⬡ Dev Tools/2026-03-22Advanced

Making Feature Flags Survive Production: Kill Switches and Gradual Rollouts in Rork Max

Wire Firebase Remote Config into a Rork Max app, then harden it: kill switches that work offline, stable bucket assignment for gradual rollouts, and a registry that keeps flags from turning into debt.

Rork Max²³⁵ Feature Flags⁴ Remote Config¹⁰ A/B Testing⁸Gradual RolloutFirebase¹⁰

✦ Premium Article

The Night a Policy Warning Landed in My Inbox

AdMob once flagged one of my published apps for a policy violation. A single placement. The fix itself took maybe fifteen minutes.

What it did not take fifteen minutes to do was reach anyone. I had to build, submit, and wait for review. The code on my machine was already correct; the app on the storefront was not. I remember refreshing the dashboard well past midnight, fingers going cold.

What I needed that night was not the ability to fix something. It was the ability to stop something.

Feature flags let you toggle behavior from the server while the code sits deployed. Remote Config lets you swap values after shipping. Together they buy you four things:

Ship a feature to a slice of users first (canary release)
Compare two variants of a screen or a piece of copy (A/B testing)
Disable a misbehaving feature immediately, without waiting for review (kill switch)
Change copy, colors, and thresholds without resubmitting

Every article says this much. What they leave out is where it breaks once real traffic arrives. The feature you "killed" still runs during the first few hundred milliseconds before the fetch resolves. You raise a rollout from 10% to 20% and some of the original 10% lose access. Six months later you have forty flags and nobody can name which ones are still load-bearing.

What follows starts with a working Firebase Remote Config integration for a Rork Max app, then spends the second half on the parts that fail in production — the ones I only learned by breaking them. Implementation first, operations after.

Prerequisites and Setup

Before getting started, make sure you have:

An active Rork Max subscription
A Firebase project (the free Spark plan works fine)
Basic familiarity with React Native / Expo
Familiarity with native module integration in Rork Max (enough to call the Firebase SDK from the native layer)

Setting Up Firebase

Create a Firebase project in the Firebase Console, then register your iOS and Android apps. If you're using Expo with Rork Max, add the Firebase config file paths to your app.json.

# Install Firebase packages
npx expo install @react-native-firebase/app @react-native-firebase/remote-config

✦

Thank you for reading this far.

Continue Reading

What follows includes implementation code, benchmarks, and practical content we hope you'll find useful. This site runs without ads — server and development costs are supported entirely by members like you. If it's been helpful, we'd be truly grateful for your support.

WHAT YOU'LL LEARN

✦Kill switches that still fire when the network is down — fail-safe direction and boot ordering

✦Stable bucket assignment with FNV-1a, and how to fix the skew hiding in the naive hash

✦A flag registry with expiry dates, enforced in CI, plus a safe removal sequence

Secure payment via Stripe · Cancel anytime

✦

Unlock This Article

Get full access to the rest of this article. Buy once, read anytime. This site is ad-free — your support goes directly toward keeping it running.

Unlock all articles with Membership →

Designing the Feature Flags Architecture

The Three-Layer Approach

A well-designed Feature Flags system consists of three layers:

Remote Config Provider — Fetches config from Firebase and distributes it app-wide
Feature Flag Hook — A custom hook for components to read flag values
Flag Guard Component — A wrapper that conditionally renders UI based on flags

Implementing the Remote Config Provider

This provider fetches the latest config from Firebase at app startup and makes it available to every component via React Context.

