Research findings — UX polish (v0.4.0)

What the UX review surfaced, what v0.4.0 adopted, and what is deferred with reasons. Companion: research-compliance.md.

Adopted

• Tabular numerals everywhere numbers align — font-variant-numeric: tabular-nums on table.data and .mono in theme.css, plus number-width: "tabular" in the Typst tables. Cheap, immediately improves scanning of score/duration columns.
• Reports as documents, not dashboards — the rebuilt Reports page leads with the compliance artifacts (document ID, type, SHA-256, preview iframe, PDF download) and demotes the v1 metric digests to “Quick digests”. This matches how the artifacts are actually consumed: preview in the browser, file the PDF.

Deferred (with reasons)

• ⌘K command palette — real value once the page count grows, but it wants a global search backend (experiments/traces/logs by id) to be more than a nav menu; bundling a half-feature into a release about document quality would dilute it. Roadmap.
• Brush/range-select on Overview charts — echarts dataZoom brush is easy to enable but needs coordinated re-querying semantics across every panel to not be misleading; deferred until the selection model is designed (likely alongside Mosaic crossfiltering, Phase 2).
• BubbleUp-style “explain this spike” — needs a designed drill-down contract (dimension ranking over the store); roadmap.
• Notebook-style ad-hoc reports (user-composed blocks) — the v2 content model (ReportDoc/Block) is built to allow this later; not exposed in v0.4.0.
• R4/R5 templates (service-level deep-dive, regulatory run-log export) — pending real auditor feedback on R1–R3.
• Triage inbox for open weaknesses — the R1 open-weaknesses table is the read model for this; the workflow itself is roadmap.