frameguard 0.6.0

FrameGuard — performance regressions, testable. Capture Flutter frame timings, classify jank, enforce budgets, compare baselines, and fail CI on measurable UI regressions. Local by default. No telemetry.

FrameGuard

Performance regressions, testable.
Automated Flutter UI performance regression detection — budgets, baselines, and evidence-backed reports you can enforce in CI.

Docs · pub.dev · CLI · Changelog · Sponsor

---

Status

Version	0.6.0
Repository	github.com/theworker02/frameguard
Package	pub.dev/packages/frameguard
Docs site	theworker02.github.io/frameguard
License	MIT
Telemetry	None — local by default

---

Why FrameGuard?

Flutter already has excellent profilers. DevTools answers:

What happened during this profiling session?

FrameGuard answers:

Did performance regress, where, why is that likely, and should this build fail?

	DevTools	FrameGuard
Goal	Interactive exploration	Automation & regression gates
Output	Timelines you interpret	Budgets, baselines, matchers, CI exit codes
Audience	Humans in a profiling session	Humans and pipelines
Telemetry	N/A	None — local by default

FrameGuard is not a thin DevTools wrapper. It turns documented Flutter frame timings into reproducible explanations.

---

Features

• Session capture via SchedulerBinding.addTimingsCallback / FrameTiming
• Refresh-rate-aware budgets (60 / 90 / 120 / 144 Hz — no hardcoded 16.67 ms dogma)
• Jank severity (healthy / minor / major / severe)
• Percentiles — p50 / p90 / p95 / p99, histograms, streaks
• Build vs raster classification (derived, never claimed as certainty)
• Regions & rebuild counts (FrameGuardRegion)
• Traces, markers, sync tasks
• Explainability + recommendations tied to evidence (FG001–FG010)
• JSON / text / HTML reports (versioned schema)
• CSV · JUnit · SARIF · Markdown exporters for CI / PR comments
• Baselines & golden files (never silently overwritten)
• Multi-run statistics (median, MAD, CI; outliers flagged, not deleted)
• Test matchers + FrameGuardTest.measure
• CLI for compare / check / export / summary / doctor / baseline / history / suggest-budget
• Local history (JSONL) for gradual drift without a backend
• Reusable GitHub Action (.github/actions/frameguard-check)
• Optional overlay & runtime budget watcher
• Capability model — Unavailable beats fake zeros

---

Install

# pubspec.yaml
dependencies:
  frameguard: ^0.6.0

For tests-only usage you can keep it under dev_dependencies. Test helpers live in a separate library:

import 'package:frameguard/frameguard.dart';
import 'package:frameguard/frameguard_test.dart';

flutter pub get

---

Quick start

import 'package:flutter/material.dart';
import 'package:frameguard/frameguard.dart';

void main() {
  FrameGuard.initialize(
    config: FrameGuardConfig(
      samplingMode: SamplingMode.balanced,
      defaultBudget: FrameBudget.forRefreshRate(60, maxJankRate: 0.02),
    ),
  );

  runApp(
    const FrameGuardScope(
      child: FrameGuardOverlay(
        compact: true,
        child: MyApp(),
      ),
    ),
  );
}

Capture a session

final session = FrameGuard.startSession(name: 'home_scroll');
// …interact with the app…
final report = await session.stop();

debugPrint(report.summary());
await report.writeJson(File('reports/home_scroll.json'));
await report.writeHtml(File('reports/home_scroll.html'));

Assert in tests

testWidgets('product list stays within budget', (tester) async {
  await tester.pumpWidget(const App());

  final report = await FrameGuardTest.measure(
    tester,
    name: 'product_list',
    action: () async {
      await tester.fling(find.byType(ListView), const Offset(0, -1000), 1000);
      await tester.pumpAndSettle();
    },
  );

  expect(
    report,
    meetsFrameBudget(
      FrameBudget.forRefreshRate(
        60,
        maxJankFrames: 2,
        maxJankRate: 0.01,
      ),
    ),
  );
});

---

Budgets & baselines

final budget = FrameBudget(
  maxJankFrames: 2,
  maxJankRate: 0.01,
  maxP95FrameTime: const Duration(milliseconds: 16),
  maxP99FrameTime: const Duration(milliseconds: 24),
);

final evaluation = report.evaluate(budget);
if (!evaluation.passed) {
  fail(evaluation.summary());
}

Optional project YAML (Dart config remains primary):

cp frameguard.yaml.example frameguard.yaml
dart run frameguard config validate

Update baselines deliberately:

dart run frameguard baseline update reports/catalog.json --out baselines/catalog.json
dart run frameguard check reports/current.json --baseline baselines/catalog.json

Exit codes: 0 pass · 1 regression · 2 invalid config/report.

