Coherence
Parts of the system tell the same story — software, organisation, and record align
What is committed in one place is what every consumer reads, decides on, and is held accountable for — without silent divergence between layers, now or over time.
Coherence is the aim that software projections, organisational records, and human understanding describe the same committed facts — not three reconcilable approximations of what might have happened.
The aim
An organisation pursuing coherence accepts that alignment is architectural, not aspirational — and that alignment must be maintained over time, not declared once at kickoff. Teams do not each maintain a partial truth and hope meetings produce consensus. A single writer publishes L0 outcomes; consumers read those outcomes; overrides and inferences are labelled until committed at a boundary. Without that infrastructure, values drift is the default: documents stay noble while practice erodes.
Coherence is not uniformity of opinion. Deliberation remains messy. Coherence begins at commitment: once authority decides, the record is one, traceable, and consumed uniformly.
In software systems
Coherent systems have one authoritative stream per invariant domain, projections rebuilt from committed events, and no operational read path that silently diverges from the log. Dashboards, APIs, and audit exports answer the same question with the same numbers because they derive from the same L0 facts.
In human organisations
Coherent organisations publish committed decisions as organisational fact. Parallel spreadsheets, shadow CRMs, and “what we actually do” cultures are treated as incoherence to be resolved — not as harmless local adaptation.
In socio-technical systems
Coherence holds when Conway’s Law is applied deliberately: domain ownership in the org chart matches single-writer boundaries in architecture, and neither layer treats the other’s shortcuts as the real record.
What threatens this aim
Failure modes: Silent override and inference as fact break coherence by letting operational truth diverge from committed truth. Impossible simultaneity breaks it by presenting aggregates that never existed at one moment. Values drift breaks it gradually — stated values and principles diverge from lived practice while the posters and principle decks still look fine.
Anti-patterns: Diffuse write authority and epistemic tier collapse structurally guarantee drift — multiple truths or mislabelled epistemic status.
Dark patterns: Segmented messaging delivers incompatible substance to different parties — coherence fails by design, not by accident.
Principled support: Single writer principle, commitment boundary, and correction events are the primary patterns that advance coherence — including coherence over time — without sacrificing speed or autonomy.