[Jammy2211]

The campaign table was too dense, status words wrapped awkwardly, and each instrument appeared as another likelihood link. The dashboard now uses three campaign columns and a single Inference Results → PyAutoLens → Imaging hierarchy with Pulse-style likelihood rows. Each likelihood opens a page with an instrument selector; Delaunay/Rectangular Numba views stay separate from JAX.

Scripts Changed

• insight/campaigns.py: Campaign / Recent progress / Next steps; omit unset last-check-in; retain full ledger.
• insight/setup_browser.*: Pulse list/selector styling, instrument navigation/history, separate implementation filters, scoped references/manifests/results, readable task tokens and headers.
• insight/board.py: remove duplicate project section and move routine provenance into collapsed Evidence details. Actual failed/stale/cached/local warnings remain visible.
• Browser/unit checks and generated dashboards updated. Original setup URLs still work.

Validation checklist

• 170 tests; Ruff/format; offline schema/feed checks; diff check.
• Real Chromium synthetic JAX/Numba and two-instrument isolation, Back navigation, legacy URLs, clipboard/no-JS/mobile checks.
• Actual captured feed preview inspected at desktop and390px; task status/header wrapping fixed.
• Independent CLEAN review; valid mixed-backend/two-instrument fixture verifies records, manifests, references and comparisons stay isolated.
• Heart STALE only: release validation incomplete: no rehearsal for current source; no RED/YELLOW reasons.

Witness: one likelihood list rather than per-instrument duplicates; instrument switches retain implementation; no mixed timing records. Current feed has no Numba measurements, so those requested view filters explicitly show no recorded results. No scientific data, campaign state or producer schema changed.

Closes #17. User-requested refinement of the approved dashboard; continuing same-session completion and merge after green CI.