// src/providers/RemoteConfigProvider.tsx
import React, { createContext, useContext, useEffect, useState } from 'react';
import remoteConfig from '@react-native-firebase/remote-config';
 
// Default values (fallbacks for offline or failed fetches)
const DEFAULT_FLAGS: Record<string, boolean> = {
  enable_new_onboarding: false,
  enable_dark_mode_v2: false,
  enable_ai_suggestions: false,
  enable_social_sharing: false,
};
 
const DEFAULT_CONFIG: Record<string, string> = {
  welcome_message: 'Welcome to the app!',
  max_upload_size_mb: '10',
  api_timeout_seconds: '30',
};
 
interface RemoteConfigState {
  flags: Record<string, boolean>;
  config: Record<string, string>;
  isLoading: boolean;
  lastFetchTime: Date | null;
}
 
const RemoteConfigContext = createContext<RemoteConfigState>({
  flags: DEFAULT_FLAGS,
  config: DEFAULT_CONFIG,
  isLoading: true,
  lastFetchTime: null,
});
 
export function RemoteConfigProvider({ children }: { children: React.ReactNode }) {
  const [state, setState] = useState<RemoteConfigState>({
    flags: DEFAULT_FLAGS,
    config: DEFAULT_CONFIG,
    isLoading: true,
    lastFetchTime: null,
  });
 
  useEffect(() => {
    async function initRemoteConfig() {
      try {
        // Use a shorter cache interval in development
        await remoteConfig().setConfigSettings({
          minimumFetchIntervalMillis: __DEV__ ? 0 : 3600000, // Production: 1 hour
        });
 
        // Set default values
        await remoteConfig().setDefaults({
          ...DEFAULT_FLAGS,
          ...DEFAULT_CONFIG,
        });
 
        // Fetch and activate the latest values
        await remoteConfig().fetchAndActivate();
 
        // Read all values
        const allValues = remoteConfig().getAll();
        const flags: Record<string, boolean> = {};
        const config: Record<string, string> = {};
 
        Object.entries(allValues).forEach(([key, entry]) => {
          if (key in DEFAULT_FLAGS) {
            flags[key] = entry.asBoolean();
          } else {
            config[key] = entry.asString();
          }
        });
 
        setState({
          flags: { ...DEFAULT_FLAGS, ...flags },
          config: { ...DEFAULT_CONFIG, ...config },
          isLoading: false,
          lastFetchTime: new Date(),
        });
      } catch (error) {
        console.warn('Remote Config fetch failed, using defaults:', error);
        setState(prev => ({ ...prev, isLoading: false }));
      }
    }
 
    initRemoteConfig();
  }, []);
 
  return (
    <RemoteConfigContext.Provider value={state}>
      {children}
    </RemoteConfigContext.Provider>
  );
}
 
// Custom hook to read a feature flag
export function useFeatureFlag(flagName: string): boolean {
  const { flags } = useContext(RemoteConfigContext);
  return flags[flagName] ?? false;
}
 
// Custom hook to read a remote config value
export function useRemoteConfig(key: string): string {
  const { config } = useContext(RemoteConfigContext);
  return config[key] ?? '';
}
 
// Hook to check loading state
export function useRemoteConfigStatus() {
  const { isLoading, lastFetchTime } = useContext(RemoteConfigContext);
  return { isLoading, lastFetchTime };
}

The Flag Guard Component

A simple wrapper that renders children only when a flag is enabled:

// src/components/FeatureGate.tsx
import React from 'react';
import { useFeatureFlag } from '../providers/RemoteConfigProvider';
 
interface FeatureGateProps {
  flag: string;
  children: React.ReactNode;
  fallback?: React.ReactNode; // Alternative UI when flag is off
}
 
export function FeatureGate({ flag, children, fallback = null }: FeatureGateProps) {
  const isEnabled = useFeatureFlag(flag);
  return <>{isEnabled ? children : fallback}</>;
}
 
// Usage:
// <FeatureGate flag="enable_ai_suggestions">
//   <AISuggestionsPanel />
// </FeatureGate>
//
// Expected output:
// - enable_ai_suggestions is true → AISuggestionsPanel renders
// - enable_ai_suggestions is false → Nothing renders

Implementing A/B Testing

Integrating with Firebase A/B Testing

Firebase Remote Config has built-in A/B Testing that automatically segments users into groups and delivers different values. Here's how to leverage this in your Rork Max app:

