Controls engineering / Article / Architecture study 01

Protect the network.
Keep the water flowing.

A reference design for separating business systems, plant controls and remote pump stations—without sacrificing the operator’s ability to run the process.

Published 5 November 2026 · v1.0 · No live utility represented

Design example, not a completed engagement. This page uses an explicitly illustrative water utility to explain engineering decisions. No plant was accessed, no controls were changed, and none of the acceptance tests has been run. It is not an assurance or compliance claim.

The operating brief

One treatment plantTwo treatment cells, local HMIs, PLCs, central SCADA and a historian. Local process control should not depend on corporate email or cloud reporting.
Remote pump & tank sitesRTUs provide local control with engineered loss-of-link behavior. A protected telemetry path supplies central visibility and permitted supervisory commands.
A connected businessOffice reporting, third-party maintenance and facilities systems all have legitimate needs. Each gets a defined interface, not broad access to the plant.

Actual assets, protocols, water-quality limits, interlocks, redundancy and fail-safe behavior must be established with utility operators and the responsible engineers. “Stop every pump” is not a universal safe state.

Engineering heuristics / Review lens

Nine rules for defensible boundaries.

Use these proposed heuristics during design reviews. They are not a compliance checklist or universal operating instructions. The source-to-decision register below distinguishes published guidance from this reference design.

Rule 01

Start with consequences

Identify what failure or manipulation would do to water quality, supply, pressure and operator visibility.

Rule 02

Separate by function and trust

Group assets by operational need and consequence. Shared Ethernet does not imply shared access.

Rule 03

Own every connection

Record endpoints, service, purpose, approving owner, review date and the consequence of blocking it.

Rule 04

Separate observation from control

Use replicated reporting; distinguish read, command and programming permissions. A port allowlist alone cannot enforce all three.

Rule 05

Limit the reach of compromise

Scope accounts, engineering workstations and remote sessions to assigned systems. Review shared infrastructure as a common failure path.

Rule 06

Make isolation operable

Define services retained, local procedures, staffing, abort limits and decision authority before an incident.

Rule 07

Prove denial and continuity

Pair each negative security check with successful alarms, control or maintenance. Unspecified thresholds mean Not ready.

Rule 08

Prove restoration

Restore version-identified configurations on isolated equipment and verify behavior; successful backup jobs are not sufficient evidence.

Rule 09

Expire exceptions

Record a named owner, rationale, compensating measures, expiry and retest. Do not let temporary broad access become permanent.

Process first / Network second

What are we actually protecting?

This simplified surface-water treatment example connects process stages to controls and consequences. Other treatment methods and site layouts will differ. Chemical storage, detailed piping, interlocks, bypasses and instrument loops are deliberately not specified.

Follow the water. Then place the controls.Illustrative conventional surface-water treatment train • not a P&ID or operating procedure01 IntakeRaw-water pumpsFlow / level / pump statusCONTROL & BOUNDARYLocal PLC / motor controlsCell A boundaryCONSEQUENCE TO REVIEWLoss: intake interruptedWrong data: poor coordination02 ClarificationMixing / flocculationSettling / sludge handlingCONTROL & BOUNDARYProcess PLC / local HMICell A boundaryCONSEQUENCE TO REVIEWLoss: treatment impairedWrong data: dosing error03 FiltrationFilters / valves / backwashTurbidity / differential pressureCONTROL & BOUNDARYFilter PLC / local HMICell B boundaryCONSEQUENCE TO REVIEWLoss: capacity reducedWrong data: quality obscured04 DisinfectionDosing / contact storageResidual / flow / levelCONTROL & BOUNDARYDosing PLC / local protectionDedicated reviewed subzoneCONSEQUENCE TO REVIEWLoss: quality at riskWrong data: unsafe dosing05 DistributionClearwell / pumps / tanksPressure / level / flowCONTROL & BOUNDARYLocal RTU / pump controlsRemote-site boundaryCONSEQUENCE TO REVIEWLoss: supply impairedWrong data: tank/pressure riskBlue arrows: simplified physical water flow. Dashed lines: control associations, not network links.SCADA supervises across approved conduits; local controllers execute the engineered sequences.Protective functions, quality limits, chemical systems and safe states require process-specific engineering.

