The Analytical Move
The raw material — Senedd proceedings, Audit Wales reports, DHCW board papers, Employment Tribunal filings, witness testimony, FOI refusals, the record at carenhs.org — becomes eleven loops and seven traps through a single analytical move: group every observation by the stock it affects and the feedback it produces.
Take a concrete example. A Senedd committee hears that health boards are building their own systems rather than using DHCW’s. That observation affects a stock (health-board trust in DHCW — declining). It also affects a flow (adoption rate — falling). The declining trust stock feeds back into DHCW’s credibility, which further reduces adoption. When you trace this chain far enough, it closes into a circle: a reinforcing feedback loop. That loop is L2 — Credibility Death Spiral.
Loops emerge when observations form a closed cycle. Traps emerge when multiple loops interact in Meadows’ recognised patterns. This is not interpretation — it is classification. The discipline provides the taxonomy; the evidence provides the instances.
The Evidence Aggregation
The discipline gives the taxonomy. What turns the taxonomy into a diagnosis is the evidence aggregation. Five years of DHCW board and committee meeting transcripts — 61 meetings in total — were processed into a knowledge graph of 1,779 nodes and 3,427 edges, supplemented by Welsh Government performance-escalation correspondence, Audit Wales structured assessments, FOI disclosures, and the ABUHB dossier (123 sources). Each node is an entity, a claim, a finding, or an evidence point. Each edge is a typed relationship between them.
Some edge types carry quantitative weight:
- 51 edges of type “approved without scrutiny” — board approvals that the published record shows received no substantive discussion before being passed.
- 45 edges of type “undeclared interest at meeting” — instances where a director participated in a discussion in which they had an undisclosed interest.
- 33 edges of type “declared nil despite” — instances where a director declared “nil” while holding a known undisclosed Professor of Practice title (Thomas 10, Evans 10, Hurle 8, Hall 5).
- 63 edges of type “failed to act on” — instances where a flagged risk was logged and no corrective action followed.
- 107 sanitisation findings plus 237 hiding-intent passages — passages where what was spoken in the room was substantively altered, softened, or deleted before publication.
These counts are the bridge between the discipline (which predicts that captured governance produces certain edge-types) and the evidence (which shows the predicted edge-types appearing at scale). They are not the only evidence — much of the diagnosis rests on qualitative passages, named quotes, and structural observations — but they are the part that quantifies what the qualitative narrative describes.
Two Clusters, One Architecture
The eleven loops fall into two clusters with fundamentally different characters.
Cluster A — five reinforcing loops of delivery failure. Hiring without delivery, credibility death spiral, annual-funding paralysis, the rebranding escape, vendor-dependency spiral. These are structural. Any national health IT body operating under similar constraints — monopoly mandate, annual funding cycle, sovereign health-board customers — would face some version of these dynamics. They are the starting point of the diagnosis, not the whole story.
Cluster B — six reinforcing loops of self-preservation. Manufactured narrative, competence void, loyalty selection, whistleblower suppression, information fortress, oversight obstruction. Cluster B is not structural accident. It is an active engine that intercepts every corrective mechanism — board scrutiny, ministerial oversight, staff feedback, external audit — before it can reach Cluster A. This is why repeated reform attempts have failed: the self-preservation engine absorbs them.
The blueprint must address both clusters. Fixing delivery structure without removing the self-preservation engine means the engine will capture the new structure. Removing leadership without fixing the structure means new leadership will face the same dynamics.
How to Read the Diagnosis
Each loop page has four sections: the pattern (the generic systems-dynamics mechanism), how it manifests at DHCW (specific evidence), the healthy alternative (what a functioning version looks like), and the escape route (which intervention in the blueprint breaks the loop). Every loop identifies a specific stock being depleted, the feedback mechanism that depletes it, and the characteristic delay that makes the damage invisible until it is severe.
Each trap page has four sections: the archetype (Meadows’ definition), the DHCW manifestation, which loops produce it, and the intervention that breaks it. Traps are how the blueprint sequences its interventions — the trap structure dictates which changes must come first.
The 18 stocks tracked across the analysis are organised by visibility on the stocks page: eight that are visible and measured, six that are visible but unmeasured, and four that are invisible but influential. The gap between what governance measures and what actually determines outcomes is itself a diagnostic finding.
What the Methodology Predicted
Systems-dynamics analyses make a strong claim: when the structural conditions are present at the start, the failure patterns will be visible from the start. This is testable. The deep evidence base tests it.
Every governance deficit pattern observed at the point of Level 3 escalation thirty-four months in was already operational at DHCW’s very first board meeting. Pre-DHCW UWTSD Professor of Practice titles awarded to three executive directors in December 2020 — four months before the founding board — were never declared at any subsequent board meeting through to escalation. The first meeting in April 2021 documents a director declaring “nil” while ten months into holding that title. The architecture was not gradually corrupted by stress. It was complete on day one.
This is the methodological pay-off. The discipline predicts that systems shaped by certain feedback loops and protected by certain self-preservation mechanisms will fail in characteristic ways. The evidence shows the predicted failure unfolding exactly as the discipline forecasts — and visible from inception, not in hindsight.