04 · Communities
WHAT TO CHECK
Define nodes, direction and weight of relations, period, sources and deduplication rule. A contract, citation, investment and flow are not interchangeable relations, and missing links caused by incomplete collection remain visible.
Community detection groups nodes under a declared function such as modularity. Betweenness measures a node's presence on shortest paths under a stated formula and normalization. Neither result names a motive by itself.
Recalculate after removing uncertain links, changing the window or varying resolution. Interpret a group only when it remains sufficiently stable and external evidence explains the relationships.
VALIDATION DOSSIER
Use the question “Which groups of actors are more densely connected to one another than to the rest of the ecosystem?” as a contract: name the owner, deadline, scope and alternatives before collecting more material. Communities can reveal coalitions, standards, schools of thought or influence chains that a list cannot show. Record what is deliberately outside scope, who can challenge the framing and what consequence follows from postponement. This prevents an attractive framework, actor list or market story from silently replacing the decision that the analysis is meant to support.
VALIDATION DOSSIER
For every material statement, retain the original source, publication date, observation period, method, unit and relevant population. Distinguish observed facts, calculated estimates, interview judgements and assumptions. Connect the register to the promised output: Interpretable groups, their bridge actors and their changes over time. Missing or contradictory evidence remains visible with a responsible owner and a collection action; it is never converted into certainty merely to complete the page.
VALIDATION DOSSIER
Define the rules before applying the analysis: inclusion criteria, identifiers, comparison set, calculation or coding convention, treatment of missing values and version date. Reproduce the same convention across every option or actor. Use the page checks as acceptance tests, beginning with “Specify the relationship type and graph period.” A conclusion is usable only when another practitioner can reconstruct how the inputs became the displayed result and identify where judgement entered.
VALIDATION DOSSIER
Challenge the preferred interpretation with at least one plausible alternative, contradictory observation and sensitivity test. Separate association from prediction and never infer intent, power or causality from a position, score or visual proximity alone. The standing limit remains: A statistical community does not automatically imply an alliance or shared intention. State which uncertainty is most likely to reverse the conclusion and what additional observation would discriminate between the competing explanations.
VALIDATION DOSSIER
Turn the analysis into a dated decision record rather than a static page. Preserve the selected option, rejected alternatives, objections, accepted risks and first reversible action. Name the signal, numerical or qualitative threshold, review owner and review date that will confirm, adapt, pause or reopen the decision. Link every later update to its author and reason so that the history remains understandable when evidence or market conditions change.
LIMIT
