Fleet Storage Telemetry — Methodology

Sample window

2026-05-15T00:00:00Z → 2026-05-22T00:00:00Z — one full week. Per-trailer windows clipped to each trailer's actual archive range (one trailer had 154 of the 168 hours; the rest had the full 168).

Sample size

• Trailers: 4 (below the playbook's N≥8 bar — explicit pilot).
• Cameras: 16 total.
• Site-type buckets with fewer than 4 cameras are marked insufficient and omit percentiles (not zero-filled).

Site-type taxonomy

Each trailer was classified by visual inspection of one live frame per camera (full-native-resolution JPEG, no playback). The classifier uses a 7-bucket scheme:

• perimeter-secure — chain-link / razor-wire perimeters, secured yards, locked utility cabinets
• warehouse-distribution — loading bays, roll-up doors, forklift activity, fleet-vehicle staging
• retail-parking — commercial parking lots, painted parking spaces, mixed pedestrian / vehicle activity
• jobsite-construction, port-industrial, corporate-campus, off-grid-rural — (zero cameras in this pilot window)

Customer brands, deployment addresses, and trailer identifiers are excluded from this published methodology.

Byte estimation

Headline: trailer-disk-allocation derivation

Source: the fleet's pre-allocated VDB-slot configuration as of 2026-05-22T13:56:21Z. Each entry carries the trailer's pre-allocated VDB slot size (~6.9 TB — total disk allocation, not bytes written this week) plus an externally-computed usage GB/day figure (pre-allocated / retention-days). The usage GB/day field is the load-bearing input.

• Per-trailer headline: gb_per_trailer_per_week = gb_per_day × 7.
• Per-camera allocation: gb_per_camera_per_week = gb_per_trailer_per_week × (recorded_minutes_camera / sum_recorded_minutes_trailer). When the trailer captured no motion in the window, fall back to uniform / n_cameras.

This is not a true two-timestamp byte delta — only one VDB snapshot was taken — but on motion-only NVRs with FDC's temporal compression it is materially closer to real archive consumption than the snapshot-sampling alternative described below.

Cross-check: snapshot sampling

For each trailer, archive timeline coverage was queried at 1-minute resolution to derive recordedSeconds per camera; then ~100 native-resolution JPEG snapshots were pulled per trailer and the mean frame byte size was multiplied by an assumed 900 frames/minute (15 fps × 60 s).

Why this is a cross-check, not the headline: these NVRs record on motion only and FDC archive uses temporal compression. Multiplying a full-size JPEG snapshot byte count by a continuous-recording rate materially over-states the on-disk cost — the pilot saw snapshot-sampling at roughly 3× the disk-allocation figure for one trailer. Published as calibration only.

Results

Site type	N cameras	Median GB/cam/wk	p25 – p75	Motion density
Perimeter / secured	4	442	379 – 510	29.5%
Warehouse / distribution	8	2,453	1,447 – 3,418	44.0%
Retail / parking	4	2,923	2,753 – 3,050	45.9%

Privacy / publication rules

• No per-trailer rows are published.
• Trailer identifiers, hostnames, IP addresses, and customer brands are excluded.
• Camera snapshot JPEGs from the site-tagging pass are kept out of published artifacts — only aggregate frame-size statistics inform the cross-check.

Excluded buckets

Site-type buckets with fewer than 4 cameras (jobsite-construction, port-industrial, corporate-campus, off-grid-rural in this pilot window) are omitted rather than zero-filled. The full N≥8 trailer roll-out scheduled later in 2026 will cover more buckets.

← Back to fleet telemetry table on /technical-specifications