react-doctor-benchmarks

Reproducible react-doctor scores for popular open-source React frontends.

The scores below are produced by GitHub Actions on a weekly cron (and on demand). Every entry is scanned with npx react-doctor@latest --json --offline against a fresh clone of the upstream repo, and the resulting JSON is committed alongside this README so the leaderboard is fully auditable.

Leaderboard

Rank	Project	Score	Errors	Warnings	Files	Commit
1	executor	🟢 ███████████████████░ 96/100	3	8	7	ec3f61e
2	nodejs.org	🟢 █████████████████░░░ 86/100	0	196	179	125b760
3	tldraw	🟡 ██████████████░░░░░░ 70/100	7	145	76	2eb9f83
4	t3code	🟡 ██████████████░░░░░░ 68/100	0	763	256	131234b
5	better-auth	🟡 █████████████░░░░░░░ 64/100	0	628	266	cf59136
6	excalidraw	🟡 █████████████░░░░░░░ 63/100	1	555	177	b2b2815
7	mastra	🟡 █████████████░░░░░░░ 63/100	23	468	207	4df7cc7
8	payload	🟡 ████████████░░░░░░░░ 60/100	4	767	424	80dabd3
9	typebot	🟡 ███████████░░░░░░░░░ 57/100	4	382	206	85eb843
10	plane	🟡 ███████████░░░░░░░░░ 56/100	9	1944	833	4c1bdd1
11	medusajs/admin	🟡 ███████████░░░░░░░░░ 56/100	10	591	247	7747d05
12	rocket.chat	🟡 ██████████░░░░░░░░░░ 51/100	57	2179	970	2a927fa
13	twenty	🔴 ██████████░░░░░░░░░░ 48/100	85	1833	1240	0f8ee57
14	unkey	🔴 ██████████░░░░░░░░░░ 48/100	27	966	380	5a6c1ad
15	shadcn/ui	🔴 █████████░░░░░░░░░░░ 46/100	15	2349	1014	fc1ca40
16	trigger.dev	🔴 ████████░░░░░░░░░░░░ 42/100	37	1942	590	749dc46
17	formbricks	🔴 ████████░░░░░░░░░░░░ 41/100	11	4056	811	69ead97
18	langfuse	🔴 ███████░░░░░░░░░░░░░ 36/100	27	3167	852	cfb88be
19	tooljet	🔴 ███████░░░░░░░░░░░░░ 33/100	190	5976	1459	f2f18d1
20	onlook	🔴 ██████░░░░░░░░░░░░░░ 32/100	65	2034	416	a242be5
21	cal.com	🔴 ██████░░░░░░░░░░░░░░ 32/100	43	1582	425	a4a01a0
22	posthog	🔴 ██████░░░░░░░░░░░░░░ 31/100	682	5304	1833	8d12f45
23	appsmith	🔴 ██████░░░░░░░░░░░░░░ 31/100	145	2450	1314	9ab4a39
24	supabase	🔴 ██████░░░░░░░░░░░░░░ 30/100	52	3304	1315	0278672
25	sentry	🔴 ██████░░░░░░░░░░░░░░ 30/100	179	3196	1652	52ecb95
26	lobehub/lobe-chat	🔴 ██████░░░░░░░░░░░░░░ 30/100	298	6474	1619	c760171
27	dub	🔴 █████░░░░░░░░░░░░░░░ 24/100	60	3355	1209	a5fa025

_{Last updated 2026-05-08T13:25:23.597Z · react-doctor 0.1.0 · 27 scored, 0 skipped/failed · raw results in results/latest.json}

How it works

1. repos.yaml lists every benchmark target with its GitHub URL, the workspace project to scan, and any per-repo overrides (skip dead-code, skip install, etc.).
2. The benchmark workflow fans out across the list using a matrix strategy. Each job clones one repo, attempts pnpm/npm/yarn/bun install --ignore-scripts (auto-detected from the lockfile), and runs npx -y react-doctor@latest <scanDir> --json --offline --fail-on none. If install fails, it falls back to --no-dead-code and scans the source anyway so a working score still lands.
3. A final publish job downloads every per-repo artifact, writes results/latest.json, regenerates this README's leaderboard table, and commits the diff back to main using the default GITHUB_TOKEN. If nothing changed (idempotent rendering), the commit step is a no-op.

The harness pins nothing about the upstream repos by default — every entry tracks HEAD of its default branch, and the SHA actually scanned is recorded in each result row so any score is reproducible. To pin a specific commit, set the ref field on a repos.yaml entry to a branch, tag, or SHA.

Consuming the leaderboard

Every CI run writes results/leaderboard.json — a slim, stable JSON blob that downstream repos can fetch and drop in. It mirrors react-doctor's LeaderboardEntry interface so a one-shot replacement is straightforward.

