Skip to content

feat(dashboard): add hierarchical category taxonomy#1560

Merged
christso merged 1 commit into
result-row-id-sidecarsfrom
category-tree-taxonomy
Jun 29, 2026
Merged

feat(dashboard): add hierarchical category taxonomy#1560
christso merged 1 commit into
result-row-id-sidecarsfrom
category-tree-taxonomy

Conversation

@christso

Copy link
Copy Markdown
Collaborator

Summary

AgentV result categories now behave as slash-delimited analytics taxonomy paths, so security/network can be inspected as both a leaf and as part of the parent security rollup. Existing flat category labels remain valid, and explicit YAML category: values continue to override derived categories after canonicalization.

The Dashboard and results API now build category trees from loaded result metadata rather than persisted tree state. Parent drilldowns include descendant suites, leaf drilldowns stay precise, and compare score distributions filter by category metadata without parsing eval_path.

Category derivation is documented for generic eval filenames, named eval files, nested folders, and explicit overrides.

Validation

  • bun run build
  • bun test packages/core/test/evaluation/category.test.ts packages/core/test/evaluation/yaml-parser-metadata.test.ts apps/cli/test/commands/results/serve.test.ts apps/dashboard/src/lib/category-tree.test.ts apps/dashboard/src/lib/score-distribution.test.ts — 159 pass, 0 fail
  • bun run typecheck
  • bun run lint
  • bun --filter @agentv/dashboard test — 147 pass, 0 fail
  • bun --filter @agentv/core test — 2077 pass, 0 fail
  • Browser/API UAT with a temporary run bundle under /tmp/agentv-cat-tree-uat: verified /categories exposes category_tree, parent security totals include security/network, /category/security shows both parent and descendant suites, /category/security%2Fnetwork shows only the leaf suite, and a narrow viewport renders the category table without obvious overlap.

Post-Deploy Monitoring & Validation

  • This is local Dashboard/read-model behavior; no external production service monitoring is required.
  • CI should cover build, typecheck, lint, and package tests for the stacked branch.
  • After the base result-row sidecar work lands, run Dashboard against a real result bundle with nested categories and confirm parent rollups, leaf drilldown, and compare score distribution category filters still use category metadata only.

Risks


Compound Engineering
GPT-5

@cloudflare-workers-and-pages

Copy link
Copy Markdown

Deploying agentv with  Cloudflare Pages  Cloudflare Pages

Latest commit: d1d734e
Status: ✅  Deploy successful!
Preview URL: https://0396bc80.agentv.pages.dev
Branch Preview URL: https://category-tree-taxonomy.agentv.pages.dev

View logs

@christso christso merged commit a55a918 into result-row-id-sidecars Jun 29, 2026
1 check passed
@christso christso deleted the category-tree-taxonomy branch June 29, 2026 04:40
christso added a commit that referenced this pull request Jun 29, 2026
* fix(results): isolate row sidecars by target bundle

* fix(dashboard): split run experiment and target columns

* feat(dashboard): add hierarchical category taxonomy

Merge PR #1560 for Bead av-k0e after independent read-only code review reported no actionable issues and verification passed.

* fix(eval): stop surfacing provider staging logs (#1561)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant