Public Records Request and Redaction Software: Surviving the Clock, the Video and the Appeal

The request that starts the weekend badly

It arrives at 4:20pm on a Friday, by email, from a local reporter. All body worn camera footage, dispatch audio, incident reports and internal email relating to a single arrest three weeks ago. That is fourteen hours of video across six officers, two hours of radio traffic, a stack of reports, and whatever turns up when somebody eventually searches the chief's mailbox. The statutory clock started when the email hit the inbox, whether or not anyone opened it. Nobody in the office is going to be able to tell you by Monday how many hours of work is actually sitting in that request.

The office already has a tracking tool. It logged the request, assigned a number and sent the requester an acknowledgement. What it cannot do is any of the work: it does not know the video exists, cannot search the email archive, cannot tell a redactor which faces in frame belong to an uninvolved juvenile, and will not produce, when the denial is challenged six months from now, a page by page record of which exemption was cited for which withheld line. That last gap is the one that ends up in front of a judge.

The clock is not the real exposure, the record of your reasoning is

Late responses are embarrassing. Undocumented withholding is expensive. Many state public records acts shift the requester's attorney fees onto the agency when the requester prevails, which turns a defensible partial denial into a budget line and a council agenda item. The question in that proceeding is never whether you were busy. It is which exemption you applied, to which specific content, and on what basis, and whether you can produce that showing without reconstructing it from memory a year later.

Most offices cannot. The withholding decision happens inside a PDF editor or a video tool, the reasoning lives in a paralegal's head or in an email thread, and the exemption log is retyped afterwards as a summary rather than captured as a byproduct of the work. Reconstructing it burns days and produces a document that a competent requester's attorney will take apart, because the summary and the actual redactions do not line up perfectly and never could.

Where GovQA, NextRequest and JustFOIA stop

These are competent request tracking products and they solve the front of the process well: a public portal, a request number, correspondence history, fee estimates, deadline reminders, a place for the requester to see status without calling the clerk. If your office is drowning in phone calls and sticky notes, they will help you tomorrow and you should look at them first.

Where they stop is fulfilment. They track that a request exists, not what is responsive to it. Search happens in your other systems and the results get dragged back in by hand. Redaction is a separate product, often Veritone Redact or an equivalent, which means the withholding decision is made in a tool that has no idea which request it belongs to, and the exemption citation gets typed into a spreadsheet after the fact. The chain of custody between what a system held, what a reviewer withheld, and what citation justified it is assembled manually every single time.

Veritone Redact and its peers are worth taking seriously for the video work itself, because automated face and plate detection with tracking across frames is genuinely hard and they have invested in it. The gap is that redaction is being sold as a media tool rather than as part of a records decision, so the audit trail you actually need lives in whatever you build or maintain around it.

The exemption log is the product

If a custom build does one thing, this is it. Every withheld element, whether a page, a paragraph, a field in a spreadsheet or a range of frames in a video, carries a citation to the specific statutory exemption, the reviewer who applied it, the timestamp, and a short justification note. Not a summary written afterwards. A record generated as a consequence of doing the review, which cannot drift from the released product because it is the same object.

Get that right and three things follow. The response letter generates itself with an accurate index rather than a paragraph of general language. An appeal is answered in an afternoon instead of a fortnight. And your own supervisors can finally see whether two reviewers are applying the same exemption to the same category of content, which they usually are not, and which is a quality problem no tracking tool will ever surface for you.

Video is where the hours go, and where the money is

Redaction of footage is measured in hours per minute of source video when a human is doing frame by frame work. That single ratio is why records units miss deadlines. The build has to attack it directly: automated detection of faces, licence plates, screens and documents in frame, object tracking so a subject stays redacted through a pan, audio redaction tied to a transcript so a reviewer can mark a spoken address in the text rather than scrubbing a waveform, and a review pass where the human confirms and corrects rather than draws.

Be honest with yourself about the residual. Detection will miss a face at the edge of frame in low light, and it will over redact a reflection. The design goal is not zero human review, it is turning the human into a checker. In our experience the useful measurement to hold a vendor to is the review time per minute of footage after automation, not the accuracy percentage of the detector, because the second number is easy to make look good on clean daytime video and tells you nothing about a foot pursuit at night.

Search has to reach the systems the request actually spans

• The records management system, for reports, arrests and case narratives, with case number and person and address as entry points.
• The digital evidence store, for body camera, in car and interview room media, matched by incident rather than by filename.
• The email and messaging archive, with custodian and date range search that a records officer can run without a legal hold tool.
• Financial, permitting and asset systems, because contract and invoice requests are a large and growing share of volume.
• A responsiveness workspace where a reviewer marks each item responsive, non responsive or withheld, and the withheld items carry citations automatically.

What this costs and how long it takes

A first release covering intake across email, portal and phone, the statutory clock with your state's calculation rules, the responsiveness workspace and the exemption log runs $60,000 to $140,000 and ships in 12 to 16 weeks. The full platform adding automated video and audio redaction, source system search connectors, fee calculation and collection, and the appeal and litigation hold workflow runs $150,000 to $350,000 phased over 6 to 12 months.

Cost moves on two things specifically. The number of source systems you want searched from one place, because each connector is real integration work and vendor cooperation varies enormously. And the volume of video, because at high volume you are making infrastructure decisions about storage and transcode capacity that a document only office never has to think about. Fee calculation is fiddlier than it looks, since most state rules distinguish between search time, review time, duplication and the cost of a redaction technician, and getting that wrong creates its own appeals.

When to buy instead of build

Buy if you are a small or mid sized city clerk with document requests, no body camera programme and a handful of appeals a year. NextRequest and GovQA will improve your life next month for a fraction of a build, and the honest answer is that your bottleneck is intake discipline rather than software architecture.

Build when at least two of these describe you. Requests routinely span video, records and email in one filing. Your office has lost a fee award, or been warned about one, over withholding you could not document. You are a police records unit whose video redaction backlog is measured in months. Or you sit at a jurisdiction where the same request arrives at five departments and nobody can see that it is one request being worked five times.

How to choose a developer

Ask them to model the exemption log on a whiteboard before you talk about anything else. The right answer connects a request to responsive items, items to withheld ranges, and each range to a statutory citation and a reviewer. If they draw requests and attachments, they have built a ticketing system and are about to learn public records law on your budget.

Ask what they have integrated on the evidence side. Pulling media out of a camera vendor's platform, at volume, without paying that vendor for the privilege, is a specific skill and the answer should name the platform. Ask how they measure redaction throughput, and listen for review minutes per footage minute rather than detector accuracy.

Then settle ownership before kickoff: the repositories, the cloud accounts, the model artefacts if any were trained on your footage, and your unrestricted right to bring in another firm. At Digital Heroes that is the default from the first commit. Start your own scoping by pulling last year's requests and sorting them by hours spent, not by count, because the answer is usually that a small number of video requests consumed most of the office.