Skip to content

Feature: build the interactive benchmark-results dashboard promised in ROADMAP.md Phase 5 #744

Description

@yakew7

Where: ROADMAP.md Phase 5, results/ (results_fairness.csv, results_performance.csv, summary.csv, figures/), profiler.html (as the existing web/CLI split precedent).

Current gap: ROADMAP.md's Phase 5 checklist has one remaining unchecked item: - [ ] Fairness dashboard for the benchmark harness results (interactive results/ explorer, mirroring the Open Dataset Profiler's web/CLI split) - every other Phase 5 item is checked off, including the Open Dataset Profiler itself and the benchmark harness that produces results/. Right now faircode benchmark only writes static CSVs and PNG figures with no browser-based way to explore/filter them (by audit, model, strategy, metric, protected attribute) the way profiler.html lets you explore a single dataset's profile interactively.

Repro:

$ grep -n "Fairness dashboard for the benchmark harness" ROADMAP.md
- [ ] Fairness dashboard for the benchmark harness results (interactive `results/` explorer, mirroring the Open Dataset Profiler's web/CLI split)

Suggested approach: a new static page (e.g. benchmark.html + assets/benchmark-dashboard.js) that fetches/parses results/results_fairness.csv and results/results_performance.csv client-side (or accepts a drag-and-drop of a user's own results/ export) and renders filterable tables/charts - audit × model × strategy × metric - reusing the existing bootstrap-CI/permutation-p-value columns already in those CSVs, styled consistently with the design tokens shared by index.html/profiler.html.

Activity

  1. yakew7 commented on Sep 30, 2026

    @yakew7
    OwnerAuthor

    Fixed in 0399ded: added benchmark.html + assets/benchmark-dashboard.js + assets/benchmark.css - a client-side, filterable/sortable explorer for results/results_fairness.csv and results/results_performance.csv, mirroring the Open Dataset Profiler's web/CLI split. Loads via a bundled-results fetch button or drag-and-drop of your own faircode benchmark export; filters by audit/strategy/model/protected_attribute/metric plus a significant-only toggle, sortable table, and a bar chart. Verified with a real DOM-stub-driven test (no jsdom in this repo) that loads/filters/sorts/tab-switches against the actual results/*.csv and checks every number against pandas ground truth computed in the test itself, so it stays correct as results/ keeps moving. This was the last unchecked item in ROADMAP.md Phase 5, now complete. Full suite (519 passed), ruff, em-dash check, broken-links check, CODEOWNERS coverage all clean.

  2. added a commit that references this issue on Sep 30, 2026
    0399ded
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requesthelp wantedExtra attention is needed

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions