The architecture
Four hops outward, zero paths back
The pattern is deliberately plain, because plainness is inspectable. Data moves outward stage by stage; each boundary is enforced by the plant’s own security controls, and at no stage does anything on the outside initiate a connection inward.
A one-way flow: historian or read-only SCADA interface, to a DMZ relay, to protocol translation and context enrichment, to the analytics estate — with a fallback noting the plant is unaffected if the outward path fails.
- 01
The source of record
The historian, or a read-only SCADA data interface, inside the control zone. It already collects; we add no agent to it.
- 02
The DMZ relay
A staging node in the industrial DMZ receives pushed data. Where the plant mandates hardware enforcement, a data diode makes the one-way property physical rather than configured.
- 03
Translation and context
Outside the control network, proprietary protocols are translated — OPC UA as the common vocabulary — and raw tags gain units, ranges and their place in the equipment hierarchy.
- 04
The analytics estate
Contextualised process data lands in the warehouse or lake beside maintenance, lab and ERP data, where analysts can finally join it to the rest of the business.
If the outward path fails
The plant does not notice. That is the design’s central test: the control side has no dependency on the pipeline, so an analytics outage is an analytics problem, never an operations one. The relay buffers and backfills when the path returns.
The vocabulary problem
A tag called FT-1042A is not yet data
Control systems speak dozens of protocols, many proprietary, and translation between them is a settled category of work — OPC UA has become the common vocabulary most modern equipment and software can meet at, and where legacy interfaces cannot, gateway translation bridges them. The genuinely hard part comes after translation: a historian export is thousands of tags with names like FT-1042A, meaningful to the instrument engineer who named them and to nobody in the analytics estate. Context enrichment — attaching each tag’s unit, valid range, equipment, and place in the plant hierarchy — is what turns a signal archive into analysable data, and it is slow, careful work done with the plant’s own engineers, because they are the only people who know what the tags mean.
This is also where latency expectations get set honestly. An outward pipeline through a DMZ delivers data in minutes — sometimes less, sometimes more, depending on the staging pattern the plant’s security architecture requires — and that is entirely sufficient for trending, efficiency analysis, maintenance planning and reporting. It is not a real-time feed, and anyone who needs to react in seconds should be looking at the control room’s own displays, which is exactly where that job belongs. We state the expected latency per feed at design time so no consumer ever mistakes analytical history for an operator screen.
The ground rules
The conditions under which we go anywhere near plant data
These are not preferences; they are the terms of engagement, agreed with your operations and OT security teams before any design work starts.
Read-only, always
Extraction uses read interfaces the historian or SCADA vendor provides; no component we deploy holds write credentials to any control-zone system.
What it protects: The impossibility, not just the policy, of our work affecting a process.
Outward initiation only
Connections are initiated from the more-trusted zone outward; nothing in the DMZ or the enterprise network opens a session into the control zone.
What it protects: A control network whose attack surface we did not enlarge.
The plant’s security authority approves the design
The architecture is documented plainly and reviewed by your OT security organisation against its zoning model before implementation — approval is theirs to give, not ours to assume.
What it protects: One security posture, owned where it must be owned.
No dependency flows inward
The control side must run identically whether the pipeline is up or down; buffering and backfill live outside, and no operational procedure may come to rely on the analytics path.
What it protects: An analytics failure that is never an operations incident.
Our role
We build the window, never a door
The division of responsibility on an OT data project must be explicit, because ambiguity here is not a contractual nicety — it is a safety matter.
- 01
The outward pipeline
We design and build the extraction, DMZ staging, translation and delivery into the analytics estate, documented plainly enough for your security team to audit every hop.
- 02
The context model
Working with your instrument and process engineers, we build the tag-to-meaning mapping — units, ranges, hierarchy — that makes the data analysable, and we maintain it as the plant changes.
- 03
The declared service
Each feed is delivered with its latency, completeness and update behaviour stated, so every consumer downstream knows exactly what kind of truth they are reading.
Everything on the control side of the boundary — the control systems themselves, changes to them, and every safety approval on the control path — remains with your engineering authority and your control-system vendors. We integrate for visibility; we do not modify control systems, and we do not accept work that asks us to.
Common questions
What buyers ask
Will connecting the historian to analytics put our control network at risk?+
It must not, and the architecture is how that is made true rather than promised. Data leaves through paths initiated from the trusted side, staged in a DMZ, with no component outside holding any ability to write inward — physically enforced by a data diode where your policy requires it. We describe the design plainly and submit it to your OT security organisation for review against your zoning model, because a pattern that depends on nobody understanding it is not a security pattern. What we will not do is claim any architecture makes the plant secure; your security posture remains owned by your own OT security team.
How fresh is plant data by the time analysts can query it?+
Minutes is the honest general answer, and the exact figure per feed is set — and stated — at design time, because it depends on the staging your security architecture requires and the batching that keeps the path efficient. That freshness is right for what the analytics estate is for: trends, efficiency, maintenance planning, reporting. It is deliberately not a real-time channel, and building one would mean weakening exactly the boundaries the design exists to keep strong. Second-by-second reaction belongs to the control room and stays there.
Our tags are twenty years of undocumented naming. Is the data still usable?+
Yes — this is the normal condition, not a special problem. Every long-running plant has tag names that encode a naming convention three revisions old and the memory of engineers who may have retired. The context work goes in priority order: we start from the analytical questions you want answered, identify the tags those questions need, and document that subset properly with your engineers rather than attempting a heroic full-catalogue cleanup that never finishes. The mapping grows use case by use case, and each documented tag stays documented — so the asset compounds instead of resetting with every project.
In this cluster
Next step
Walk us through your estate
An architecture session with your operations and OT security teams that draws the outward data path on paper — sources, DMZ stages, protocols, latencies — before anything touches a network.

