Framework · what it is

Every setting traces to the business outcome it serves, and every outcome to the measure that proves it.

Value is the frame

One chain, stored as links a machine can walk.

  1. 1 · Outcome What the business gets, in its own words, with an owner, a metric, a baseline and a dated target. “A new starter has a working port the same day they ask for it.”
  2. 2 · Requirement What the business asked the network to do, and the technical requirements derived from it.
  3. 3 · Decision What was chosen, what was turned down and why, who decided, and the assumptions the choice depends on.
  4. 4 · Implementation The part of the data model and the configuration the decision explains.
  5. 5 · Evidence The checks and telemetry that prove the chain holds today, kept with a date.
The evidence flows back up: the measurement is how the outcome reports where it stands.
A break anywhere is a finding. UnexplainedA setting with no decision. UnjustifiedA decision with no requirement. UnreportableAn outcome nothing measures.
VALUE · THE FRAME for which outcome? Observe Model Codify Automate Augment Govern
The six disciplines

One loop, not six projects.

  1. ObserveWhat is actually happening? Telemetry and checks produce the evidence and the outcome measures.
  2. ModelWhat should the network be? One source of truth for intent.
  3. CodifyWhat does the organisation know? Decisions, their reasons and their assumptions, in structured records.
  4. AutomateWhat can the system do? Repeatable decisions become deterministic changes.
  5. AugmentWhere can AI amplify judgement? AI reads and writes the same records as the engineers.
  6. GovernHow do we know it is safe? Every change is traced, checked and accepted by a named person.
What makes it different

Assumptions that announce when they stop holding.

Most decision records are written once and never revisited. In CIK, each decision lists what it assumes in a form a machine can check. When an assumption stops holding, the decision is reopened instead of waiting for someone to remember it.

“One collector is enough.” Review due Apr 2027
“No legitimate endpoint exceeds the storm-control limit.” Breached
In the 03:00 scenario: decision ADR-0013 reopened for review
The seven value categories

What a leader is asked for, and how it is measured.

CostDo more with the same spendRun cost per site, circuits sized to measured use
ResilienceFewer, shorter incidentsChange failure rate, time to restore
SpeedSay yes to the business fasterLead time for a new site, desk or segment
ScaleGrowth without headcount growing at the same rateChanges per engineer
ControlPass the audit without a fire drillHours to produce evidence, open findings
ContinuityThe person leaving is not a crisisShare of the estate with recorded reasoning
AI readinessUse AI safelyShare of questions an agent answers correctly from the records

Money is shown where a conversion can be defended, and every conversion is labelled as an assumption with an owner. Build your value case

One chain, filled in

From the published campus design, with real records.

The chain also shows what it cannot prove yet. The second segmentation requirement, restriction within a segment, fails its check: the policy is decided (ADR-0022) but not yet modelled.

See the evidence
  • OUT-CTL-SEG.yaml, requirements.yaml, ADR-0009.md
  • The camera VRF stanza, and checks.json dated 2026-10-05
  1. OutcomeOUT-CTL-SEG Segment isolation can be shown to an auditor from current records, on the day. Owner: the business. Measured 0 of 2 segmentation requirements proven as published, 1 of 2 now; target 2 of 2 by 30 April 2027 (illustrative target).
  2. RequirementREQ-BR-02 → REQ-TR-03 Segment voice, data, wireless, camera and IPTV so one domain cannot degrade or expose another → macro-segmentation by VRF.
  3. DecisionADR-0009 One segment and one VRF per device class. Rejected: VLANs in one routing table, separated by access lists. Decided by elo33011 and ppklau.
  4. Implementationlogical topology model, rendered config vrf definition VRF-CAMPUS-CAMERAS With its own route target.
  5. EvidenceCHK-SEG-01, CHK-SEG-03Passed 5 October 2026 6 of 6 segments have a VRF; 6 distinct VRFs for 6 segments; 0 pairs share a route target.
How the pieces fit

CIK is not a product. It is a set of records and the links between them.

Kept in the firm's own repository, with the tools the firm already runs on either side. Each piece is shown by the job it does; the product named under it is the one used in this work, and any product that does the same job can take its place.