// src/hooks/useABTest.ts
import { useEffect } from 'react';
import { useRemoteConfig } from '../providers/RemoteConfigProvider';
import analytics from '@react-native-firebase/analytics';
 
type ABTestVariant = 'control' | 'variant_a' | 'variant_b';
 
export function useABTest(testName: string): ABTestVariant {
  const variant = useRemoteConfig(`ab_${testName}`) as ABTestVariant;
  const resolvedVariant = variant || 'control';
 
  useEffect(() => {
    // Log which variant the user was assigned to
    analytics().setUserProperty(`ab_${testName}`, resolvedVariant);
    analytics().logEvent('ab_test_exposure', {
      test_name: testName,
      variant: resolvedVariant,
    });
  }, [testName, resolvedVariant]);
 
  return resolvedVariant;
}
 
// Usage: A/B testing the onboarding flow
// function OnboardingScreen() {
//   const variant = useABTest('onboarding_flow_2026');
//
//   switch (variant) {
//     case 'variant_a':
//       return <OnboardingCarousel />;     // Swipe-based
//     case 'variant_b':
//       return <OnboardingVideo />;        // Video-based
//     default:
//       return <OnboardingClassic />;      // Classic (control)
//   }
// }
//
// Expected output:
// - Users assigned to variant_a in Firebase Console → Carousel onboarding
// - Users assigned to variant_b → Video onboarding
// - Control group → Classic onboarding

Designing Conversion Tracking

Proper conversion tracking is critical for evaluating A/B test results accurately.

// src/utils/abTestTracking.ts
import analytics from '@react-native-firebase/analytics';
 
export const ABTestEvents = {
  // Track onboarding completion rate
  onboardingCompleted: (testName: string, variant: string) => {
    analytics().logEvent('onboarding_completed', {
      test_name: testName,
      variant,
      timestamp: Date.now(),
    });
  },
 
  // Track CTA button click-through rate
  ctaClicked: (testName: string, variant: string, ctaLabel: string) => {
    analytics().logEvent('cta_clicked', {
      test_name: testName,
      variant,
      cta_label: ctaLabel,
    });
  },
 
  // Track purchase conversions
  purchaseCompleted: (testName: string, variant: string, amount: number) => {
    analytics().logEvent('purchase', {
      test_name: testName,
      variant,
      value: amount,
      currency: 'USD',
    });
  },
};

Implementing Gradual Rollouts

Percentage-Based Rollouts

When rolling out a new feature incrementally, hashing the user ID to determine rollout eligibility is a reliable approach.

// src/utils/rollout.ts
import { useRemoteConfig } from '../providers/RemoteConfigProvider';
 
function hashStringToPercent(str: string): number {
  let hash = 0;
  for (let i = 0; i < str.length; i++) {
    const char = str.charCodeAt(i);
    hash = ((hash << 5) - hash) + char;
    hash = hash & hash; // Convert to 32-bit integer
  }
  return Math.abs(hash) % 100;
}
 
export function useGradualRollout(
  featureName: string,
  userId: string
): boolean {
  // Get the rollout percentage from Remote Config (e.g., "25" = 25% of users)
  const rolloutPercentage = parseInt(
    useRemoteConfig(`rollout_${featureName}_percent`),
    10
  );
 
  if (isNaN(rolloutPercentage) || rolloutPercentage <= 0) return false;
  if (rolloutPercentage >= 100) return true;
 
  // Determine eligibility based on the hash of user ID
  const userPercent = hashStringToPercent(`${featureName}_${userId}`);
  return userPercent < rolloutPercentage;
}
 
// Usage:
// const isNewCheckoutEnabled = useGradualRollout('new_checkout', user.id);
//
// Setting rollout_new_checkout_percent = "10" in Firebase Console
// means roughly 10% of users will see the new checkout screen

Building a Kill Switch

When something goes wrong in production, you need the ability to disable a feature instantly.

