ClusterOS has diagnosed 377 innovation clusters across 57 countries. The 26 ecosystems figure refers to the UK National Behavioural Diagnostic, where every Combined Authority, devolved nation, and city region is read together as one system. What the framework finds is not what each ecosystem does differently — it is how consistently they do the same things.
The diagnostic tracks nine stalls — behavioural substitutions a cluster makes when it cannot decide. Aggregated across 232 cluster diagnostics in the corpus's current-schema pool, every stall is observable; the distribution is not uniform.
The corpus contains 377 cluster diagnostics in total; figures in this section are computed against the 232 with stack readings under the current diagnostic schema. See methodology footnote in Section 4.
Source — diagnostic library, canonical-schema pool. Bars sum to 100% of the pool per stall. Stabilising Around Incumbents fires at high or medium in 90% of clusters; Forgiving Instead of Redesigning is indeterminate in 86%. The framework refuses to assert what it cannot observe.
The most comprehensive diagnostic reading we have produced. Every UK Combined Authority, devolved nation, and city region — diagnosed through the same framework, cross-referenced at entity level.
Read the UK diagnostic →Eight clusters in one Scottish supercluster — Advanced Manufacturing, Critical Technologies, Digital Creative, FinTech, Innovation Ecosystem, Life Sciences, Net Zero, Space — read together to surface ecosystem-level dynamics.
Read the Glasgow diagnostic →Our first international supercluster diagnostic. Space, tourism, defence, MedTech — ten clusters diagnosed with steward-held data from the Orlando Economic Partnership.
Read the Orlando diagnostic →Every cluster the diagnostic has read, browsable. Each profile carries its own stall map, configuration breakdown, and leverage hypotheses, cross-referenced to the corpus-wide patterns above.
Browse the cluster library →Where stalls describe individual behavioural substitutions, configurations describe how those substitutions stack — the structural shapes a cluster falls into when several stalls fire together. Six configurations dominate the canonical-schema pool.
Source — diagnostic library, canonical-schema pool. The pool is composed of 173 UK NBD clusters re-stacked under canonical schema and 59 international clusters carrying canonical-schema outputs; share-of-pool figures should be read as a UK-weighted reading of the corpus, not a global one.
Cross-cut the canonical pool by sector and configuration and three readings sharpen: capital-intensive sectors converge on a three-stack backbone, capital-light sectors do not, and the smallest, most-anchored sector reads with zero variance across its members.
Coverage — 212 of 232 canonical-pool clusters classified by sector. 20 clusters in long-tail sectors (logistics, food and drink, tourism, aquaculture, mining tech) are unclassified for this grid.
Coordination–Intermediary–Activity, Extraction–Intermediary, and Governance Capture all clear 75% confident in Energy and Net Zero, Advanced Manufacturing, and Quantum and Photonics. Where the underlying work needs capital, physical anchors, and long timelines, the corpus shows the same three-stack signature regardless of geography.
FinTech sits below the cross-sector mean on every backbone configuration: Coordination–Intermediary–Activity 48.3 vs 72.8, Extraction–Intermediary 51.7 vs 72.8, Governance Capture 44.8 vs 70.7, Permission–Validation 41.4 vs 54.5. The highest reading is 51.7%. Capital-light sectors do not converge to a backbone — the configuration profile is flatter, not tighter.
All five Quantum and Photonics clusters fire on five of six configurations at 100% confident; only Volume–Tolerance misses. Zero variance across five clusters in five configurations means the diagnostic is reading sector-level structure rather than cluster-level variation — the most extreme expression of the backbone effect.
Five cross-continental resemblance pairs from the corpus. Each pair belongs to a multi-cluster resemblance group at cosine similarity 0.97 or higher — measured on the confidence-weighted stall and stack signature.
Method — pairwise cosine on a confidence-weighted signature covering each cluster's nine-stall confidence map and any STK-XX stacks present, threshold 0.97 for group inclusion. Pool: 391 diagnostic runs with parseable stall fingerprint. 26 multi-cluster groups; 21 span more than one country; 21 span more than one continent.
Diagnostic reports surface leverage hypotheses — small, reversible interventions that change the structure registering in the diagnostic. Across the 170 UK NBD clusters with typed leverage hypotheses, the corpus generates 2,177 hypotheses that classify cleanly into five categories.
Source — UK National Behavioural Diagnostic, leverage stage. 2,177 hypotheses classified cleanly into the five categories shown; a long tail of OTHER classifications was excluded from this chart pending classifier refinement.
Information flow leads — close to half of the typed hypotheses surface a previously-unreported metric, a feedback loop that does not return, or a flow that is invisible to the actors who would respond to it. Boundary adjustment and timing or sequencing follow at similar weight; both change who or when, not what. Constraint shift and coupling exposure carry the smaller share — leverage that changes a binding rule, or that routes a transaction directly between two actors who currently use an intermediary.
The skew toward information flow is consistent with what the diagnostic finds elsewhere in this page — Coordinating Instead of Deciding fires in 69% of canonical-pool clusters, Stabilising Around Incumbents in 90%, and the Coordination–Intermediary–Activity Configuration in 71%. Where coordination substitutes for decision and intermediation substitutes for connection, the most-available leverage is restoring the metric that lets actors see what the coordination is doing.
15 minutes. No data required. Identify stall patterns, see the configurations they form, and read the leverage entry points the framework would surface.
Start the self-diagnostic →Your evidence, your actors, your clusters. Cross-referenced at entity level. Output: ecosystem-level diagnostic with leverage hypotheses calibrated to your context.
Request a diagnostic →