Cell placement is a proposed refinement of the overview: disinfection needs its own reviewed boundary. It is not an assertion that Cell B or a separate safety instrumented system fits every facility.

Download process diagram · SVG
Incorrect data is different from missing data. A clearly stale pressure value may trigger an operator response; a plausible but wrong value can mislead control and decision-making. Review both conditions and the independent means of detecting them.

Operation / Containment / Recovery

Change access without improvising the process.

Each mode is a pre-engineered policy, not a command to disconnect equipment. Restrictions should target affected services while preserving only the paths operations can still trust and safely use.

Operating modes: preserve essential control.Proposed communication policies • not a live status display • all transitions require named authorityNORMALApproved routine operationBusiness dataAvailable via DMZVendor maintenanceApproved window onlySCADA ↔ RTURequired telemetry/controlLocal PLC / protectionEngineered operationWHO AUTHORIZES?Operations + access ownerauthorize support sessions.RESTRICTEDContain a suspected access pathBusiness dataSuspend if implicatedVendor maintenanceRevoke / blockSCADA ↔ RTURetain if trusted and neededLocal PLC / protectionApproved fallback onlyWHO AUTHORIZES?Incident lead + operationsselect a rehearsed policy.RECOVERYRestore trust before connectivityBusiness dataRestore after validationVendor maintenanceNew approval / credentialsSCADA ↔ RTUValidate, then reconnectLocal PLC / protectionVerify logic and behaviorWHO AUTHORIZES?Controls + operations approve;incident lead releases restrictions.Normal → Restricted: credible incident or approved precaution. Restricted → Recovery: scope contained.Recovery → Normal: signed functional and security checks. Failed validation → remain isolated / reassess.If telemetry itself is compromised, do not retain it blindly: use the site-specific incident and safety plan.

Download operating modes · SVG

Enter restricted mode

Record the incident or precaution, affected paths and authority. Validate local staffing, alarm availability and approved fallback. Apply the rehearsed policy; monitor process trends.

Enter recovery

Establish containment and configuration provenance. Validate replacements, credentials and restored logic on isolated systems. Prevent replay of stale supervisory commands.

Return to normal

Reconnect in stages. Witness security and functional checks, record exceptions and obtain named signoff. Unexpected behavior stops restoration and triggers the approved rollback or isolation plan.

Evidence required before any transition

Named decision-maker, time, policy/configuration version, trusted paths retained, alarm and process baseline, owner-approved tolerances, rollback trigger and operator acknowledgment. Tests T03, T05, T06, T08, T10 and T11 provide the existing review hooks. None has been executed.

These modes apply NCSC isolation-planning guidance to the reference utility; the particular retained paths and transition gates are our proposed design.

Worked example / PS-01 remote pump station

One station. Three operating conditions.

Follow the control responsibility—not just the network cable.

PS-01 is an illustrative distribution booster station with a local pressure sensor, pump drive, RTU and local operator interface. These identifiers belong only to this walkthrough; they are not a real asset inventory. The site-specific hydraulic design, pump arrangement and protection remain unresolved.

PT-01 pressure feedback → RTU-01; LOP-01 local operator panel ↔ RTU-01. Local control stays at PS-01. Arrows show conceptual communication, not wiring or a deployable configuration.

01 / Normal

Supervise centrally. Control locally.

PT-01 supplies local pressure feedback. RTU-01 executes the approved control sequence and permitted commands to VFD-01/P-01. SCADA-01 displays values, quality, operating mode and alarms through GW-01.

Operator: checks fresh telemetry and authority before an approved supervisory command. Programming is a separate F06 maintenance activity.

Evidence: baseline trends, command acknowledgments, access decisions and configuration versions. Verify an approved command works while an unauthorized programming attempt is denied on lab endpoints.

02 / WAN unavailable

Make lost visibility unmistakable.

After the approved timeout, SCADA-01 marks remote values stale and raises the configured communications alarm. RTU-01 follows its engineered loss-of-link behavior using valid local inputs; WAN loss alone does not define whether a pump should stop or continue.

Operator: follows the approved response and local staffing plan. Local alarms may remain available, but delivery to central SCADA cannot be assumed during the outage.

Evidence: simulator timeline, stale indication, local trends and response record. If PT-01 also fails, this is a separate fault requiring an approved response—not the same fallback.

