A security property is a number, not a thing. It is measured in watched-tile-seconds per day, and most sites score closer to zero than they think.
The first page of Google for “security property” is a sequence of component lists: cameras, access control, alarms, intercoms, dashboards. None of those lists answer the one operational question that matters. How many of a property’s camera-seconds each day are actually being watched by something that can raise an alert? This guide gives that question a unit, walks through the arithmetic, and shows how a single HDMI cable into an existing DVR moves a typical multifamily site from about 3% coverage to a constant 100%.
See a live watched-tile-second count on a real DVRThe SERP for “security property” has one specific blind spot
Open the first ten results. You will find Pelco, Allstar Security, Monitex, Deep Sentinel, Avigilon, Mammoth Security, Omnilert, AAOA, Multifamily Executive, and Silva Consultants. Every one of them organizes the page around a list of components. Video surveillance. Access control. Alarms and sensors. Communication systems. Unified management platforms. The writing is fine, and the lists are mostly correct. But there is a specific question none of them answer.
“If I install all five of those components tomorrow, how much of my property is being watched per second?”
That is the question a property manager is implicitly asking when they type this keyword. They are not sourcing a shopping list. They are trying to figure out whether they have a security property or just a recording property. The two words sound the same. They are not. Recording accumulates evidence after the fact. Security intervenes before it. The variable that separates them is the attention factor on each camera tile, and that variable has a unit: tile-seconds per day.
“Cameras record 24/7 but nobody watches the feeds. Checking footage after an incident can take hours or days of manual scrubbing.”
Cyrano product description, llms.txt
The five variables that define a security property
Every operational definition of “security property” reduces to these five variables. Multiply them and you get a scalar per-day score for your site. Every product in the category is trying to move one or more of these numbers; most of them only move the first.
1. Tile count (C)
How many camera tiles the site has in total. The DVR's multiview grid is usually the honest answer: the number of tiles it draws on the monitor. Typical multifamily sites run 8 to 25. Large jobsites can reach 25 on a single recorder. C is the only variable vendor product pages discuss.
2. Attention factor (A)
Fraction of each tile's seconds that something capable of raising an alert is actually attending to that tile. A single guard at a 16-tile monitor delivers A ≈ 0.02 to 0.06. Event-triggered remote monitoring delivers A ≈ 0.05 to 0.15. HDMI retrofit AI delivers A = 1.0 on every tile.
3. Duty-cycle seconds (S)
How many seconds per day the property needs watched. For most multifamily, construction, and commercial sites, S = 86,400 (the whole day). For sites with daytime leasing staff plus after-hours risk, the S that matters is the 40,000 to 50,000 unattended seconds.
4. Latency to response (L)
Seconds from a tile's first anomalous pixel change to a human or system action. L gates how much of the watched-tile-second is convertible into a prevention rather than a documentation. 1.1s on HDMI retrofit. 30 to 300s on remote monitoring. 600 to 1200s on guard-dispatch paths.
5. Event bandwidth ceiling (B)
Bytes per event leaving the property. Constrains what capture architectures work on the site's uplink. ~240 KB per event on HDMI retrofit. Continuous full-frame cloud AI can hit gigabytes per camera per day and disqualifies itself on cellular-backed jobsites.
The arithmetic, on a typical 16-camera multifamily site
These are the numbers behind the score. The day is a constant. The tile count is whatever the DVR grid says. The attention factor is the lever that every security purchase is implicitly buying, and the number vendors almost never cite. This is what the arithmetic looks like in four rows, one per common approach, at the same 16-camera site.
The attention factor for a single guard is bounded above by 1/C, because a human can fixate on one tile at a time. The multiview is the operational upper bound for a human watcher; no amount of training moves the ceiling. An AI detector attends to every tile in parallel, which is why A = 1.0 is reachable by the architecture and unreachable by the staffing.
Where the watched-tile-seconds come from, physically
The diagram below shows the actual physical path. The cameras on the left already feed into the DVR, which already draws the composite frame. The new thing is the HDMI cable in the middle. On the right, tile-seconds get produced at a constant 1.0 attention factor and are converted into event packets only when a zone rule fires.
