Integrated observation systems, designed end to end
CloudWitness is the engineering practice of A Cloud Of Witnesses LLC. We design computer integrated systems that observe a site, move its signals safely and produce evidence that outlives the moment. Every service below is delivered together, as one accountable arrangement.
Observation Network Integration
An observation network is the connective tissue of the whole chamber. We begin by surveying the structures, power rooms and cable routes that already exist, then design a network that carries every observing device on a documented path. Cameras, gateways, environmental probes and edge nodes are assigned to segmented virtual networks so that a fault in one branch cannot cascade into another. We specify industrial switches with appropriate temperature ratings, plan power over Ethernet budgets carefully, and install surge protection on every external copper run.
Documentation is part of the deliverable, not an afterthought. Each drop, patch panel port and device address is recorded in a schedule that a future technician can follow without guesswork. We configure managed switching with port security, monitor link health continuously, and provide a simple topology drawing that shows how a signal travels from its sensor to its destination. When an installation spans multiple buildings, we design the inter-building links and the failover behaviour that keeps the chamber observing during a single path failure.
The outcome is a network that behaves like the felt-lined rim of an observation box: quiet, ordered and reliable, holding every moving part in its place so the events inside remain visible.
Sensor Telemetry Pipelines
A telemetry pipeline is the route a reading takes from a probe to a place where it can be trusted. We select sensors to match the physical quantity being measured, then build the ingestion path that carries those readings without loss. Protocols such as Modbus, MQTT and simple serial streams are normalised into a single internal schema, so downstream consumers do not need to know which device produced a value.
Reliability is engineered into the pipeline rather than assumed. Edge collectors buffer locally when the network is unavailable and replay their backlog once it returns. We define schema contracts so that a firmware update cannot silently change the meaning of a field. Duplicate suppression and clock alignment keep the stream coherent, and every record carries enough context to be interpreted years later, including device identity, calibration state and the exact moment of capture.
For sites with high sample rates we tune sampling windows, decimation and aggregation so that the pipeline stays within its budget while retaining the detail that matters. The result is a stream of readings that reads like a set of vapour trails: continuous, comparable and impossible to mistake for one another.
Event Recording Systems
Recording systems capture what happened in a form that can be reviewed afterwards. We design the recording layer so that video, audio, instrument output and metadata are bound together on a common timeline. Time synchronisation across all recorders is mandatory, because an unsynchronised record is an argument waiting to happen. We use network time sources, verify drift at commissioning and recheck it during stewardship visits.
Retention rules are defined with the operator before a single byte is written. We set separate policies for short-term operational review and long-term evidential holding, and we make sure the two never collide. Storage media is chosen for sustained write endurance, and arrays are configured so that a single disk failure degrades performance rather than ending the record. Where a site requires it, we implement recording failover so that a failed unit hands over to a standby within seconds.
Retrieval is designed with the same care as capture. Operators can jump to a moment by event, by alarm or by time, and export a clip with its surrounding context intact. A recording that cannot be found quickly is barely a recording at all.
Edge Analytics Deployments
Edge analytics move interpretation out of the data centre and into the chamber itself. When bandwidth, latency or privacy rules limit what may leave a site, we place small compute nodes close to the sensors so that classification, counting and filtering happen locally. Only the results that matter travel onward, which reduces network cost and keeps sensitive imagery on premises.
We size the edge hardware to the workload, choosing ruggedised enclosures where the environment demands them and planning thermal headroom so that sustained inference does not throttle. Models are packaged with their dependencies and their versions recorded, so that a result can always be traced to the exact software that produced it. We implement health checks that detect a stalled or overheated node and raise an alarm rather than silently dropping analysis.
Because edge decisions are consequential, we favour transparent, testable logic and we validate it against recorded data before it goes live. When the model sees something, the trail it draws should be one that an operator can understand and defend.
Archive and Audit Trails
An archive is the memory of the observation box, and an audit trail is the proof that the memory has not been tampered with. We build tiered storage that moves older material from fast media to economical media while keeping it retrievable. Integrity checks run on a schedule so that bit rot is discovered early, and a copy is kept in a separate location whenever the retention obligation justifies it.
Every access is logged. The audit trail records who viewed a record, when they viewed it and what they did with it, producing a chain that stands up to review. Where a regulation demands immutability, we apply write-once retention windows so that a record cannot be altered before its disposition date. Deletion is itself an audited event, scheduled and documented rather than improvised.
Search is the point of all this effort. We build indexes that let an investigator find a record by time, device, site or event type, and we export results in formats that survive the software that produced them. Evidence should outlive the tools that captured it.
Dashboard and Alarm Builds
A dashboard is where the chamber finally faces its operator. We design views that answer the questions an operator actually asks: is the system healthy, is anything outside its expected range, and what needs attention first. Live state, trends and historical context sit together so that a decision can be made without hunting across five different tools.
Alarm logic is deliberately restrained. We tune thresholds against real baseline data, add dwell times and hysteresis to suppress transient noise, and separate informational notices from genuine escalations. Each alarm carries a plain statement of what changed, why it matters and what the operator should do next. Escalation paths are defined in advance, with named responsibility at each tier so that nothing waits for a person who is not there.
We test alarm behaviour by injecting realistic conditions before commissioning, and we review the alarm history with operators during stewardship. A system that cries wolf is ignored, and an ignored system is blind. Our dashboards are built to preserve attention, not to spend it.