03 / Recovery

Reconnect visibility before authority.

Confirm the cause is resolved and GW-01/RTU-01 configurations are trustworthy. Reconcile data quality, timestamps, current mode, active alarms and pending commands. Stale commands must not execute merely because the link returns.

Operator: witnesses current state before supervisory control is released under an approved sequence. A cyber incident also invokes the restricted-mode approval gates above.

Evidence: command lifecycle results, restored alarm delivery, access logs and named acceptance. Unexpected actuation or missing alarms stops the exercise and invokes the approved contingency.

Design gate—not a pump-control recipe. Timeout, pressure limits, sensor-fault behavior, fallback duration, restart conditions and remote/local authority are TBD. Operations and the responsible controls/process engineers must approve them. No setpoints or live operating instructions are supplied here.

Rehearse T06 before considering a site exercise

  1. Load version-identified RTU/HMI models into an isolated simulator; set the owner-approved limits and a healthy baseline.
  2. Simulate loss of the WAN path only. Record local behavior, SCADA stale flags, alarms and operator response.
  3. Separately model bad local sensor quality and unavailable local staffing. Confirm these conditions do not silently inherit the WAN-only response.
  4. Restore the simulated link with a pending command scenario. Verify fresh state, explicit authority and safe command rejection/expiry behavior before resuming supervision.
  5. Compare observations to signed criteria. Record deviations; do not advance to a live change window until they are resolved.

Evidence status: Proposed. Simulator-tested: no. Site-verified: no. This narrative describes expected behavior, not an executed simulation.

Requirements → decisions → evidence

Make every design choice traceable.

R-PS identifiers are proposed requirements for this example. Flow IDs refer to the existing communication matrix; T IDs refer to the acceptance plans. Each row remains Proposed / NOT RUN.

Need / requirementDecision & connectionVerification: denial + useful operationEvidence / accountable roleOpen acceptance gate
R-PS01
Local operation during WAN loss
RTU-01 owns the approved local sequence; F07 carries supervision.T06 / T08: isolate the simulated WAN; confirm approved local behavior and operator access.Logic version, trend and outage timeline / Controls + operationsFallback envelope, timeout and supported outage duration TBD.
R-PS02
Trustworthy operator visibility
SCADA-01 shows freshness and quality; no silent use of stale values.T06: missing updates produce stale status and alarm within the approved limit; current values return with validated quality.Timestamped HMI captures and alarm record / SCADA ownerFreshness limit, alarm latency and acknowledgment procedure TBD.
R-PS03
Limit changes to authorized engineers
Separate operational F07 from time-bound F06 programming; deny D01 office access.T01 / T04: unauthorized engineering access denied; approved monitoring and scoped maintenance still function in the lab.Rule counters, access logs, project diff / Controls + networkOEM enforcement of read/write/program permissions and endpoints TBD.
R-PS04
No unexpected actuation on recovery
Validate command freshness, current authority and pending state before release.T06: stale/expired commands do not execute; a newly authorized command works in the simulator.Command and acknowledgment timeline / Controls + operationsProtocol-specific expiry, queue and restart semantics TBD.
R-PS05
Recover known configurations
F11 protected backups for RTU, HMI and gateway; isolated restore first.T10: restore to spare/simulator, verify versions and expected control/alarm behavior.Hashes, restore duration and functional results / Recovery ownerRecovery time/point targets and compatible spare hardware TBD.
R-PS06
Retain required process protection
Define local protective functions and sensor-fault handling through hazard review.T11 / T06: specialist-reviewed simulation demonstrates approved response; access restrictions do not impair protection.Hazard review, cause/effect record and test traces / Process + controlsProtection design and independent validation required; ordinary RTU is not assumed safety-rated.

Guidance basis: existing NIST/NCSC decision register. These requirements and test mappings are our proposed application of that guidance, not quoted mandates.

Commissioning / Witnessed acceptance

Earn the right to reconnect.

A review checklist with gates, evidence and a stop rule—not a live switching procedure.

Current state: NOT READY. Named owners, limits and evidence are unassigned. All checks below are unexecuted. Marking a document complete cannot substitute for witnessed test results.

C01 / AuthorizeNOT RUN

Approve scope and process limits

  • Identify equipment, dependencies, owner, tester and operations witness.
  • Approve the change window, communications plan and exact affected paths.
  • Fill all pressure, latency, freshness, outage and recovery criteria; no TBDs.

