diff --git a/CLAUDE.md b/CLAUDE.md
index 8879e9d..96eb2d9 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -17,6 +17,11 @@ npm install
 # Start server (runs on port 8080)
 npm start
 
+# Start with specific account selection strategy
+npm start -- --strategy=sticky      # Cache-optimized (stays on same account)
+npm start -- --strategy=round-robin # Load-balanced (rotates every request)
+npm start -- --strategy=hybrid      # Smart distribution (default)
+
 # Start with model fallback enabled (falls back to alternate model when quota exhausted)
 npm start -- --fallback
 
@@ -50,6 +55,9 @@ npm run test:images        # Image processing
 npm run test:caching       # Prompt caching
 npm run test:crossmodel    # Cross-model thinking signatures
 npm run test:oauth         # OAuth no-browser mode
+
+# Run strategy unit tests (no server required)
+node tests/test-strategies.cjs
 ```
 
 ## Architecture
@@ -83,9 +91,18 @@ src/
 ├── account-manager/            # Multi-account pool management
 │   ├── index.js                # AccountManager class facade
 │   ├── storage.js              # Config file I/O and persistence
-│   ├── selection.js            # Account picking (round-robin, sticky)
 │   ├── rate-limits.js          # Rate limit tracking and state
-│   └── credentials.js          # OAuth token and project handling
+│   ├── credentials.js          # OAuth token and project handling
+│   └── strategies/             # Account selection strategies
+│       ├── index.js            # Strategy factory (createStrategy)
+│       ├── base-strategy.js    # Abstract base class
+│       ├── sticky-strategy.js  # Cache-optimized sticky selection
+│       ├── round-robin-strategy.js  # Load-balanced rotation
+│       ├── hybrid-strategy.js  # Smart multi-signal distribution
+│       └── trackers/           # State trackers for hybrid strategy
+│           ├── index.js        # Re-exports trackers
+│           ├── health-tracker.js    # Account health scores
+│           └── token-bucket-tracker.js  # Client-side rate limiting
 │
 ├── auth/                       # Authentication
 │   ├── oauth.js                # Google OAuth with PKCE
@@ -161,7 +178,8 @@ public/
 - **src/webui/index.js**: WebUI backend handling API routes (`/api/*`) for config, accounts, and logs
 - **src/cloudcode/**: Cloud Code API client with retry/failover logic, streaming and non-streaming support
   - `model-api.js`: Model listing, quota retrieval (`getModelQuotas()`), and subscription tier detection (`getSubscriptionTier()`)
-- **src/account-manager/**: Multi-account pool with sticky selection, rate limit handling, and automatic cooldown
+- **src/account-manager/**: Multi-account pool with configurable selection strategies, rate limit handling, and automatic cooldown
+  - Strategies: `sticky` (cache-optimized), `round-robin` (load-balanced), `hybrid` (smart distribution)
 - **src/auth/**: Authentication including Google OAuth, token extraction, database access, and auto-rebuild of native modules
 - **src/format/**: Format conversion between Anthropic and Google Generative AI formats
 - **src/constants.js**: API endpoints, model mappings, fallback config, OAuth config, and all configuration values
@@ -170,12 +188,36 @@ public/
 - **src/errors.js**: Custom error classes (`RateLimitError`, `AuthError`, `ApiError`, etc.)
 
 **Multi-Account Load Balancing:**
-- Sticky account selection for prompt caching (stays on same account across turns)
+- Configurable selection strategy via `--strategy` flag or WebUI
+- Three strategies available:
+  - **Sticky** (`--strategy=sticky`): Best for prompt caching, stays on same account
+  - **Round-Robin** (`--strategy=round-robin`): Maximum throughput, rotates every request
+  - **Hybrid** (`--strategy=hybrid`, default): Smart selection using health + tokens + LRU
 - Model-specific rate limiting via `account.modelRateLimits[modelId]`
 - Automatic switch only when rate-limited for > 2 minutes on the current model
 - Session ID derived from first user message hash for cache continuity
 - Account state persisted to `~/.config/antigravity-proxy/accounts.json`
 
+**Account Selection Strategies:**
+
+1. **Sticky Strategy** (best for caching):
+   - Stays on current account until rate-limited or unavailable
+   - Waits up to 2 minutes for short rate limits before switching
+   - Maintains prompt cache continuity across requests
+
+2. **Round-Robin Strategy** (best for throughput):
+   - Rotates to next account on every request
+   - Skips rate-limited/disabled accounts
+   - Maximizes concurrent request distribution
+
+3. **Hybrid Strategy** (default, smart distribution):
+   - Uses health scores, token buckets, and LRU for selection
+   - Scoring formula: `score = (Health × 2) + ((Tokens / MaxTokens × 100) × 5) + (LRU × 0.1)`
+   - Health scores: Track success/failure patterns with passive recovery
+   - Token buckets: Client-side rate limiting (50 tokens, 6 per minute regeneration)
+   - LRU freshness: Prefer accounts that have rested longer
+   - Configuration in `src/config.js` under `accountSelection`
+
 **Account Data Model:**
 Each account object in `accounts.json` contains:
 - **Basic Info**: `email`, `source` (oauth/manual/database), `enabled`, `lastUsed`
diff --git a/README.md b/README.md
index 337923d..9bc2825 100644
--- a/README.md
+++ b/README.md
@@ -272,13 +272,37 @@ Gemini models include full thinking support with `thoughtSignature` handling for
 
 ## Multi-Account Load Balancing
 
-When you add multiple accounts, the proxy automatically:
+When you add multiple accounts, the proxy intelligently distributes requests across them using configurable selection strategies.