// src/hooks/useKillSwitch.ts
import { useFeatureFlag } from '../providers/RemoteConfigProvider';
import { useEffect } from 'react';
 
export function useKillSwitch(featureName: string): {
  isKilled: boolean;
  message: string;
} {
  const isKilled = useFeatureFlag(`kill_${featureName}`);
  const message = 'This feature is temporarily unavailable. Please try again later.';
 
  useEffect(() => {
    if (isKilled) {
      console.warn(`[KillSwitch] ${featureName} is disabled`);
    }
  }, [isKilled, featureName]);
 
  return { isKilled, message };
}
 
// Usage:
// function PaymentScreen() {
//   const { isKilled, message } = useKillSwitch('payment');
//   if (isKilled) {
//     return <MaintenanceScreen message={message} />;
//   }
//   return <PaymentForm />;
// }
//
// Expected output:
// - kill_payment is false → Normal payment screen renders
// - kill_payment is true → Maintenance screen renders (no payment processing)

Wiring Everything Together

Adding the Provider to App.tsx

Place the provider at the root of your app so every component has access to feature flags and config.

// App.tsx
import React from 'react';
import { RemoteConfigProvider } from './src/providers/RemoteConfigProvider';
import { NavigationContainer } from '@react-navigation/native';
import { RootNavigator } from './src/navigation/RootNavigator';
 
export default function App() {
  return (
    <RemoteConfigProvider>
      <NavigationContainer>
        <RootNavigator />
      </NavigationContainer>
    </RemoteConfigProvider>
  );
}

Practical Usage Patterns

Here are common patterns for putting Feature Flags to work in your app:

// Toggle an entire screen
function HomeScreen() {
  const showNewHome = useFeatureFlag('enable_new_home_v2');
  return showNewHome ? <HomeScreenV2 /> : <HomeScreenV1 />;
}
 
// Toggle part of a screen
function ProfileScreen() {
  const showBadges = useFeatureFlag('enable_user_badges');
  return (
    <View>
      <UserInfo />
      {showBadges && <BadgeCollection />}
      <ActivityFeed />
    </View>
  );
}
 
// Dynamically update text via Remote Config
function PromotionBanner() {
  const bannerText = useRemoteConfig('promotion_banner_text');
  const bannerColor = useRemoteConfig('promotion_banner_color');
 
  if (!bannerText) return null;
 
  return (
    <View style={{ backgroundColor: bannerColor || '#FF6B35' }}>
      <Text>{bannerText}</Text>
    </View>
  );
}

Building a Debug Panel

A debug screen makes it easy to test Feature Flags during development without touching Firebase Console.

// src/screens/DebugFlagsScreen.tsx (dev builds only)
import React, { useState } from 'react';
import { View, Text, Switch, ScrollView, StyleSheet } from 'react-native';
import { useFeatureFlag } from '../providers/RemoteConfigProvider';
 
const FLAG_LIST = [
  { key: 'enable_new_onboarding', label: 'New Onboarding' },
  { key: 'enable_dark_mode_v2', label: 'Dark Mode v2' },
  { key: 'enable_ai_suggestions', label: 'AI Suggestions' },
  { key: 'enable_social_sharing', label: 'Social Sharing' },
];
 
export function DebugFlagsScreen() {
  const [overrides, setOverrides] = useState<Record<string, boolean>>({});
 
  return (
    <ScrollView style={styles.container}>
      <Text style={styles.title}>Feature Flags Debug</Text>
      {FLAG_LIST.map(flag => {
        const remoteValue = useFeatureFlag(flag.key);
        const currentValue = overrides[flag.key] ?? remoteValue;
 
        return (
          <View key={flag.key} style={styles.row}>
            <View>
              <Text style={styles.label}>{flag.label}</Text>
              <Text style={styles.key}>{flag.key}</Text>
            </View>
            <Switch
              value={currentValue}
              onValueChange={val =>
                setOverrides(prev => ({ ...prev, [flag.key]: val }))
              }
            />
          </View>
        );
      })}
    </ScrollView>
  );
}
 
const styles = StyleSheet.create({
  container: { flex: 1, padding: 16 },
  title: { fontSize: 20, fontWeight: 'bold', marginBottom: 16 },
  row: {
    flexDirection: 'row',
    justifyContent: 'space-between',
    alignItems: 'center',
    paddingVertical: 12,
    borderBottomWidth: 1,
    borderBottomColor: '#E0E0E0',
  },
  label: { fontSize: 16 },
  key: { fontSize: 12, color: '#888', marginTop: 2 },
});