Gate: signed scope and limits. Owner: operations + controls lead.

C02 / PrepareNOT RUN

Prove readiness and recovery

  • Capture passive baseline, current alarms and configuration versions.
  • Verify recoverable backups and isolated restore evidence (T10).
  • Confirm local staffing, access, contingency procedure and rollback authority.

Gate: restoration rehearsed; evidence retained. Owner: recovery + operations.

C03 / RehearseNOT RUN

Test on isolated equipment

  • Execute R-PS01–06 checks with approved thresholds.
  • Pair denied access with successful monitoring/control; test WAN loss, sensor fault and recovery separately.
  • Resolve unexpected outputs, stale command behavior and missing alarms.

Gate: witnessed results accepted; simulator limitations recorded. Owner: controls + network + operations.

C04 / PilotNOT RUN

Use a separately approved site plan

  • Only after C01–03, approve a limited site method with hold points.
  • Maintain operator authority and monitor process, alarms and load against baseline.
  • No unplanned cable pulls, live logic edits or broad scans; this page is not execution authorization.

Gate: paired security and process evidence meets signed criteria. Owner: site operations + change lead.

C05 / RecoverNOT RUN

Restore in controlled stages

  • Validate trusted configuration, telemetry quality and current operating mode.
  • Resolve pending commands before enabling approved supervisory authority.
  • Witness alarm delivery and permitted control; retain restriction if validation fails.

Gate: recovery tests pass without unexpected actuation. Owner: controls + operations; incident lead when applicable.

C06 / AcceptNOT RUN

Close with evidence, not confidence

  • Record results, timestamps, witnesses, versions and deviations against each R-PS requirement.
  • Resolve failed checks; document approved noncritical exceptions with expiry and compensating measures.
  • Obtain operations/engineering signoff, update as-built records and assign the next review date.

Gate: no unresolved safety-critical failure; named acceptance and handover. Owner: accountable utility owner.

Stop rule: unexpected actuation, loss of required alarms, invalid local sensing, process limits exceeded, or loss of operator control stops the exercise. The operator follows the approved contingency; the change lead invokes the pre-approved rollback only when safe. Reapplying a firewall configuration does not automatically restore a safe process state.

Minimum evidence record

Requirement/test ID · environment (simulator or site) · equipment and configuration versions · approved expected result and threshold · observed result · pass/fail/not run · UTC/local timestamps · tester and witness · evidence location · deviation and corrective action · signoff and next review date.

Keep actual site addresses, logic, credentials and detailed operating limits in the utility’s controlled records. A future public case study should contain only approved, sanitized evidence.

01 / Trust boundaries

Separate the process, not just the address ranges.

The industrial DMZ provides controlled exchange services. Behind it, the site operations, engineering, management and process zones have distinct access policies. Facilities, guest devices and safety-related systems do not inherit general production access.

Water utility network diagram showing enterprise and vendor entry to an industrial DMZ, a site firewall, OT service enclaves, separated treatment cells, remote telemetry and independent protective functions.

On a small screen, scroll the diagram horizontally for detail. Arrows show selected connection initiators, not all data directions. The flow matrix below includes engineering, logging and recovery paths omitted from this overview.

Download presentation diagram · SVG

VLANs define groups; enforced routing policy defines access. Avoid dual-homed workstations, unmanaged cellular routers and vendor tunnels that bypass the boundary. Boundary devices and management paths need their own tested recovery and redundancy design.

02 / Communication contract

Every permitted conversation has a purpose.

This is a proposed allowlist, not a deployable firewall configuration. All unlisted inter-zone traffic is denied by default after dependencies are discovered and approved. “TBD” entries are unresolved design gates; do not substitute broad port ranges. Replies on an established session are different from permission to initiate a new session.