How tile-seconds are generated from an existing DVR
No camera is replaced. No firmware is flashed. No RTSP credential is needed. The HDMI signal the DVR is already emitting to a guard monitor is the signal the adapter reads. The tile-seconds are a byproduct of the inference pass; the event packets are only triggered when a configured zone rule matches.
“Security property” as a noun vs as a measurement
The phrase is doing two different jobs in English right now. One is a label for a building. The other is a measurement of how much of that building is being attended to per second. The difference shows up as soon as you try to answer “is this a security property?” with a number.
Two ways the phrase is used
A security property is a building that has security things installed on it. Cameras. Access control. An alarm panel. Maybe a guard post near the entrance. The label is satisfied by the presence of components. The property manager can point at them. The board can see them during the tour. Everyone nods.
- Satisfied by component presence, not by coverage
- Does not distinguish between recording and watching
- Cannot answer: how many tile-seconds per day?
- Rewards procurement, not operations
Four approaches, scored at the same 16-camera site
If “security property” is a measurement, the right comparison is in the measurement’s units. Below is the same site, same cameras, compared across four common approaches. The rightmost column is the annualized dollars per million watched-tile-seconds. That is the price of the output, not the price of the input.
Four security-property configurations, scored in tile-seconds
Same 16 cameras, same 86,400 seconds in a day. The only variable is the attention factor and the latency it is paired with.
| Feature | Guards, mobile patrol, and remote human video | HDMI retrofit AI on existing DVR (Cyrano) |
|---|---|---|
| Watched-tile-seconds per day | 0 to ~138,240 (A=0 to 0.10) | 1,382,400 (A=1.0) |
| Latency from pixel change to phone | 30 s to 20 min | ~1.1 s |
| Cameras replaced | 0 (but gear drift is common) | 0 |
| Install time on a running DVR | Weeks (guard RFP) to immediate (patrol) | Under 2 minutes |
| Off-property bytes per event | Continuous stream or none | ~240 KB (thumb + clip + JSON) |
| Bounded by human fixation ceiling (1/C) | Yes | No |
| Annualized cost at 16 cameras | $12k to $220k | ~$2,850 (incl. hardware in year 1) |
| $/million watched-tile-seconds/yr | $87 to >$1,000 | ~$5.64 |
| Works on analog BNC cameras | Yes, if visible on monitor | Yes (DVR normalizes) |
| Scales across a portfolio without DVR harmonization | Varies | Yes (HDMI is universal) |
How to measure your own security property this afternoon
The score is knowable without buying anything. The walk below produces a current watched-tile-seconds number for the site you manage today. If the number embarrasses you, that is information. If it does not, the current posture is defensible and you are done.
Scoring your site in an afternoon
1. Count the tiles your DVR actually draws
Walk to the DVR. Look at the monitor. Count the tiles on the screen. That is C. Do not count cameras the DVR lost connection to six months ago, even if they are still on the invoice. C is the honest version, not the paid version.
2. Identify who or what is attending to that monitor right now
If it is a guard, write down the shift hours. If it is a leasing agent, write down the hours the agent is at the desk and facing the monitor. If no one is attending it, write A = 0. This is the second most lied-about number at a multifamily property, after occupancy.
3. Multiply
watched_tile_seconds_per_day = C × A × S. For one guard on a 16-tile monitor for an 8-hour shift, that is 16 × 0.0625 × 28,800 = 28,800. Note that A is capped at 1/C for a single human watcher. The ceiling matters.
4. Add your latency to response
From the guard's first glance at the anomalous tile, how long until someone is acting on it? If the guard has to call dispatch, dispatch has to call a patrol, the patrol has to arrive, that is L = 600 to 1200 seconds. Write L next to the watched-tile-second number. The two numbers travel together.
5. Compare to the HDMI retrofit ceiling
The retrofit ceiling at the same site is C × 1.0 × 86,400, which is 1,382,400 at 16 cameras, paired with L ≈ 1.1 s. The gap between your current number and that ceiling is the gap between “has cameras” and “has a security property” in the measured sense.
Does your property currently secure, or only record?