Stable URL (always main, always the latest run):

https://raw.githubusercontent.com/millionco/react-doctor-benchmarks/main/results/leaderboard.json

Schema:

interface ConsumerLeaderboard {
  schemaVersion: 1;
  generatedAt: string;          // ISO 8601 UTC
  doctorVersion: string | null; // e.g. "0.0.47"
  source: { repo: string; path: string; docs: string };
  entries: Array<{
    slug: string;               // "tldraw"
    name: string;               // "tldraw"
    githubUrl: string;          // "https://github.com/tldraw/tldraw"
    packageName: string;        // workspace project name passed via --project
    score: number;              // 0–100
    errorCount: number;
    warningCount: number;
    fileCount: number;          // affectedFileCount in react-doctor's JsonReport
    commitSha: string | null;   // SHA we actually scanned
    scannedAt: string;
  }>;                           // sorted desc by score
}

Example: regenerate react-doctor's leaderboard-entries.ts from CI:

# .github/workflows/refresh-leaderboard.yml in millionco/react-doctor
name: refresh leaderboard
on:
  schedule:
    - cron: "0 7 * * 1"  # one hour after react-doctor-benchmarks runs
  workflow_dispatch:

jobs:
  refresh:
    runs-on: ubuntu-latest
    permissions: { contents: write, pull-requests: write }
    steps:
      - uses: actions/checkout@v4
      - name: fetch latest leaderboard
        run: |
          curl -sSfL \
            https://raw.githubusercontent.com/millionco/react-doctor-benchmarks/main/results/leaderboard.json \
            -o /tmp/leaderboard.json
      - name: codegen leaderboard-entries.ts
        run: node scripts/codegen-leaderboard.mjs /tmp/leaderboard.json \
              > packages/website/src/app/leaderboard/leaderboard-entries.ts
      - uses: peter-evans/create-pull-request@v6
        with:
          title: "chore: refresh leaderboard from react-doctor-benchmarks"
          commit-message: "chore: refresh leaderboard"
          branch: chore/refresh-leaderboard

Where scripts/codegen-leaderboard.mjs is whatever projection makes sense for your downstream — typically:

// scripts/codegen-leaderboard.mjs
import { readFileSync } from "node:fs";
const data = JSON.parse(readFileSync(process.argv[2], "utf8"));
const rows = data.entries.map((e) => `  ${JSON.stringify({
  name: e.name, githubUrl: e.githubUrl, packageName: e.packageName,
  score: e.score, errorCount: e.errorCount, warningCount: e.warningCount,
  fileCount: e.fileCount,
})},`).join("\n");
process.stdout.write(`// Auto-generated from ${data.source.repo} on ${data.generatedAt}\n` +
  `export const RAW_ENTRIES = [\n${rows}\n];\n`);

The blob is rewritten on every CI run, so even when scores don't change the generatedAt timestamp does — you can safely diff or skip in your downstream codegen.

Adding a project

Open a PR that adds an entry to repos.yaml. The schema is defined and validated in scripts/lib/config.ts:

- slug: my-project          # kebab-case, must be unique
  name: my-project          # display name in the leaderboard
  githubUrl: https://github.com/owner/repo
  scanDir: apps/web         # optional, default "."
  project: "@my/web"        # optional, passes --project to react-doctor
  packageManager: pnpm      # optional, auto-detected from lockfile
  skipDeadCode: false       # optional, true → pass --no-dead-code
  skipInstall: false        # optional, true → don't install (implies skipDeadCode)

Once merged, the entry shows up the next time the workflow runs (weekly cron, or click Run workflow on the benchmark action).

Reproducing locally

pnpm install
pnpm tsx scripts/benchmark-repo.ts dub        # scan one entry
pnpm tsx scripts/aggregate.ts                  # collect results/per-repo/*.json → results/latest.json
pnpm tsx scripts/render-readme.ts              # splice into README between markers

Set REACT_DOCTOR_VERSION=1.2.3 to pin a specific upstream version (defaults to latest).

Layout

Path	What
repos.yaml	Benchmark targets (canonical source of truth).
results/latest.json	Aggregated leaderboard data; auto-generated.
results/per-repo/<slug>.json	Per-repo result; auto-generated.
scripts/	The harness (matrix prep, single-repo benchmark, aggregator, README renderer).
.github/workflows/benchmark.yml	The automation.

Credits

Thirteen of the entries (tldraw, excalidraw, twenty, plane, formbricks, posthog, supabase, onlook, payload, sentry, cal.com, dub, nodejs.org) were originally compiled by hand for the react.doctor/leaderboard page in millionco/react-doctor. This repo turns that snapshot into a self-updating, auditable benchmark and adds ten more popular OSS React apps.

License

MIT.