IDInitiator → destinationDecision / serviceRestrictionAccountable ownerTest
F01Reporting users → DMZ replicaAllow: HTTPS 443, named application/usersRead replicated metrics only; no control API.IT + data ownerT01 / T02
F02OT historian exporter → DMZ brokerAllow: TLS-protected replication; exact port TBDNew sessions originate in OT; stateful replies allowed. Data export is not a hardware one-way boundary.OT + data ownerT02
F03Approved vendor → DMZ access brokerConditional: HTTPS 443 + MFANamed identity, device posture, site approval, short access window; no routed vendor VPN into OT.OT access ownerT03
F04DMZ broker → OT jump host → assigned engineering stationConditional: product-specific session transportSeparate allowlists on each hop. No broad desktop access or general file transfer; broker compromise remains a risk.OT engineeringT03 / T04
F05SCADA server → assigned cell PLC/RTUAllow: exact OEM control protocol, port and endpoints TBDOnly approved operational reads/writes. Verify direction with actual protocol. Port filtering alone does not distinguish read from write.Process + controlsT01 / T05
F06Assigned engineering station → assigned controllersConditional: exact OEM engineering services TBDApproved change ticket, backup, local operator authorization and time-bound access. No programming from office machines.Controls leadT04
F07SCADA master → remote RTUs through telemetry gatewaysAllow: selected telemetry protocol inside authenticated tunnel where supportedSite-specific addresses; no general branch-to-branch access. Legacy radio/serial requires protocol and latency review.Telemetry leadT06
F08OT update/transfer service → DMZ stagingConditional: approved HTTPS retrievalValidate signature/hash, malware checks, compatibility and change approval. Staging does not authorize installation.OT maintenanceT07
F09OT assets → local DNS / time / identityAllow: individually documented infrastructure servicesDo not share corporate privileged accounts. Specify ports and authentication behavior from real dependencies, not a generic all-services rule.OT platform ownerT08
F10OT collectors → DMZ log relay → enterprise SIEMAllow: approved TLS log transportBuffer locally; no SIEM-initiated management path back into controllers. Avoid loading fragile endpoints with agents.SOC + OTT09
F11OT backup service ↔ approved OT assetsConditional: documented backup transportDedicated credentials and endpoints; protected offline/immutable recovery copy. Validate backup load and restore on spare equipment.Recovery ownerT10
D01Enterprise / public internet / facilities → PLC or RTUDenyIncludes IPv6, wireless, cellular, alternate VPNs and dual-homed laptops.Network ownerT01
D02Cell A ↔ cell B; remote site ↔ remote siteDeny unless explicitly engineeredDocument process dependencies before enforcement. No any-to-any lateral route.Controls + networkT05
D03General OT / vendor sessions → safety engineeringDeny by defaultStatus-only interfaces and maintenance access require separate safety review; never assume an ordinary PLC is a safety system.Process safety ownerT11

Roles are proposed; named owners remain unassigned. For production, add exact addresses, protocol versions, rule IDs, change tickets, rule expiry, IPv6 policy and documented dependencies.

03 / Resilience walkthrough

What happens when a connection disappears?

These are design expectations and evidence requirements, not simulated test results.

Corporate IT unavailable

Keep local operations available

Design expectation: local PLC control and required OT services continue within the approved operating envelope. Business reports may become stale and must show that clearly.

Prove it in a lab: remove simulated corporate identity/WAN dependencies, exercise operator sign-in and verify alarms, time synchronization and subsequent recovery. Do not disconnect a real plant as a demonstration.

Pump-site WAN lost

Let the remote process follow its engineered fallback

Design expectation: the RTU follows approved loss-of-communications logic; central SCADA flags stale telemetry. A local operating procedure addresses conditions that exceed that fallback.

Prove it in simulation: verify timeout, stale flags, safe local control, operator response and reconnection. Evaluate queued commands; do not assume automatic replay is safe.

Vendor access revoked

End support access without ending operations

Design expectation: revoking a vendor session stops remote maintenance access while local operators retain monitoring and control.

Prove it on a test workstation: expire the approved window, attempt reconnection and measure revocation delay. Confirm no persistent tunnel, unmanaged account or scheduled tool remains.

04 / Prove both sides

Blocked where required. Working where needed.

Before any live test: obtain operations and controls-engineering approval, define the safe process envelope and abort thresholds, verify backups, identify a rollback owner and establish an observation baseline. Start with configuration review, passive traffic observation and an isolated lab. Do not run exploit tests, broad scans, packet floods, live logic changes or unplanned disconnects against production.