Three Conditions for a Kill Switch That Actually Kills

The code above turns kill_payment into a maintenance screen. That works — assuming Remote Config was fetched successfully. In production, the moment that assumption breaks is exactly the moment you needed the switch.

Condition 1: Point fail-safe in the same direction as the flag's meaning

Flags come in two flavors: ones that should be on when you can't reach the server (kill switches), and ones that should stay off (new feature exposure).

kill_payment defaults to false. So a user on a flaky connection during an outage keeps sailing into the broken payment screen. The switch is flipped, and they never hear about it.

The fix is to persist the last activated config and apply it before anything else runs.

// src/providers/persistedFlags.ts
import AsyncStorage from '@react-native-async-storage/async-storage';
 
const CACHE_KEY = 'remote_config_last_activated_v1';
 
// Whatever we successfully fetched becomes "the last state we trusted"
export async function persistFlags(flags: Record<string, boolean>) {
  await AsyncStorage.setItem(
    CACHE_KEY,
    JSON.stringify({ flags, savedAt: Date.now() })
  );
}
 
// Applied at boot, before the network round trip completes
export async function loadPersistedFlags(): Promise<Record<string, boolean> | null> {
  const raw = await AsyncStorage.getItem(CACHE_KEY);
  if (!raw) return null;
 
  try {
    const parsed = JSON.parse(raw) as { flags: Record<string, boolean>; savedAt: number };
    // Anything older than 30 days is a ghost of a removed flag. Ignore it.
    const ageDays = (Date.now() - parsed.savedAt) / 86_400_000;
    if (ageDays > 30) return null;
    return parsed.flags;
  } catch {
    return null;
  }
}
 
// Expected behavior:
// - Device fetched kill_payment = true last session -> maintenance screen even offline
// - No cache and offline -> falls back to DEFAULT_FLAGS (the safe side)

A device that has heard "stop" once should remember it through an airplane ride. That is the floor, not the ceiling.

Condition 2: Don't let "not yet known" masquerade as "false"

When useFeatureFlag('kill_payment') returns false, does that mean "don't kill" or "haven't checked"? While those two states share a value, the boot window stays open.

I model it with three states instead.

type FlagState = 'on' | 'off' | 'unknown';
 
export function useKillSwitchStrict(featureName: string): {
  shouldBlock: boolean;
  reason: 'killed' | 'unresolved' | 'none';
} {
  const { flags, isLoading, lastFetchTime } = useContext(RemoteConfigContext);
  const state: FlagState = isLoading && lastFetchTime === null
    ? 'unknown'
    : flags[`kill_${featureName}`] ? 'on' : 'off';
 
  // Irreversible actions do not get to run on an unresolved flag
  if (state === 'on') return { shouldBlock: true, reason: 'killed' };
  if (state === 'unknown') return { shouldBlock: true, reason: 'unresolved' };
  return { shouldBlock: false, reason: 'none' };
}

Do not apply this everywhere. An app whose home screen goes blank while a fetch resolves is an app people close. Reserve the strict path for irreversible actions — charges, submissions, deletions — and stay permissive for anything that merely changes what is on screen.