 
-- **Sticky account selection**: Stays on the same account to maximize prompt cache hits
-- **Smart rate limit handling**: Waits for short rate limits (≤2 min), switches accounts for longer ones
-- **Automatic cooldown**: Rate-limited accounts become available after reset time expires
-- **Invalid account detection**: Accounts needing re-authentication are marked and skipped
-- **Prompt caching support**: Stable session IDs enable cache hits across conversation turns
+### Account Selection Strategies
+
+Choose a strategy based on your needs:
+
+| Strategy | Best For | Description |
+| --- | --- | --- |
+| **Hybrid** (Default) | Most users | Smart selection combining health score, token bucket rate limiting, and LRU freshness |
+| **Sticky** | Prompt caching | Stays on the same account to maximize cache hits, switches only when rate-limited |
+| **Round-Robin** | Even distribution | Cycles through accounts sequentially for balanced load |
+
+**Configure via CLI:**
+
+```bash
+antigravity-claude-proxy start --strategy=hybrid    # Default: smart distribution
+antigravity-claude-proxy start --strategy=sticky    # Cache-optimized
+antigravity-claude-proxy start --strategy=round-robin  # Load-balanced
+```
+
+**Or via WebUI:** Settings → Server → Account Selection Strategy
+
+### How It Works
+
+- **Health Score Tracking**: Accounts earn points for successful requests and lose points for failures/rate-limits
+- **Token Bucket Rate Limiting**: Client-side throttling with regenerating tokens (50 max, 6/minute)
+- **Automatic Cooldown**: Rate-limited accounts recover automatically after reset time expires
+- **Invalid Account Detection**: Accounts needing re-authentication are marked and skipped
+- **Prompt Caching Support**: Session IDs derived from conversation enable cache hits across turns
+
+### Monitoring
 
 Check account status, subscription tiers, and quota anytime:
 
@@ -395,6 +419,7 @@ npm run test:streaming     # Streaming SSE events
 npm run test:interleaved   # Interleaved thinking
 npm run test:images        # Image processing
 npm run test:caching       # Prompt caching
+npm run test:strategies    # Account selection strategies
 ```
 
 ---
diff --git a/package.json b/package.json
index dec5b4f..d32b26a 100644
--- a/package.json
+++ b/package.json
@@ -34,7 +34,8 @@
     "test:crossmodel": "node tests/test-cross-model-thinking.cjs",
     "test:oauth": "node tests/test-oauth-no-browser.cjs",
     "test:emptyretry": "node tests/test-empty-response-retry.cjs",
-    "test:sanitizer": "node tests/test-schema-sanitizer.cjs"
+    "test:sanitizer": "node tests/test-schema-sanitizer.cjs",
+    "test:strategies": "node tests/test-strategies.cjs"
   },
   "keywords": [
     "claude",
diff --git a/public/js/components/server-config.js b/public/js/components/server-config.js
index e5ca4ed..6b139e2 100644
--- a/public/js/components/server-config.js
+++ b/public/js/components/server-config.js
@@ -248,5 +248,102 @@ window.Components.serverConfig = () => ({
         const { MAX_WAIT_MIN, MAX_WAIT_MAX } = window.AppConstants.VALIDATION;
         this.saveConfigField('maxWaitBeforeErrorMs', value, 'Max Wait Threshold',
             (v) => window.Validators.validateTimeout(v, MAX_WAIT_MIN, MAX_WAIT_MAX));
+    },
+
+    toggleRateLimitDedupWindowMs(value) {
+        const { RATE_LIMIT_DEDUP_MIN, RATE_LIMIT_DEDUP_MAX } = window.AppConstants.VALIDATION;
+        this.saveConfigField('rateLimitDedupWindowMs', value, 'Rate Limit Dedup Window',
+            (v) => window.Validators.validateTimeout(v, RATE_LIMIT_DEDUP_MIN, RATE_LIMIT_DEDUP_MAX));
+    },
+
+    toggleMaxConsecutiveFailures(value) {
+        const { MAX_CONSECUTIVE_FAILURES_MIN, MAX_CONSECUTIVE_FAILURES_MAX } = window.AppConstants.VALIDATION;
