A policy brief reaches a minister's desk. Two of its four key claims trace back to a single working paper — not three independent sources. The third claim was contested in a 2021 OECD report nobody cross-checked. The brief sounds authoritative. Nobody in the chain had the infrastructure to know it wasn't.
This is not an analyst failure. It is a structural failure of how evidence-based policy work is done.
Every major policy review begins by reconstructing evidence that was gathered for the last one. Six months of institutional knowledge evaporates when the project ends. The analyst who built the previous brief has moved on. The literature they curated is in a folder nobody can find.
A single working paper gets cited by 12 subsequent papers — all of which cite each other. Standard tools see 13 sources and report high consensus. Epistamate's diversity weighting detects the cascade. Claims that trace to one original source are flagged, not amplified.
Policy briefs report what is known. They are professionally reluctant to name what is not known — because gaps read as incompleteness. But the gaps are often the most consequential part of the brief. Epistamate makes gaps first-class outputs, with importance ratings.
A parliamentary question lands. A committee wants a brief in 48 hours. Existing knowledge cannot be rapidly queried for what has already been established. Everything restarts. The second time this topic comes around, the same work is done again — at crisis speed.
When a policy recommendation is scrutinised months later, the original evidence basis is often unrecoverable. What was known, what was contested, what gaps were acknowledged at the time of decision — none of it is in the brief. It was in the analyst's working notes, which are gone.
Session N+1 retrieves verified claims from session N. A policy analyst running a brief on carbon pricing in Q4 has access to the verified findings from the Q2 session — without rebuilding them. Contradictions between sessions are preserved as typed objects, not silently resolved. The knowledge base grows with each session.
The confidence formula's diversity weighting scores cross-provider consensus differently from single-source amplification. A claim corroborated by three independent Tier 1 sources scores higher than a claim cited by 15 papers all tracing to one original. This distinction is structurally invisible to any tool that counts citations rather than sources.
Every research session produces a typed gap list alongside the claim vault. Gaps have importance ratings. They accumulate across sessions and narrow as evidence arrives. A briefing that says "we do not yet have reliable evidence on X — importance: HIGH" is more trustworthy than one that elides the uncertainty.
When the parliamentary question arrives, the quick scan retrieves what is already verified. Phase 2 deepens on what has changed or is new. The adversarial challenge phase runs regardless — but it runs faster when the base is already established. The brief ships with confidence scores visible to the reader.
When a decision is logged, the full evidence state is preserved at that moment — verified claims, contested claims, acknowledged gaps, source tiers, confidence scores. This is not a compliance feature added afterward. It is the output of how the engine works by default.
UN agency situation reports, OCHA Humanitarian Needs Overviews, UNDP policy assessments — evidence synthesis under time pressure, across conflicting organisational sources, with accountability to funding bodies. Gap tracking and contradiction detection are exactly what's missing from these workflows.
Impact assessments, comitology documents, parliamentary questions, and committee briefs all require evidence that is traceable, contestable, and auditable. The engine's Article 12 compatible Decision Log is not a bolt-on — it is the structural output of the pipeline.
Library of Parliament units, congressional research services, and independent legislative advisory bodies do structured evidence work under deadline pressure with no institutional memory between legislative cycles. The knowledge graph is precisely what's missing.
We are talking to policy organisations, think tanks, and legislative research bodies in active development. If you recognise your workflow in this page, we would like to understand the specific problem before describing the solution.