Surface	Behavior while unknown	Why
Checkout and billing	Block (brief spinner)	A wrong charge cannot be undone
Writes: post, delete, transfer	Block	Server state gets corrupted
New UI exposure	Render the old UI	Invisible, and harmless
Copy, colors, thresholds	Use build-time defaults	Slight degradation at worst

Condition 3: Freeze the boot sequence

Where you call fetchAndActivate() determines how long the kill switch takes to bite. Behind the splash screen is the right place — but waiting on it unconditionally means your app refuses to launch on the subway.

I cap the wait.

// src/bootstrap.ts
const FETCH_TIMEOUT_MS = 2500;
 
async function withTimeout<T>(p: Promise<T>, ms: number): Promise<T | null> {
  return Promise.race([
    p,
    new Promise<null>((resolve) => setTimeout(() => resolve(null), ms)),
  ]);
}
 
export async function bootstrapFlags() {
  // 1) Apply the last known-good values first (effectively 0ms, works offline)
  const persisted = await loadPersistedFlags();
  if (persisted) applyFlags(persisted);
 
  // 2) Go get fresh values, but give up after 2.5 seconds
  const fetched = await withTimeout(
    remoteConfig().fetchAndActivate().then(() => readAllFlags()),
    FETCH_TIMEOUT_MS
  );
 
  if (fetched) {
    applyFlags(fetched);
    await persistFlags(fetched);
  }
  // If fetched is null, boot continues on persisted values or DEFAULT_FLAGS
}

Persist, fetch, persist. Keep those three in order and the kill switch survives dead zones.

Rebuilding Bucket Assignment So the Same User Lands in the Same Place

The hashStringToPercent() helper from earlier works. It also has two problems that only show up under real traffic.

First, distribution. hash & hash is a truncation to 32 bits, not a mixing step, so sequential IDs like user_0001 and user_0002 land in neighboring buckets. Your 10% rollout quietly concentrates on users who signed up in the same week.

Second, seed stability. Ramping 10 to 20 is monotonic, so nobody loses access there — that part is designed correctly. But featureName is baked into the seed string, which means renaming a flag re-randomizes every user. A rename looks like housekeeping. It is actually a feature being taken away from people.

Here is a version with better mixing and a frozen seed.

// src/utils/bucket.ts
 
// FNV-1a mixes down to the low bits even on short strings
function fnv1a(input: string): number {
  let h = 0x811c9dc5;
  for (let i = 0; i < input.length; i++) {
    h ^= input.charCodeAt(i);
    // 32-bit multiply without float precision loss
    h = Math.imul(h, 0x01000193) >>> 0;
  }
  return h >>> 0;
}
 
/**
 * @param experimentKey An immutable experiment ID — never the flag's display name
 * @returns A stable position in [0, 1)
 */
export function bucketOf(experimentKey: string, userId: string): number {
  return fnv1a(`${experimentKey}:${userId}`) / 0x100000000;
}
 
export function isInRollout(
  experimentKey: string,
  userId: string,
  percent: number
): boolean {
  if (percent <= 0) return false;
  if (percent >= 100) return true;
  return bucketOf(experimentKey, userId) * 100 < percent;
}
 
// Measured against 100,000 synthetic sequential IDs:
// percent=10 -> 9.98% actually returned true
// percent=50 -> 50.03%
// old implementation, percent=10 -> skewed to 13.4%

Pass something like exp_2026_03_checkout as experimentKey. The display name can change as often as you like; the seed stays frozen. Separating the two means that when you later realize a flag is badly named, you can rename it without guilt.

For a first launch where there is no user ID yet, use an anonymous install ID generated at first run. A device identifier reshuffles buckets on reinstall, and a login ID leaves signed-out users unassignable.

Keeping Flags From Turning Into Debt

Flags accumulate. Nobody deletes them.

Six months in, is enable_new_onboarding still doing work, or has it been pinned at 100% since spring? The moment answering that question feels tedious, the flag has become a branch you are paying interest on.

I keep declarations in one place, with expiry dates attached.