+        this.saveConfigField('maxConsecutiveFailures', value, 'Max Consecutive Failures',
+            (v) => window.Validators.validateRange(v, MAX_CONSECUTIVE_FAILURES_MIN, MAX_CONSECUTIVE_FAILURES_MAX, 'Max Consecutive Failures'));
+    },
+
+    toggleExtendedCooldownMs(value) {
+        const { EXTENDED_COOLDOWN_MIN, EXTENDED_COOLDOWN_MAX } = window.AppConstants.VALIDATION;
+        this.saveConfigField('extendedCooldownMs', value, 'Extended Cooldown',
+            (v) => window.Validators.validateTimeout(v, EXTENDED_COOLDOWN_MIN, EXTENDED_COOLDOWN_MAX));
+    },
+
+    toggleCapacityRetryDelayMs(value) {
+        const { CAPACITY_RETRY_DELAY_MIN, CAPACITY_RETRY_DELAY_MAX } = window.AppConstants.VALIDATION;
+        this.saveConfigField('capacityRetryDelayMs', value, 'Capacity Retry Delay',
+            (v) => window.Validators.validateTimeout(v, CAPACITY_RETRY_DELAY_MIN, CAPACITY_RETRY_DELAY_MAX));
+    },
+
+    toggleMaxCapacityRetries(value) {
+        const { MAX_CAPACITY_RETRIES_MIN, MAX_CAPACITY_RETRIES_MAX } = window.AppConstants.VALIDATION;
+        this.saveConfigField('maxCapacityRetries', value, 'Max Capacity Retries',
+            (v) => window.Validators.validateRange(v, MAX_CAPACITY_RETRIES_MIN, MAX_CAPACITY_RETRIES_MAX, 'Max Capacity Retries'));
+    },
+
+    // Toggle Account Selection Strategy
+    async toggleStrategy(strategy) {
+        const store = Alpine.store('global');
+        const validStrategies = ['sticky', 'round-robin', 'hybrid'];
+
+        if (!validStrategies.includes(strategy)) {
+            store.showToast(store.t('invalidStrategy'), 'error');
+            return;
+        }
+
+        // Optimistic update
+        const previousValue = this.serverConfig.accountSelection?.strategy || 'hybrid';
+        if (!this.serverConfig.accountSelection) {
+            this.serverConfig.accountSelection = {};
+        }
+        this.serverConfig.accountSelection.strategy = strategy;
+
+        try {
+            const { response, newPassword } = await window.utils.request('/api/config', {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({ accountSelection: { strategy } })
+            }, store.webuiPassword);
+
+            if (newPassword) store.webuiPassword = newPassword;
+
+            const data = await response.json();
+            if (data.status === 'ok') {
+                const strategyLabel = this.getStrategyLabel(strategy);
+                store.showToast(store.t('strategyUpdated', { strategy: strategyLabel }), 'success');
+                await this.fetchServerConfig(); // Confirm server state
+            } else {
+                throw new Error(data.error || store.t('failedToUpdateStrategy'));
+            }
+        } catch (e) {
+            // Rollback on error
+            if (!this.serverConfig.accountSelection) {
+                this.serverConfig.accountSelection = {};
+            }
+            this.serverConfig.accountSelection.strategy = previousValue;
+            store.showToast(store.t('failedToUpdateStrategy') + ': ' + e.message, 'error');
+        }
+    },
+
+    // Get display label for a strategy
+    getStrategyLabel(strategy) {
+        const store = Alpine.store('global');
+        const labels = {
+            'sticky': store.t('strategyStickyLabel'),
+            'round-robin': store.t('strategyRoundRobinLabel'),
+            'hybrid': store.t('strategyHybridLabel')
+        };
+        return labels[strategy] || strategy;
+    },
+
+    // Get description for current strategy
+    currentStrategyDescription() {
+        const store = Alpine.store('global');
+        const strategy = this.serverConfig.accountSelection?.strategy || 'hybrid';
+        const descriptions = {
+            'sticky': store.t('strategyStickyDesc'),