The difference reduces to a handful of yes/no questions. Each “no” is a tile-second you are paying for and not collecting. Each “yes” moves the property along the coverage curve. The questions are the shape of the diligence that a plaintiff’s negligent-security expert would walk a jury through in 2026.
The diligence questions that convert recording into security
- Is something attending to every tile on the DVR every second, or only some tiles some of the time?
- When a rule fires (person in a restricted zone, package dwelling past a threshold, tailgate at the garage gate), does an alert reach an on-call phone within single-digit seconds?
- Is the attention factor on each tile independent of whether a human is physically present at the monitor at that moment?
- Does the system work without replacing cameras, rewiring infrastructure, or reconfiguring the DVR vendor's cloud app?
- Does only a bounded, predictable payload (thumbnail + short clip + JSON) leave the property per event, with no continuous full-frame upload?
- If the new device unplugs tomorrow, does the guard monitor keep working and the DVR keep recording, so the failure mode is strictly additive?
- Can you produce the site's watched-tile-seconds-per-day score for this month, this quarter, and this year on demand, for insurance and legal review?
Score your property in a 15-minute call
Bring your DVR brand, your tile count, and roughly what you pay for current monitoring. We'll compute your current watched-tile-seconds number with you on a screen share, then show what the HDMI retrofit ceiling would be on the same cameras.
Book a 15-minute walkthrough →The one number at the bottom of the page
At a typical 16-camera multifamily property on a five-year-old DVR, with one 8-hour guard shift at a monitor, the current watched-tile-seconds-per-day score is roughly 28,800. The HDMI retrofit ceiling on the same cameras is 1,382,400. The multiplier is:
more watched-tile-seconds per day, on the same physical cameras, for about 2% of the cost of the next-cheapest comparable-coverage path
Frequently asked questions
What is 'security property' when you use the phrase operationally, not as a marketing term?
Operationally, a security property is the scalar number of watched-tile-seconds your site produces per day. A tile-second is one camera view being attended to for one second by something capable of raising an alert. Multiply the number of cameras by the number of seconds each one is attended to, sum across the 86,400 seconds in a day, and you have the number. A 16-camera site with zero active monitoring produces zero watched-tile-seconds per day, even if every camera is recording. A 16-camera site with one 24/7 guard posted at a monitor produces roughly 86,400 watched-tile-seconds per day (one tile of attention, all day). A 16-camera site with a retrofit that watches every tile every second produces 1,382,400 watched-tile-seconds per day. The address on the sign is the same. The security property is not.
Where does the ~3% coverage figure for traditional setups come from?
Three sources. First, a stationary guard watching a 16-tile monitor can only fixate on one tile at a time, which is 1/16 or 6.25% of the grid, and in practice fixation drifts and drops to 2 to 4% attention-normalized. Second, mobile patrols typically hit 2 to 4 stops per night at 5 to 10 minutes per stop, which is 2 to 5% of the night covered on any given camera. Third, alarm sensors cover doors, windows, and glass-break, which is about 10% of the physical surface area of a property but fires only on trip events and contributes near-zero continuous tile-seconds. Average across the three and the working number is under 5%. Most multifamily sites sit closer to 3%.
Why is the DVR's HDMI multiview output the relevant capture point for this calculation?
Because it is the one cable in the building on which every camera already appears. A DVR draws a 2x2, 3x3, 4x4, or 5x5 grid (tiles) and emits it over HDMI so a guard can watch a monitor. Whether a given camera is analog BNC, PoE IP, or wireless, the DVR has already normalized it into a tile inside the composite frame. A device that reads that HDMI signal gets all 25 tiles in one pass, so one inference pass per composite frame generates detections on every tile simultaneously. That is how you push the attention factor on each tile from a fraction to a constant 1.0.
What actually leaves the property when a detection fires?
Exactly three artifacts per event. A thumbnail crop of the triggering tile, roughly 320x240 pixels, about 18 KB. A six-second clip of that same tile bracketing the detection moment, about 220 KB. A JSON metadata blob with the zone identifier, dwell time in seconds, wall-clock time, class (person, vehicle, package), and confidence score. Total ~240 KB per event. The full composite frame, the other 24 tiles that did not fire, any audio, any raw telemetry, none of that crosses the property boundary. The calculation is deliberate: 240 KB fits comfortably inside cellular uplink budgets at construction jobsites, and it is small enough that a privacy-sensitive property manager can recite exactly what leaves the building when asked.