WHERE WORK ARRIVES CIK · THE FIRM'S OWN RECORDS, IN GIT AI, ON THE SAME RECORDS INTENT AND AUTOMATION OBSERVABILITY becomes a requirement breaks an assumption agents draft proposals · a named person accepts decisions explain model sections check results telemetry measures outcomes, tests assumptions Requests, changes Jira · ServiceNow Incidents ServiceNow · PagerDuty Records outcomes · requirements · decisions assumptions · standards · designs Lineage and validator walks the links, reports every break Evidence checks and measures, dated MCP server search · why · impact what-if, read-only Agents any model answer and propose Source of truth Infrahub · here YAML Render Jinja2 templates Pipeline GitHub Actions Automation Ansible · Kriten The network switches · routers · WAN Collectors gNMI · model-driven telemetry Time-series store Prometheus · InfluxDB Dashboards · Grafana
Built in the campus proof Partly: modelled, not running on devices Target state: not built Proposals, accepted by a named person The pipeline passes locally and has not yet run on the hosting platform.
What each piece does, and what exists today
PieceIts jobUsed hereIn the campus proof
Requests and changes Where the business asks for something. A request becomes a requirement, or a proposal against an existing one. Jira, ServiceNow Target stateNot connected.
Incidents and outages Where things go wrong. An incident raises a drift item or marks an assumption as broken, which reopens the decision. ServiceNow, PagerDuty Target stateNot connected; the 03:00 scenario was played by hand.
CIK records Outcomes, requirements, decisions, assumptions, standards and designs, as files a person and a machine can both read. Git, the CIK schema Built8 outcomes, 10 requirements, 26 decisions.
Lineage and validator Walks the links and reports every break: a setting with no reason, an outcome nothing measures. CIK tools Built
Evidence Checks against the model and rendered configuration, and outcome measures, each with a date. CIK checks Built25 checks, 21 passing.
Source of truth What the network should be: the data models the configuration is rendered from. Infrahub; NetBox or Nautobot can take its place. The campus uses YAML with JSON Schema. BuiltBuilt as YAML. The Infrahub schema extension is a draft.
Render and pipeline Turns the model into configuration, then validates, checks and gates every change. Jinja2, GitHub Actions BuiltThe pipeline passes locally; not yet run on the hosting platform.
Automation engines Push the change to devices, as a reviewed job. Ansible, Kriten PartlyPlaybooks exist in the design. Not run on devices.
Collectors and store Collect what the network is actually doing. gNMI and model-driven telemetry, Prometheus or InfluxDB PartlySubscriptions modelled, each sensor group linked to the outcome it serves. No collector running.
Dashboards and alerts Show the measures to the people who act on them. Grafana Target state
Tool access for AI Gives any model the same records an engineer uses, through a standard interface, read-only except for proposals. MCP BuiltRead-only: search, why, impact, and a what-if that re-runs every check on a copy. Writing proposals is target state.
Agents Answer “why is it like this?” and “what would this change touch?”, and draft proposals. A named person accepts every change. Any model BuiltUsed to build the proof. A demonstration agent answers questions about the campus; the with-and-without comparison is on the proof page.
The records stay in the firm's repository.No vendor holds the reasoning.
Every tool reads and writes through the same records,so the pipeline, the dashboards and the agent give the same answer.
A tool can be replaced without losing the knowledge.Changing the source of truth or the automation engine changes the links, not the decisions.
Not only network

The same links work for cloud or security.

  Network · built Cloud · illustration Security · illustration
OutcomeSame-day desk movesNew product environment in a dayAudit evidence on the day
RequirementOnboard moves without service riskSelf-service environments within guardrailsPrivileged access is reviewed
DecisionRouted access fabricAccount per product, shared landing zoneJust-in-time access, no standing admin
ImplementationData model, rendered configInfrastructure codeIdentity policy
EvidenceFabric checksPolicy-as-code resultsAccess review records

Structure only: the cloud and security columns are illustrations, not built.

What it looks like for each role.

Target state · illustrations
On-call engineer, 2 a.m.Observed state, design intent and the decision behind it, together on one screen.
Design engineerThe history of past decisions surfaced before a change is built, and the blast radius computed.
Senior leadershipThe same answer whether asked of an engineer or of the system, because it is the same source.
Business and financeConsumption and cost reconciled to departments, with a scenario planner for consolidation.
Start with one design. Get in touch