeText "is yet to introduce video redaction features," while JusticeText's own homepage lists redaction, clipping, and video syncing among its exhibit-prep features. Both claims come from interested parties.

Reduct handles video transcription and clipping, self-reporting 94% accuracy across 90+ languages. That figure has a catch the vendor itself flags: transcription misses specialized legal terms, which matters when you're building a suppression record around "Miranda" or "consent search." CaseMark targets bodycam summaries, self-reporting that manual transcript review "often tak[es] 4-6 hours per video" and that its 10-section summaries generate "in minutes." Those are CaseMark's own numbers.

Then there's the free DIY stack: local Whisper plus NotebookLM. It works fine for one interview. At case scale it breaks down, because you get no speaker-diarization QC, no citations back to the source moment, no redaction capability, and no audit trail. Uploading client material to general chatbots also raises confidentiality questions under Model Rule 1.6.

Two honest caveats apply across the board. First, virtually every performance figure above is vendor-reported with no independent verification. Second, no vendor in this comparison publishes pricing on the pages we reviewed, which means you're requesting quotes before you know the number.

Evidence Copilot approaches the same problem differently. Every AI answer cites the exact second of video or page of PDF it came from, so you can verify before you rely. AI output is always marked pending human review, originals stay immutable, and auto-signals on ingest surface report-vs-bodycam inconsistencies before you think to ask. Pricing is published: Solo $39/user/mo, Firm $99/user/mo, PD Office $750/attorney/yr, with no sales wall.

If you're evaluating tools against your actual workflow, our pages on discovery intake and the grounded discovery review workflow show how the pieces fit together.

How to Run the Review: Preservation, Comparison, First Pass

The clock starts ticking the moment you file your first discovery request. If you wait, you lose. Video files, 911 audio, and jail call recordings are often stored on agency servers with automatic retention schedules. They may be overwritten before you even get a chance to review them.

Send a preservation request alongside your initial discovery demand. Do not wait for the prosecutor to respond. The NACDL Fourth Amendment Center offers an Omnibus Tech Discovery List that serves as a credible, non-vendor baseline for these technical requests. It helps ensure you ask for the right digital artifacts without relying on a specific software provider’s template.

Once the files arrive, the real work begins. You are not just collecting documents; you are comparing them. A discovery checklist only helps when every item is compared, not just collected. The case usually lives in the gap between two sources that were never read side by side.

Consider a common scenario: the CAD log shows an officer arrived at 21:14. The bodycam footage starts at 21:31. The police report skips the 17-minute gap entirely. A gap like that can decide a suppression motion. You need to see all three sources simultaneously to spot the discrepancy.

This is where discovery review software fits into the workflow. Use AI as a first pass, nothing more. It cannot judge a witness, decide which conflict to use, or replace your own review of the recordings. It flags potential inconsistencies between the report and the video, but you make the legal determination.

For a concrete implementation path, look at how discovery intake turns a chaotic email dump into a structured case file. Then, follow the discovery review workflow to cross-source the evidence.

Here is the practical sequence for your first pass:

• Ingest all sources immediately upon receipt to start the clock on review.
• Run transcription and OCR to make audio and scanned PDFs searchable.
• Compare the police report narrative against the bodycam timeline.
• Check the CAD log against the 911 call audio for entry time discrepancies.
• Review jail calls for statements that contradict the initial report.

Evidence Copilot’s workspace handles this by surfacing candidate inconsistencies before you even ask. Every AI output is marked as pending human review, ensuring you retain full control over the legal strategy.

The volume of evidence is staggering. Felony attorneys handle an average of 4–6 hours of bodycam footage per case, often within 21 days of filing. That is hundreds of hours of video to get through. You need a system that shows you the exact source moment behind every fact. If a tool cannot show you the page behind each fact, it is not ready for a suppression motion.

Start with the preservation request. Then, build a review process that relies on grounded citations and human judgment.