Decision Lineage & Provenance
The Inquiry: Organizational decisions are the unit of governance accountability, yet no infrastructure exists to capture the provenance of decisions — why they were made, by what authority, under what constraints, and with what expected outcomes — as a structural by-product of organizational operation. What is the state of the art across provenance, design rationale, institutional memory, process mining, AI accountability, and audit compliance? Where does each tradition's coverage end? Is the gap between these traditions disciplinary (solvable within existing fields) or structural (requiring new infrastructure)?
Falsifiable formulation: If any existing standard, framework, tool, or research tradition provides infrastructure that captures organizational decision provenance — the full chain of why, by whose authority, under what constraints, with what alternatives considered, and with what expected outcomes — at the moment of action and as a by-product of operation rather than a separate documentation task, then the structural gap claimed here does not exist.
The data-to-decision gap is architectural, not extensional.
The W3C PROV specification defines provenance for "a piece of data or a thing." The provenance taxonomy (Buneman's why/where, Cheney's how, Herschel's what-for/what-form/what-from) provides a sophisticated vocabulary — but entirely for data. The questions organizational governance asks operate at a different abstraction level. Data provenance answers: which data influenced this output? Decision provenance must answer: why was this decision made, by whose authority, under what constraints, considering what alternatives, and expecting what outcomes? PROV's Entity/Activity/Agent model does not contain the constructs for authority delegation, constraint context, committed intent, or epistemic classification of knowledge states. The gap is not a missing PROV extension — it is a different layer of the stack.
Singh, Cobbe & Norval (2019) named this gap as "decision provenance" and proposed using data provenance methods to expose "decision pipelines." Their insight was correct — the accountability problem lives in decision chains, not just data chains. But their proposal remains within the data-flow paradigm: trace the data that flows into and out of decisions. The deeper requirement is infrastructure that captures the governance context of the decision itself — the authority, the constraints, the rationale — at the moment of commitment.
The cognitive load diagnosis resolves the design rationale adoption puzzle.
Fifty years of design rationale research (Rittel 1973 → Ahmeti et al. 2024) produced increasingly lightweight documentation approaches without solving the adoption problem. The pattern: IBIS required structured argumentation networks. gIBIS made them graphical. QOC simplified the notation. ADRs reduced it to a template. Lightweight ADRs cut it to a single-page file. Each iteration reduced the burden. None eliminated it.
The cognitive load framework explains why: every approach in this tradition requires extraneous cognitive load — stepping outside the work to document the work. The designer must shift attention from the design decision to the meta-task of recording the design decision. This is not a tooling problem. It is a structural property of the documentation paradigm: when rationale capture is separate from rationale use, it imposes load that practitioners rationally reject. Nonaka & Takeuchi's (1995) SECI model adds depth: design decisions are substantially tacit knowledge — acquired through experience and not fully articulable. Attempting to externalize them after the fact is inherently lossy. The resolution requires making governance context a structural requirement of the operation itself — capturing rationale at the moment of commitment through the structural requirements of governance primitives, not through a separate documentation step.
Institutional memory literature confirms: memory loss is structural, preservation must be structural.
Walsh & Ungson's (1991) five retention facilities show that organizational memory is a distributed structural property, not a database. When any facility degrades, memory degrades. Darr, Argote & Epple (1995) provided the empirical anchor: knowledge depreciation is rapid even in standardized service operations (pizza franchises). Benkard (2000) confirmed it in aircraft manufacturing. These findings demolish the naive assumption that storing information prevents memory loss. Pollitt's (2000) four amnesia types — not documented, records lost, archives inaccessible, records available but unused — describe failure modes of storage technology, not of memory infrastructure.
The synthesis across Wegner (1987), Ren & Argote (2011), Ackerman & Halverson (2000), and Fiedler & Welpe (2010) is that memory preservation is a function of organizational structure, not storage capacity. Governance infrastructure that makes decision context a structural by-product of operation addresses the root cause identified by this tradition.
The retrospective/prospective distinction is architecturally fundamental.
Process mining (van der Aalst) and data provenance (Buneman/Cheney/Herschel) operate retrospectively — analyzing event logs and data transformation records after the fact. Audit standards (21 CFR Part 11, SOX 404) mandate retrospective records. Design rationale approaches attempt prospective capture but as a separate documentation burden. Kroll (2021) identifies the infrastructure gap explicitly — practitioners lack the tools to meet traceability requirements.
The synthesis reveals that prospective capture — capturing decision context at the moment of action, when reasoning is still active in working memory — is architecturally necessary. And it must be intrinsic to the operation. The xAPI standard (F11) validates that statement-based capture works at scale, but xAPI statements are epistemically flat: any actor can assert anything, no distinction between authoritative declarations and derived inferences, no authority delegation model. The gap between xAPI's deployed capability and governance requirements defines the infrastructure space.
Organizations make decisions constantly — yet no infrastructure captures why those decisions were made, by whose authority, under what constraints, or with what expected outcomes. This survey examines six independent research traditions spanning five decades: data provenance, design rationale, institutional memory, process mining, AI accountability, and audit compliance. Each tradition independently establishes the requirement for organizational decision provenance and documents the consequences of its absence, but none provides infrastructure that captures decision context as a structural by-product of operation. The W3C PROV standard provides mature data provenance; nothing equivalent exists for decisions. The 50-year design rationale adoption failure is structurally diagnosed through cognitive load theory: every approach requires stepping outside the work to document the work. Institutional memory research confirms that preservation requires structural infrastructure, not better storage. The convergence of six traditions on the same gap — at their intersection, not within any one of them — constitutes the strongest evidence that the gap is architectural, not disciplinary.
"A record that describes the people, institutions, entities, and activities involved in producing, influencing, or delivering a piece of data or a thing." — Moreau & Missier (2013), W3C PROV-DM Recommendation