Each test needs a security result and a positive operational result. Establish numerical targets for alarm latency, data freshness, packet loss, recovery time, access revocation and process limits before execution. Unspecified targets mean Not ready, not Pass. Emergency access must remain auditable and must not bypass safety controls.

T01NOT RUN

Office-to-controller isolation

Lab, then approved boundary check

From a designated IT test host, attempt one approved connection to a safe test endpoint representing the controller zone; inspect rule counters. Do not scan a production PLC.

Pass + continuity
Denied with matching firewall evidence; ordinary operator monitoring stays available.
Stop / rollback trigger
Unexpected reachable path, loss of alarms, or new controller communication faults.
Evidence to retain
Timestamped rule ID, deny log, endpoint result, operator trend capture.
T02NOT RUN

Reporting without control access

Lab → observation

Show a changing synthetic process value on the DMZ replica and dashboard; attempt a forbidden reverse session to the isolated test endpoint.

Pass + continuity
Report freshness meets the owner-approved threshold; reverse session denied; plant loop unaffected.
Stop / rollback trigger
Reporting opens a route into control or replication overloads the historian.
Evidence to retain
Source/replica timestamps, deny log, historian load baseline.
T03NOT RUN

Vendor access expires

Lab → approved maintenance session

Try an unapproved user, then an approved named user to a test workstation; expire and revoke the session.

Pass + continuity
Unapproved entry denied; authorized task works; revocation meets agreed time; session is attributable.
Stop / rollback trigger
Access exceeds destination scope, cannot be terminated, or local operator loses control.
Evidence to retain
Approval, MFA outcome, session log, expiry and termination timestamps.
T04NOT RUN

Engineering access is scoped

Lab / spare controller only for writes

Use an approved engineering station to back up and perform a harmless lab change; confirm another workstation is blocked.

Pass + continuity
Approved engineering works only for assigned equipment; known-good logic can be restored in the lab.
Stop / rollback trigger
Unexpected logic change, access to other cells, backup inconsistency.
Evidence to retain
Change ticket, logic checksum, configuration diff, restore evidence.
T05NOT RUN

Cell containment with process continuity

Lab; production change window only after signoff

Use synthetic endpoints across cells to verify deny rules; observe actual approved SCADA traffic passively.

Pass + continuity
Cross-cell test denied; required SCADA values and alarms remain within signed-off latency/loss limits.
Stop / rollback trigger
Stale safety-relevant data, alarm loss, unintended process change.
Evidence to retain
Allowed/denied logs, baseline comparison, operator signoff.
T06NOT RUN

Remote-site loss of communications

Simulator first; controlled exercise only

Simulate WAN loss and recovery; verify RTU loss-of-link logic, alarm delivery/stale indication and reconnection.

Pass + continuity
Site follows its approved safe operating envelope; stale data is clear; recovery does not replay unsafe commands.
Stop / rollback trigger
Process leaves approved envelope, alarms disappear, or uncontrolled restart.
Evidence to retain
Simulator traces, configured timeout, alarm timeline, restart review.
T07NOT RUN

Controlled update transfer

Lab / staging only

Transfer an approved signed package and a harmless hash-mismatched sample; do not install updates on live controls.

Pass + continuity
Approved file arrives intact; altered sample rejected; no automatic production install.
Stop / rollback trigger
Unapproved file accepted or transfer path bypasses staging.
Evidence to retain
Hashes/signatures, approvals, rejection log, compatibility result.
T08NOT RUN

Loss of enterprise dependencies

Lab / tabletop, then approved resilience window

Model corporate identity, WAN and cloud unavailability; exercise local OT sign-in and the documented emergency-access process.

Pass + continuity
Essential local operation and accountability continue for the agreed outage duration; recovery reconciles access logs.
Stop / rollback trigger
Operators locked out, clock drift exceeds limit, or required control dependency fails.
Evidence to retain
Dependency map, sign-in outcomes, clock offsets, recovery timeline.
T09NOT RUN

Visibility without process overload

Passive observation

Generate one benign boundary-denied test and verify delivery to the monitoring team; compare controller/network load against baseline.

Pass + continuity
Alert arrives within agreed time with useful context; control performance remains within limits.
Stop / rollback trigger
Monitoring causes load spikes or blocks operational traffic.
Evidence to retain
Alert receipt, timestamps, resource/traffic trends.
T10NOT RUN

Restore what we protect

Isolated spare / simulation

Restore a selected PLC project, HMI configuration and firewall policy to nonproduction equipment; verify versions and expected behavior.