Is 'security property' the same as 'secured premises' in the legal sense of landlord duty of care?
No, and that is the entire point. The legal phrase describes a duty: the landlord owes a tenant reasonable measures given foreseeable risk. The operational phrase is a measurement: the site currently produces N watched-tile-seconds per day. Courts have not started reasoning in tile-seconds yet, but insurers and risk managers already do, under different names. The gap is closing. If a property has prior-incident notice, a 16-camera DVR, and is still producing a three-digit number of watched-tile-seconds per day, a plaintiff's expert is going to point at the unused HDMI port on the back of the DVR and call that unreasonable. The measurement makes the argument concrete in a way the legacy phrasing does not.
Does the number have to come from AI? Could a remote human-monitoring service produce the same tile-seconds?
In principle yes, in practice no. A remote video monitoring service charges per watched-hour per camera. Covering 16 cameras 24/7 with active human attention is 384 human-hours per day, which prices in the range of $36,000 to $120,000 per year per property. Most operators buy a small fraction of that and route the rest to exception monitoring triggered by motion events, which means the effective attention factor on each tile sits well below 0.1 for most of the day. AI on the HDMI output runs inference every frame on every tile regardless of whether motion is detected, which keeps the attention factor at a constant 1.0. Same metric, different cost structure.
How quickly can a property be converted from the low-coverage regime to the 100% regime?
Under two minutes of physical install on a running DVR, plus a few minutes of calibration. The five motions are: unplug the HDMI cable from the DVR to the guard monitor, plug it into the Cyrano unit's HDMI input, run a short HDMI cable from the unit's passthrough to the monitor, connect ethernet, plug in power. The guard monitor keeps showing picture the whole time because the passthrough is live. Calibration is a one-time step where the installer labels which tile is which camera and draws zero-masks over the DVR's clock, channel bug, and per-tile name strip so the detector does not fire on overlays. First detection can fire within five minutes of the cable going in.
What is the event-to-phone latency, and why does it matter for the measurement?
About 1.1 seconds end-to-end in typical deployments. Latency matters because watched-tile-seconds measure potential coverage. The realized value of a watched tile is gated by how quickly a human can act on a detection. At 1.1 seconds from frame to phone, an on-call operator is still inside the pre-action window for most incidents at a multifamily property (the median package theft is 60 to 120 seconds from approach to departure). At 30 seconds of latency, you are documenting. At 10 minutes (the rough average for guard dispatch from a remote monitoring center), you are reviewing. Coverage density at the frame has to be paired with response latency at the phone for the tile-seconds to translate into prevented incidents.
What about privacy? 100% coverage of every tile sounds invasive.
Three design points blunt the privacy concern. First, inference runs on the adapter at the property, not in a vendor cloud. No continuous full-frame stream leaves the building. Second, only the three per-event artifacts cross the property boundary, and only when a rule actually fires, not on every person who walks past a camera. Third, no biometric index is constructed. No face embeddings are saved, no plate strings are extracted, no gait vectors are stored. The system classifies at the category level (person, vehicle, package) and fires on zone-and-dwell rules the operator configured, which is the behavior most jurisdictions treat as commercial video surveillance rather than biometric surveillance.
What does the whole-property monthly number look like at a typical 16-camera multifamily site?
Hardware is a one-time $450 for the adapter. Software is $200 per month starting month two. At 16 cameras that works out to about $13 per camera per month. Compare to remote human video monitoring at $36,000 to $120,000 per year, a 24/7 guard post at $150,000 to $220,000 per year, or a full camera rip-and-replace to a smart-camera vendor at $50,000 to $100,000+ in capital plus ongoing cloud subscription fees. The same watched-tile-second output, on the same cameras, for roughly 1 to 4% of the cost of the next-cheapest path that produces comparable coverage.
Comments (••)
Leave a comment to see what others are saying.Public and anonymous. No signup.