Footage review should be a list of events, not a pile of hours.

Because the DVR hands them no other option. A consumer DVR stores a raw H.264 or H.265 blob per channel, addressable by timestamp and nothing else. With no index, the only way to find an event is to play the blob back faster than real time and watch it with human eyes. The 4x recommendation is not a review strategy, it is an admission that no real review tool exists on the recorder. The review workflow should not be playback at all. It should be a filter over a list of events that were already classified at the moment they happened.

It is a table where each row is one classified incident with a timestamp, a camera label, an event class, and a thumbnail. Most recent event on top. To review an overnight shift you scroll the list from 6 a.m. down to 10 p.m. the night before, glancing at each thumbnail for half a second. A quiet property produces five to twenty rows. A busy one produces two hundred. Either way, review time is O(events), not O(hours). Compare that to scrubbing eight hours of footage across sixteen cameras at 4x, which is thirty-two camera-hours of real-time video condensed to eight viewing-hours. One shift review. One work day lost.

Cyrano taps the HDMI output of your existing DVR. Every DVR drives a local monitor via HDMI, displaying all the camera feeds composited into a multiview grid (2x2, 3x3, 4x4, and so on). Cyrano splits that signal with an HDMI tap, feeds it into a small edge box, and runs a person and vehicle classifier on each tile of the composite in parallel. Up to 25 tiles per HDMI input at 30 frames per second, 38 millisecond inference latency per frame. The classifier writes a row every time something crosses a threshold. No camera replacement. No RTSP credentials. No cloud upload of video. One HDMI cable in, one HDMI cable out to your monitor, one network cable for the review dashboard.

Because Cyrano reads them off the DVR's own pixels. Every DVR paints the camera name onto each tile of the multiview as a small text strip. That strip is called the cam_name_strip in the Cyrano layout schema. At install time, once Cyrano has locked onto the multiview layout, it runs OCR across each tile's name strip and populates the tile.label field of the event record. The review dashboard's camera filter chips are rendered straight from those labels. Property managers already know those names because that is what they see on the monitor. Reviewing by name survives channel swaps, cable re-runs, and multiview layout changes. Reviewing by channel number does not.

Median capture-to-reviewable latency is 7 to 8 seconds on a 25-tile multiview, with a 5 to 15 second envelope. That is end-to-end: HDMI frame grab, per-tile crop, overlay mask, classifier inference, event-row write, thumbnail render, push to dashboard. If someone steps into the mailroom alcove at 2:04:12 a.m., the row is sitting on top of the review feed at 2:04:20 a.m. This matters for overnight shift review, because by the time the morning shift takes over, last night is already fully reviewable. Nothing is batch-indexed at end of day.

Sometimes, which is why the DVR's own recording stays untouched as a fallback. Cyrano classifies person_in_zone, loiter, vehicle_dwell, tamper, and package_tamper today. About 99 percent of post-incident review questions a property manager actually asks are answered from the event list. For the 1 percent case (a subpoena asking for a specific two-hour window with no events, or the 5 minutes leading up to a flagged event) the full DVR recording is still there to pull from. Every event thumbnail links to the underlying clip on the DVR so you can open the full context without re-scrubbing.

Set the time range to the overnight window (say 10 p.m. to 6 a.m.), leave the camera filter empty to see every tile, and let the event-class filter show person_in_zone, loiter, and tamper. The thumbnail strip loads. A quiet night is five to fifteen thumbnails. Triage: glance at each thumbnail, click the one or two that look interesting, the full 30-second clip opens inline, reviewer decides whether to escalate. Typical shift review is 10 to 20 minutes of attention on a 25-camera property, replacing what was 3 to 4 hours of scrubbing.

Motion-triggered recording is the first thing every installer tries and it breaks in two predictable ways. First, it creates recording gaps, because motion detection at the camera level misses the start of slow events and drops segments in low light. Second, on a busy outdoor camera it triggers almost continuously, producing hours of flagged material that still needs to be reviewed. Motion is not a review tool, it is a storage tool. Review needs classification (person, vehicle, loiter, tamper) plus a timeline view across tiles. Motion gives you neither.

Most do. Hikvision's iVMS, Dahua's SmartPSS, Reolink's client, and the long tail of vendor apps all offer a motion-segment view, a playback timeline, and sometimes a weak per-camera object filter. Review of a single camera on a single DVR is serviceable in those apps. The moment you need to review two cameras at once, or ten, or a portfolio spanning four recorders from two vendors, each app becomes its own silo and cross-camera review becomes a tab-switching exercise. Moving review up into an event index layer sitting above the recorder makes the recorder brand irrelevant to the review experience.

Stop scrubbing. Write down the four things that actually matter for the event you care about: the tile name on the monitor (Mailroom, Lobby, Loading Dock), the event class (person, vehicle, loiter, tamper), the approximate time window, and the property. If your DVR does not let you filter on those four fields, you are not reviewing, you are watching tape. The answer is not faster scrubbing. The answer is a review layer on top of the recorder that builds the event list while the DVR keeps recording.