One public problem, allowed to become politically complicated.

A lived experience must be preserved before administrators classify, frame, and professionalise it into a policy problem.

What caused the service problem, who is affected, which frame is most useful, and which institution has standing or jurisdiction are not yet settled.

The original complaint and competing candidate frames become inspectable rather than being silently replaced by one administrative definition.

Epistemic competence can justify influence without independently creating a right to rule.

Which historical and institutional conception of good political order should guide the relationship between expertise, citizens, and office-holders?

The case records a competence-without-sovereignty rule instead of treating expertise as self-authorizing.

The best-informed actor is not necessarily the actor legally or democratically authorized to bind the public.

Delegated powers, review routes, principal-agent risks, and de facto influence may not align neatly with formal authority.

Jurisdiction and the human authorization point become explicit parts of the dossier.

Aggregate improvement can coexist with concentrated harm, unequal standing, and competing claims about liberty and equality.

Which distributions count as politically relevant and which remedies are required remain partly normative questions.

Stakeholder positions and concentrated effects are made visible before a preferred solution is selected.

A preferred method or discipline must not become the definition of the political object before alternatives are examined.

Different perspectives identify different units, mechanisms, evidence standards, and blind spots.

Rival ways of knowing are recorded without forcing them into one premature synthesis.

Capability, model agreement, technical quality, or procedural control cannot be laundered into political legitimacy.

Procedural power and informal influence may shape outcomes even where formal authority remains human.

AI-supported functions and prohibited authority substitutions are separated explicitly.

Research questions, measures, populations, comparison logic, assumptions, ethics, and uncertainty should be set before convenient data dictate the method.

No single method can answer every descriptive, causal, predictive, normative, or authority claim in the case.

Evidence needs and limitations are specified before results are interpreted.

Meaning, context, testimony, interpretation, and researcher judgment are evidence, but they require their own standards of provenance and challenge.

Sampling, access, interpretation, positionality, translation, contradictory cases, and saturation claims remain contestable.

Qualitative evidence is preserved for the claims it can support rather than treated as decorative anecdote.

A precise estimate may still answer the wrong measurement question if the concept, population, denominator, or operationalization is poorly chosen.

Missingness, sampling, subgroup validity, weighting, and measurement choices affect what the numbers can establish.

Quantitative claims are attached to explicit measurement rules and uncertainty.

Association, prediction, and before-and-after comparison do not automatically identify a causal effect.

Counterfactuals, confounding, identification assumptions, heterogeneity, and transportability limit causal confidence.

Causal claims must state the comparison and assumptions that make them credible.

Scale and automation can reproduce a category error faster; computational sophistication does not make categories or objectives neutral.

Model error, subgroup performance, reproducibility, corpus provenance, and version changes remain material to interpretation.

Computational results become inspectable evidence rather than an opaque authority claim.

Synthesis must preserve non-comparability, minority interpretations, conflicting findings, and uncertainty where convergence is not justified.

Some evidence supports common conclusions while other claims remain disputed or conditional.

The dossier records supported, disputed, and unresolved findings instead of manufacturing consensus.

Public input, representation, standing, information integrity, and political authorization are related but distinct.

Recruitment, access barriers, missing voices, misinformation, moderation, and representation claims limit what participation can establish.

Public voice enters the decision record without being mistaken for automatic representativeness or authorization.

The strongest epistemic recommendation and the authorized public decision may legitimately differ, but the difference must be visible.

Values, rights, political trade-offs, administrative feasibility, and residual empirical uncertainty remain at the decision point.

The authorized proposition, rationale, authority basis, dissent, advice, and any override are preserved as a distinct record.

Implementation and monitoring generate evidence for learning, not automatic policy reversal or proof of cause.

Observed failures or gains may reflect implementation, context, measurement, causal mechanisms, or the original policy design.

Reopening triggers, rollback feasibility, adaptation, and any need for reauthorization are recorded without erasing the earlier decision.