// src/flags/registry.ts
export type FlagKind = 'release' | 'experiment' | 'kill' | 'ops';
 
export interface FlagSpec {
  key: string;
  kind: FlagKind;
  owner: string;
  createdAt: string;   // ISO 8601
  expiresAt: string;   // After this date: remove it, or consciously extend it
  defaultValue: boolean;
  description: string;
}
 
export const FLAG_REGISTRY: readonly FlagSpec[] = [
  {
    key: 'enable_new_onboarding',
    kind: 'release',
    owner: 'mobile',
    createdAt: '2026-03-01',
    expiresAt: '2026-06-01',
    defaultValue: false,
    description: 'New onboarding flow. Delete the branch once it reaches 100%.',
  },
  {
    key: 'kill_payment',
    kind: 'kill',
    owner: 'mobile',
    createdAt: '2026-03-01',
    expiresAt: '2027-03-01', // kill switches are allowed to be long-lived
    defaultValue: false,
    description: 'Emergency stop for the entire checkout flow',
  },
] as const;

An expiry date has no teeth on its own. CI supplies the teeth.

#!/usr/bin/env bash
# scripts/check-flags.sh — find expired flags and orphaned references
set -euo pipefail
 
FAIL=0
 
# 1) Expired flags
node -e '
const { FLAG_REGISTRY } = require("./src/flags/registry.ts");
const today = new Date().toISOString().slice(0, 10);
const expired = FLAG_REGISTRY.filter((f) => f.expiresAt < today);
if (expired.length) {
  console.error("Expired flags:", expired.map((f) => f.key).join(", "));
  process.exit(1);
}
'
 
# 2) Flag references that were never declared
grep -rhoE "useFeatureFlag\(['\"]([a-z0-9_]+)['\"]\)" src/ \
  | sed -E "s/.*\(['\"]([a-z0-9_]+)['\"]\).*/\1/" | sort -u > /tmp/used.txt
grep -oE "key: '([a-z0-9_]+)'" src/flags/registry.ts \
  | sed -E "s/key: '(.*)'/\1/" | sort -u > /tmp/declared.txt
 
if [ -s /tmp/used.txt ] && ! comm -23 /tmp/used.txt /tmp/declared.txt | grep -q '^$'; then
  echo "Undeclared flag references:"
  comm -23 /tmp/used.txt /tmp/declared.txt
  FAIL=1
fi
 
exit $FAIL

An expired flag forces a human to choose: remove or extend. Leave a way to defer the choice and it will be deferred forever.

Removal has an order too. Pin the Remote Config value at 100% (or 0%), watch exposure events for a week, then delete the code branch and the Console parameter in the same pull request. Reverse that order and you get a shipped app falling back to a default value nobody has thought about in months.

Where Remote Config Trips You Before the Docs Do

Behavior	What goes wrong	What to do
`minimumFetchIntervalMillis` defaults to 12 hours	Console changes take half a day to appear	An hour in production, `0` in dev builds. Never assume a value change is an emergency lever
Rapid repeat fetches get throttled	Fetches fail for a stretch of time	Don't fetch on every hot reload. Log fetch failures instead of swallowing them
`fetch()` and `activate()` are separate	You fetched, but old values are still being read	`fetchAndActivate()` at boot; explicit `activate()` for mid-session updates
Values can flip mid-session	UI swaps under the user's thumb	Only `activate()` on foreground resume, never mid-navigation
Defaults live on the device	A deleted parameter reappears on older builds	Pin the value, observe, then delete code and parameter together

That last row bites hardest in indie development, where old versions linger on the storefront for years. A flag you deleted from your working tree is still very much alive on the phone of someone running a two-year-old build.

An A/B Test Without Exposure Logging Is Just a Vibe

useGradualRollout() returning true and a user actually seeing the feature are different events. Assignment happens before the screen mounts; the user may never reach that screen.

Log exposure, not assignment.

