@omni-oss/perf-dashboard

Graphs and dashboards for @omni-oss/task-bench performance-comparison data. Turns the benchmark artifacts published by publish-perf-comparison.yaml into a single self-contained report with two views:

• Cross-tool — one omni version vs. the other runners in the same runs.
• Version history — omni over its own releases, on a normalized slice (Linux · full preset · resource runs); ineligible versions are dropped and shown in a mandatory exclusion panel.

See DESIGN.md for the full architecture. The pipeline is source → normalize → analyze → Chart IR → render, with pluggable data sources and renderers, and a generic, domain-agnostic output IR.

CLI

# Full report (all views) from the published GitHub data → self-contained HTML.
perf-dashboard report \
  --source github --repo omni-oss/performance-comparison \
  --renderer html --out ./site

# From local `task-bench suite --json` output (single file or a folder of runs).
perf-dashboard report --source local-fs --path ./data.json --renderer markdown --out .

# Narrow to one view and spotlight a specific omni version.
perf-dashboard report --source github --repo omni-oss/performance-comparison \
  --views cross-tool --spotlight 0.4.1 --renderer html --out ./site

# Inspect the normalized data (no analysis) for debugging.
perf-dashboard inspect --path ./data.json

Sources: local-fs (--path), github (--repo owner/repo, optional --ref, --token/GITHUB_TOKEN). Renderers: json, markdown, html.

Library

import {
  GitHubDataSource,
  HtmlRenderer,
  run,
} from "@omni-oss/perf-dashboard";

const output = await run({
  source: new GitHubDataSource({ owner: "omni-oss", repo: "performance-comparison" }),
  renderer: new HtmlRenderer(),
});
// output.files -> [{ path: "index.html", content, mime }]

composeReport(runs, opts) is the pure core (no IO) if you want to bring your own source/renderer; it returns a Report — a serializable description of every view, chart, and data point.

Configuration

• PERF_DASHBOARD_SCENARIO_ALIASES — JSON object of { alias: canonical } scenario-name pairs, merged over the checked-in constant map (env wins). Used by the version-history view to bridge scenario renames across releases.