The discovery dump lands the week before the preliminary hearing: 40 hours of bodycam footage spread across dozens of files with names like X13qrt9__x3455, two interrogation videos, a stack of scanned police reports, and a 911 call you haven't heard yet. You have maybe three working days to get through it. If you're a solo defender or you're carrying a public-defender caseload, there's no associate to delegate to and no budget for a vendor that charges per hour of reviewed footage. Here's the workflow that makes it survivable. It's the same structure experienced defenders use, whether they're working with transcripts from a transcription service, a local Whisper setup, or a purpose-built workspace like Evidence Copilot. The tools change; the method doesn't. And the method matters more than the tools, because the failure mode in a big bodycam review is never watching too slowly. It's finding the one moment that matters and then being unable to put your hands on it again two weeks later.
Key points
- Triage the dump before you watch anything: inventory the files, sort by recording officer and date, and start with the footage that brackets the report's central events.
- Word-level timestamps and speaker diarization are the difference between a transcript you can use for impeachment and one you have to re-verify from scratch.
- The highest-value findings come from comparing the police report against the footage, not from watching footage in isolation; look for events the report describes that the camera doesn't show, and vice versa.
- Every key moment should be saved with a citation to the exact second, so cross-exam prep and suppression drafting pull from a verified record instead of memory.
- Redaction for discovery turnover should always produce a new derived file; the original footage stays untouched and every edit is auditable.
Resist the urge to open the first video file. Start with the file list itself. Sort everything by date, then by recording device or officer if the metadata makes that possible. In most dumps, the same event shows up in multiple officers' footage with different start times, and knowing that before you watch saves you from reviewing the same arrest four times. Next, read the police report first, not the footage. Mark every factual claim in the report that could be checked against a recording: when the stop started, what the officer says about consent to search, what was said during the encounter, when Miranda warnings were given. That list of checkable claims is your review plan. You're not watching 40 hours of video; you're verifying a specific set of assertions against a specific set of recordings. This is also where you note what's missing. If the report describes a K-9 deployment but no file in the dump covers that window, that's a discovery follow-up you want to send now, not in week six.
A bodycam transcript is only useful if you can trust two things: who is speaking, and when. Automatic transcription without speaker diarization gives you a wall of text where officer and client are indistinguishable, and without word-level timestamps you can't cite anything. When you're building a cross-exam or drafting a motion to suppress, "the officer said something like this at some point in the video" is worthless. "Officer Reyes, at 00:41:17, on camera 3" is a question you can ask on the stand. If you're running a local Whisper setup, budget real time for QC. Whisper does solid raw transcription but no diarization, and it routinely stumbles on legal terms and police jargon: 'Terry stop,' 'Giglio,' surnames, street names. Every one of those errors is a landmine if the transcript ends up in front of a court. Whatever tool you use, spot-check the segments that matter most: the consent conversation, the Miranda exchange, any statement the report characterizes as an admission. One workflow note that saves hours: rename speakers once, and make that rename propagate across the whole transcript. If the diarization labeled your client 'Speaker 2' in 300 segments, you want 'Mr. Delgado' in all 300, not a manual find-and-replace project.
The money in a bodycam review is almost never in the footage alone. It's in the delta between what the report says and what the camera captured. Three patterns come up constantly: First, the report describes something the footage doesn't show. The report says the client 'consented to a search of the vehicle.' The footage shows the officer saying 'you're fine with me looking around, right?' and the client saying 'I guess.' Those are different facts, and the difference is the suppression motion. Second, the footage shows something the report omits. A 20-minute gap between the stop and the Miranda warning, visible on the timestamps, that the report compresses into 'shortly after.' A second officer's comments that never made the narrative. Third, the tone is different. Reports read calm and procedural; the audio captures something else. Jurors and judges react to that gap. Go through your list of checkable claims from Step 1 and log each one as verified, contradicted, or unverifiable. Anything contradicted becomes a candidate inconsistency worth its own note, with both the report language and the transcript excerpt saved side by side. Tools that surface these comparisons automatically, like the ingest-time inconsistency signals in Evidence Copilot, can catch report-vs-bodycam and statement-vs-jail-call mismatches before you think to ask, but the manual pass is worth doing regardless, because you know the theory of the case and the software doesn't.