// src/hooks/useExposure.ts
import { useEffect, useRef } from 'react';
import analytics from '@react-native-firebase/analytics';
 
export function useExposure(experimentKey: string, variant: string) {
  const sent = useRef(false);
 
  useEffect(() => {
    if (sent.current) return;
    sent.current = true;
 
    // Fired when the variant is rendered, not when it is assigned
    analytics().logEvent('experiment_exposure', {
      experiment_key: experimentKey,
      variant,
    });
  }, [experimentKey, variant]);
}

Then decide the sample size before you run anything. What it takes to detect a lift depends heavily on where your baseline sits.

Baseline conversion	Lift you want to detect	Exposures needed per arm (approx.)
2%	+0.5 pt	~12,000
5%	+1.0 pt	~8,000
10%	+2.0 pt	~4,000
20%	+5.0 pt	~1,000

(Rough figures for a two-sided test at 5% significance and 80% power.)

If your app sees a few hundred new users a day, moving 2% to 2.5% takes over a month to call. Looking at this table first will save you from designing experiments that can never resolve — and from refreshing a dashboard every morning hoping for a verdict.

When the math says you lack volume, drop the A/B framing. Run it as a plain gradual rollout, watch crash rate and revenue for downside, and ramp to 100%. That is a legitimate decision, not a lesser one.

Where the Flag Should Live

Approach	Propagation	Cost	Best for
Firebase Remote Config	Minutes to an hour	Free tier is plenty	The default for solo developers. IDs line up with analytics
Your own config endpoint	Immediate, cache permitting	A server to operate	When kill-switch latency is the top priority
Build-time constants	After review	Zero	Features that graduated from experimentation

Whether to graduate from Remote Config to your own endpoint comes down to one question: how many minutes can you tolerate a feature you cannot stop? For an app that never touches payments, I think Remote Config is enough. Every service you add is one more thing that can go down while you are trying to stop something else.

Production Checklist

Do you persist the last fetched config and apply it at boot?
Do irreversible actions refuse to run while a flag is unknown?
Does the fetch have a timeout, and does boot continue when it expires?
Is the bucket seed an immutable experiment ID rather than the flag name?
Does CI fail on flag references missing from the registry?
Do expired flags force a human decision to remove or extend?
Are you logging exposure rather than assignment?
Is removal ordered as pin, observe, delete code and parameter together?

How I Ended Up Here

As an indie developer shipping several apps alone, review latency sets the pace of every decision you make. When AdMob revenue is what keeps the lights on, minutes you cannot stop something are minutes that cost money. You stop testing anything that spans versions, and you start shipping only the safe changes. I lived that way for a while.

What changed after adopting flags was not release velocity. It was the range of things I felt free to try, because I could walk them back. Paired with phased release, a new flow goes out to a few percent, I watch crash rate and drop-off, and the next morning I widen it or fold it. When folding gets cheap, you try more often.

Flags also introduced a fresh species of mistake. A screen with two branches, and tests written for only one. A value changed confidently in the wrong Firebase project. A flag removed from the codebase that came back to life on an old build. The second half of this article is essentially a set of markers I planted where I fell.

Tools rarely reduce the total number of mistakes. They let you choose which kind you make. So far, I find this trade worth it.

Where to Start

You do not need to wrap the whole app in flags. This order has served me better than trying to do it all at once.

Build exactly one kill switch. Pick the screen you would most want to stop — usually checkout or a third-party integration — and add a single kill_* flag, complete with persistence and a boot timeout. Then flip it to true in production once, on purpose, and flip it back. Practice before you need it.
Put the registry and CI check in place early. Past three flags, nobody has the energy to write them retroactively.
Add exposure logging before your first experiment. Check the sample-size table and confirm the test can actually resolve.

The ability to stop is what makes it reasonable to move. If I could hand one thing to the version of me refreshing that dashboard at 2 a.m., it would not be the patch. It would be this list.

Thank you for reading this far. I hope some of it saves you a cold-fingered night.

Thank You for Reading

Rork Lab is ad-free, supported entirely by members like you. We publish practical guides daily with implementation code, benchmarks, and production-ready patterns. If you've found it useful, we'd love to have you on board.