The word criminal in criminal trespass lives in one field on the event. Here is the field, and the six values it takes.

Civil trespass is a tort: the owner of the property sues the intruder for damages or an injunction, the standard of proof is preponderance of the evidence, and the remedy is usually money. Criminal trespass is a prosecutable offense under a state penal code or a federal statute: the state files charges, the standard of proof is beyond a reasonable doubt, and the penalty is a fine, jail, or both. Every state's criminal trespass statute adds at least one element beyond presence on the property, and the most common added element is notice. The subject had to know (or had constructive notice) that entry or remaining was not permitted. That is the notice element, and it is the line that separates a civil tort from a criminal charge. Without evidence supporting the notice element, a charging prosecutor usually declines the case.

The taxonomy varies by state, but the patterns collapse into six forms. One, posted_sign: a visible no-trespass notice at the entry point. Two, fenced_enclosure: a physical fence, wall, or enclosure that is plainly designed to exclude. Three, locked_entry: a gate, door, or barrier that is locked or otherwise physically controlled. Four, personal_communication: a verbal or written demand to leave given by the owner or an authorized person. Five, prior_warning_served: a previously served written trespass warning on file. Six, marked_boundary: a painted line, colored marker, or distinctive paint mark on trees or posts that several state statutes (Texas, Arkansas, Oklahoma, Florida, among others) explicitly accept as notice. A charging decision names which of the six the footage supports. A plain DVR cannot name any of them without a human scrub.

During install, after plugging the unit into the DVR's HDMI out, the operator draws the zone polygons through the dashboard (gates, lobby ingress, parking rows, rear lots, posted fence lines, leasing office approaches). Each polygon carries a notice_type tag drawn from the six values: posted_sign, fenced_enclosure, locked_entry, personal_communication, prior_warning_served, marked_boundary. A polygon can carry more than one tag (for example a gate that is both locked and posted). The tag is set once and sticks with the zone. Every track that crosses the polygon inherits the tag on its event record.

The array is keyed on the zone's notice_type and is populated with the measured behavioral evidence that supports it. For a posted_sign zone, the array includes sign_polygon_id, orientation_dwell_ms (how long the track's bounding-box orientation aligned with the sign), and the approach path. For a fenced_enclosure zone, it includes enclosure_geometry (a shape descriptor drawn at install) and path_break_at_boundary, a true/false on whether the track's heading broke exactly at the fence line. For a locked_entry zone, it includes the entry_state (locked, unlocked, forced_open) at the moment of the crossing, and for a tailgating event the follow_gap_seconds. For a personal_communication zone, it includes the staff_presence_timestamp and the dwell_after_verbal_warning_seconds. For a prior_warning_served zone, it includes a prior_warning_track_match pointer back to the earlier clip where the warning was served. For a marked_boundary zone, it includes the marker color and the marker_polygon_id plus orientation_dwell_ms. These are the fields a charging affidavit cites, not human prose.

The natural-language index accepts a description or attribute query (for example, man in dark jacket with black backpack, east gate, between 2 and 4 AM, any night this week). The index returns candidate tracks across the full retention window on the DVR. For the prior_warning_served element, the workflow is: pull the current event, open the natural-language query panel, paste the subject's description from the current event, and restrict the search to prior events on the same property. The returned candidates show the prior clip where the warning was served. The prior_warning_track_match pointer on the current event is produced from that retrieval. The system does not persistently store a face embedding across sessions; the match is produced at query time from the existing DVR footage.

At least a dozen states, with Texas, Arkansas, Oklahoma, Florida, Missouri, Kansas, North Carolina, South Carolina, and Tennessee among the most specific. Texas Penal Code 30.05(b)(2)(F) names purple paint on a post or tree as effective notice. Arkansas Code 5-39-201 accepts purple paint. Florida recognizes distinctive signs on fences and gates. Oklahoma accepts purple paint under Title 21 Section 1835.2. The details vary: some states specify the height and spacing of the marks, some require a particular color, some accept any color if distinctive. The practical implication for footage is the same across statutes: the clip has to show that the marker was in view at the crossing point. A marked_boundary zone with orientation_dwell_ms on the marker_polygon is the evidentiary record.

Because a DVR records pixels without a tag that says where the pixel was, what the pixel showed, or whether the subject in the pixel noticed it. The recording is raw video plus a channel number. A question like, did the subject approach a posted sign, does not have a field on the file to answer; a human has to scrub the tape, identify the sign, identify the subject, and narrate what they saw. That narration does not rise to the evidentiary bar of a retrievable field on a payload. It is reconstructive testimony, which is how contested trespass packets die. Cyrano adds the tagged field at the source, at the moment the track crosses the zone, so the field does not need to be reconstructed later.

No. One Cyrano unit plugs into the existing DVR or NVR over HDMI, reads the composite multiview that recorder is already outputting to the guard or leasing-office monitor, and runs per-tile tracking and event tagging on that composite. No camera replacement, no RTSP negotiation, no firmware change. A single unit indexes up to 25 camera feeds. The evidence chain still terminates at the DVR's native recording, which is the chain courts have already accepted in prior trespass prosecutions. Cyrano is the retrieval and event-tagging layer on top of that chain.

Target band on a 16 to 25 camera property with a 48-hour query window is under 60 seconds from a plain-English query to a ranked candidate list with the notice_evidence array attached. The packet that leaves the building is: the DVR's native clip segment, the zone polygon, the notice_type on the zone, and the notice_evidence array on the event. An investigator can attach the packet to a charging affidavit and point at the exact statutory subsection the notice_type answers. Compared to a 4 to 6 hour manual scrub on a consumer-grade DVR, that is roughly a 200x reduction in the retrieval step that determines whether a case gets charged.

Hardware is $450 one-time per unit. Software is $200 per month starting in month two. Install time is under 30 minutes per property: plug the unit into the DVR's HDMI out, connect it to the property network, draw the zone polygons, tag each polygon with a notice_type, and the unit begins indexing. One unit covers up to 25 camera feeds. For context, a single dedicated contract guard on a 12-hour post runs $3,000 to $5,000 per month per property and cannot see every feed at once. The product ships the behavioral evidence chain that criminal-trespass charging requires, on the DVR that is already in the closet.