As you review, save key moments as timeline entries with a citation to the exact source: the file, the timestamp, the quoted line. This sounds like overhead. It's the opposite. It's what lets you answer, three weeks later when you're drafting the suppression motion at 11 p.m., the question 'where was that again?' in ten seconds instead of re-skimming an hour of footage. The same log feeds everything downstream. Cross-exam questions pulled from quoted transcript lines. Exhibit lists tied to real files. The case brief you hand a co-counsel or a supervisor. If your workspace drafts these for you, as Evidence Copilot's cross-exam and exhibit-list playbooks do, the drafts are only as good as the underlying log, so the log is where your effort pays off. A caution that applies to any AI-assisted review: treat every AI-generated summary, suggested question, or flagged inconsistency as unverified until you've checked the citation yourself. The output should say so, and if a tool doesn't mark AI output as pending human review, that's a tool problem. You're the attorney of record. The citation either checks out or it doesn't.
If any of this footage gets shared with the defense side of discovery, or if you need to turn over redacted media, keep one rule: the original file is never modified. Frame-accurate video redaction (a rectangle drawn over a bystander's face, blacked or blurred), audio redaction that mutes exactly the spoken words you select, and PDF redaction should all produce new derived files, with the source untouched and every redaction action logged. Watch for tools that offer AI-suggested redaction regions for faces, PII screens, and license plates. Useful, but they should arrive as suggestions you approve, never as edits applied automatically. An auto-applied redaction that hits the wrong frame in discovery turnover is the kind of mistake that ends up in front of a judge. If you're syncing multi-camera footage of one event, bodycam plus interview room plus dashcam, look for a tool that shows its work on the sync. Evidence Copilot, for example, reports the match evidence behind its sync proposals (something like '31 word matches, ±0.2 s spread') and never applies them automatically. You want to see that math before you trust a master timeline.
Budget matters, especially for PD offices and solos. Evidence Copilot publishes its pricing openly: Solo at $39/user/month with 15 media-processing hours included, Firm at $99/user/month with 60 hours, and a PD Office tier at $750/attorney/year with 100 hours. Overage is metered but hard-capped at 3x the included hours, so a heavy intake month can't produce a surprise invoice. There's a 7-day full-access trial with no card required, and a 14-day money-back guarantee on paid plans. For comparison: JusticeText runs roughly $1,200 per attorney per year, so the PD tier comes in about 38% lower. General-purpose tools are cheaper up front but cost you elsewhere. Sonix is generic transcription with no diarization QC for legal terms and no citation structure. The free DIY stack (local Whisper plus NotebookLM) works for a single short video and breaks at case scale: no speaker QC, no citations into the video, no redaction, no audit trail. CoCounsel, at around $400/user/month, is a research and memo tool that doesn't review evidence media at all. And the platform most defenders are forced to receive discovery through, Axon Evidence, is built for the prosecution side: slow player, no analysis, no workflow for the person whose client is in the footage.
Forty hours of bodycam is not a watching problem. It's a verification problem: check the report's claims against the recordings, log every key moment with a citation to the exact second, and keep the originals immutable while redacted copies go out the door. Do that, and the hearing prep writes itself, because everything you need is one click away from the source. If you want to see this workflow on your own discovery rather than a demo case, Evidence Copilot runs a 7-day full-access trial with no card, and you can book a 20-minute walkthrough at evidencecopilot.com/book or start on the Solo plan at evidencecopilot.com/signup. Every AI output in the workspace is marked for human review, and the product never makes legal conclusions. The judgment stays yours. The tool just makes sure you can always find the moment that supports it.