Pass + continuity
Restore meets approved recovery time and recovery-point targets, with credentials and dependencies available.
Stop / rollback trigger
Restore fails, backup is incomplete, or evidence cannot establish version integrity.
Evidence to retain
Backup age, hashes, elapsed restore time, functional checks.
T11NOT RUN

Protective-function independence

Safety specialist-led review / simulator

Review safety interfaces and maintenance access; simulate loss of supervisory communications without manipulating a live protective function.

Pass + continuity
Documented protection remains effective according to the process hazard analysis; unauthorized access is blocked.
Stop / rollback trigger
Any unexpected safety dependency or impaired protective function.
Evidence to retain
Safety review, interface inventory, simulation results; separate validation required.

05 / Change without losing control

Four gates before production acceptance.

Observe

Inventory equipment and every access path. Capture passive traffic over representative operating and maintenance cycles. Confirm the process owner’s limits and dependencies.

Rehearse

Test rules, remote access, loss of services and restoration in an isolated environment. Resolve every TBD and preserve a known-good configuration.

Pilot

Use an approved window and a limited boundary. Operator stays in control. Observe alarms and process trends; stop and restore the approved configuration on any abort trigger.

Accept

Review paired security and continuity evidence. Obtain operations and engineering signoff. Recheck after material changes and on an owner-approved schedule.

Evidence packet

Keep UTC and local timestamps, tester and witness, approved scope, asset/configuration versions, baseline and post-change trends, firewall decisions, alert/session records, rollback outcome and unresolved exceptions. Store sensitive plant detail in the utility’s controlled repository, not a public portfolio.

What remains to be engineered

Teaching boundary: the separate pump demonstration illustrates failure and recovery concepts. It does not validate this architecture or any plant acceptance test. Site-specific engineering, hazard analysis and independent acceptance evidence remain necessary.

Decision register / Source-backed review

Why these architecture choices?

These government publications are stronger foundations than a vendor brochure. They support principles, not certification of this design. Exact topology, protocols, equipment and acceptance thresholds remain engineering decisions.

DecisionPrimary publicationHow we apply itWhat it does not establish
Industrial DMZ and process zonesNIST SP 800-82 Rev. 3 · Final, September 2023. Full publication PDF. See network segmentation discussion and Figure 16.Separate enterprise, exchange services and process control. Our F01–F07 paths make permitted communication explicit.Does not prescribe our exact number of zones or firewall rules. Revision 4 is an initial public draft as of this review, not the final baseline.
Brokered maintenance accessNIST SP 1800-45 — Water and Wastewater Build Architecture (OT Remote Access) · Final, June 24, 2026. Full publication PDF.The publication demonstrates three representative remote-access implementations for utilities of different capacities. Use it to evaluate F03/F04 against actual staffing and support needs; test authentication, scope and expiry in T03.We have not reproduced its vendor configurations or validated a specific product. A demonstrated example is not a complete security assessment of our utility design.
Contain compromise inside OTNCSC secure connectivity, Principle 6 · Official guidance. Full principles PDF.Keep treatment cells and engineering access separately controlled; test the lateral restrictions in T04/T05. Protect gateways as significant shared dependencies.Segmentation cannot make an already-authorized malicious control command safe. Application permissions and process protection remain necessary.
Isolation without blind shutdownNCSC secure connectivity, Principle 8 — Establish an isolation plan · Official guidance.Prepare and rehearse service-specific isolation and continued local operation. T06/T08 test dependencies and recovery before any live exercise.No universal safe pump state, outage duration or automatic permission to disconnect telemetry. Utility operators and responsible engineers approve these choices.

Verified 29 October 2026. These links lead to the publishers; no third-party paper has been rehosted. The reference topology, flow matrix and acceptance plans are our original interpretation and remain unvalidated.

Sources & interpretation

Grounded in guidance. Adapted through engineering.

This original reference design uses segmentation and controlled remote access principles. The topology, flow IDs, acceptance tests and scenario expectations are proposed engineering choices for this example, not requirements quoted from a standard.

Sources checked 29 October 2026. No assertion of certification, compliance, a complete threat model or security of any actual facility.