---

CLI

dart run frameguard help
dart run frameguard init
dart run frameguard doctor
dart run frameguard list
dart run frameguard explain report.json
dart run frameguard frames report.json --janky-only
dart run frameguard top report.json
dart run frameguard check reports/ --baseline baselines/ --profile mid_range --require-profile
dart run frameguard suggest-budget reports/home.json --format dart
dart run frameguard history append reports/home.json
dart run frameguard summary report.json --format markdown
dart run frameguard batch reports/ --format junit --out-dir reports/exports
dart run frameguard watch reports/ --once
dart run frameguard completions --shell bash

Full command map: doc/cli.md · docs/cli.html.

---

CI

Gate performance reports in GitHub Actions:

- uses: theworker02/frameguard/.github/actions/frameguard-check@main
  with:
    report: reports/
    baseline: baselines/
    profile: mid_range
    require-profile: 'true'

Or call the CLI directly:

dart run frameguard check reports/ --baseline baselines/ --require-profile --profile mid_range
dart run frameguard summary reports/catalog.json --format markdown >> "$GITHUB_STEP_SUMMARY"

More: doc/ci.md.

---

Regions, traces, markers

FrameGuardRegion(
  name: 'product_grid',
  child: ProductGrid(),
);

await FrameGuard.trace('open_product', () async {
  await Navigator.of(context).push(...);
});

FrameGuard.mark('products_loaded', metadata: {'count': 42});

await FrameGuard.measureTask('parse_catalog', () => parseCatalog(data));

---

Scenarios & statistics

final result = await FrameScenarioRunner(
  scenario: const FrameScenario(name: 'catalog_scroll', warmupFrames: 30),
  runs: 5,
  budget: FrameBudget.forRefreshRate(120),
).runAggregated((i) async {
  await scrollCatalog();
});

debugPrint(result.summary());

Details: doc/scenarios.md.

---

Platform support

Platform	Frame timings	Image cache	Overlay	Notes
Android / iOS	Yes	Yes	Yes	Prefer profile for gates
Desktop	Yes	Yes	Yes
Web	Best-effort	Best-effort	Yes	Absolute ms budgets vary by browser

Native extras (JankStats, signposts, GPU counters) are Unavailable until real adapters ship — we will not invent zeros. See FrameGuard.capabilities and platform adapters.

---

Privacy

• No analytics
• No telemetry
• No accounts
• No required dashboard
• Reports stay on disk under your control

---

Documentation

Product site	Guide, CLI, diagnostics, brand
Brand kit	Voice, color, logo
Docs index	Package guides
API cookbook	Common workflows
CI integration	Pipelines & Action
Diagnostics FG001–FG010	What each finding means
Contributing	Dev workflow
Changelog	SemVer history
Security	Vulnerability reporting
Code of Conduct	Community norms
Sponsor	thanks.dev / GitHub Sponsors

API reference (after publish): pub.dev/documentation/frameguard

---

Example & benchmarks

cd example && flutter run --profile
flutter test
flutter test benchmark/overhead_benchmark.dart

The example app includes intentional jank scenarios (rebuild storm, CPU stall, raster stress, …) so you can validate FrameGuard against known behavior.

---

Roadmap (honest)

Shipped foundations: sessions, budgets, baselines, CI, explainability, exporters, scenario stats, local history, budget suggestions.

Next (when evidence exists — never as fake features):

• Deeper GC / memory correlation where public APIs allow
• Android JankStats / iOS signpost adapters
• DevTools extension
• Historical trend dashboards (still local-first)

---

Contributing

PRs welcome. Please read CONTRIBUTING.md and the Code of Conduct.

dart format .
flutter analyze
flutter test

Bug reports & ideas: Issues

---

License

MIT — see LICENSE.

Support development

• thanks.dev/u/gh/theworker02
• GitHub Sponsors @theworker02

---

_{FrameGuard — make performance regressions testable the way functional regressions already are.}