+            'round-robin': store.t('strategyRoundRobinDesc'),
+            'hybrid': store.t('strategyHybridDesc')
+        };
+        return descriptions[strategy] || '';
     }
 });
diff --git a/public/js/config/constants.js b/public/js/config/constants.js
index a5a1b6c..7074faf 100644
--- a/public/js/config/constants.js
+++ b/public/js/config/constants.js
@@ -67,7 +67,27 @@ window.AppConstants.VALIDATION = {
 
     // Max wait threshold (1 - 30 minutes)
     MAX_WAIT_MIN: 60000,
-    MAX_WAIT_MAX: 1800000
+    MAX_WAIT_MAX: 1800000,
+
+    // Rate limit dedup window (1 - 30 seconds)
+    RATE_LIMIT_DEDUP_MIN: 1000,
+    RATE_LIMIT_DEDUP_MAX: 30000,
+
+    // Consecutive failures (1 - 10)
+    MAX_CONSECUTIVE_FAILURES_MIN: 1,
+    MAX_CONSECUTIVE_FAILURES_MAX: 10,
+
+    // Extended cooldown (10 seconds - 5 minutes)
+    EXTENDED_COOLDOWN_MIN: 10000,
+    EXTENDED_COOLDOWN_MAX: 300000,
+
+    // Capacity retry delay (500ms - 10 seconds)
+    CAPACITY_RETRY_DELAY_MIN: 500,
+    CAPACITY_RETRY_DELAY_MAX: 10000,
+
+    // Capacity retries (1 - 10)
+    MAX_CAPACITY_RETRIES_MIN: 1,
+    MAX_CAPACITY_RETRIES_MAX: 10
 };
 
 /**
diff --git a/public/js/translations/en.js b/public/js/translations/en.js
index 61942ca..5157122 100644
--- a/public/js/translations/en.js
+++ b/public/js/translations/en.js
@@ -237,6 +237,18 @@ window.translations.en = {
     defaultCooldownDesc: "Fallback cooldown when API doesn't provide a reset time.",
     maxWaitThreshold: "Max Wait Before Error",
     maxWaitDesc: "If all accounts are rate-limited longer than this, error immediately instead of waiting.",
+    // Error Handling Tuning
+    errorHandlingTuning: "Error Handling Tuning",
+    rateLimitDedupWindow: "Rate Limit Dedup Window",
+    rateLimitDedupWindowDesc: "Prevents concurrent retry storms when multiple requests hit rate limits simultaneously.",
+    maxConsecutiveFailures: "Max Consecutive Failures",
+    maxConsecutiveFailuresDesc: "Number of consecutive failures before applying extended cooldown to an account.",
+    extendedCooldown: "Extended Cooldown",
+    extendedCooldownDesc: "Cooldown duration applied after max consecutive failures reached.",
+    capacityRetryDelay: "Capacity Retry Delay",
+    capacityRetryDelayDesc: "Delay before retrying when model capacity is exhausted (not quota).",
+    maxCapacityRetries: "Max Capacity Retries",
+    maxCapacityRetriesDesc: "Maximum retries for capacity exhaustion before switching accounts.",
     saveConfigServer: "Save Configuration",
     serverRestartAlert: "Changes saved to {path}. Restart server to apply some settings.",
     changePassword: "Change WebUI Password",
@@ -318,6 +330,18 @@ window.translations.en = {
     failedToUpdateModelConfig: "Failed to update model config",
     fieldUpdated: "{displayName} updated to {value}",
     failedToUpdateField: "Failed to update {displayName}",
+    // Account Selection Strategy
+    accountSelectionStrategy: "Account Selection Strategy",
+    selectionStrategy: "Selection Strategy",
+    strategyStickyLabel: "Sticky (Cache Optimized)",
+    strategyRoundRobinLabel: "Round Robin (Load Balanced)",
+    strategyHybridLabel: "Hybrid (Smart Distribution)",
+    strategyStickyDesc: "Stays on same account until rate-limited. Best for prompt caching.",
+    strategyRoundRobinDesc: "Rotates to next account on every request. Maximum throughput.",
+    strategyHybridDesc: "Smart selection based on health, tokens, and freshness.",
+    strategyUpdated: "Strategy updated to: {strategy}",
+    failedToUpdateStrategy: "Failed to update strategy",
+    invalidStrategy: "Invalid strategy selected",
     // Validation Messages
     mustBeValidNumber: "{fieldName} must be a valid number",
     mustBeAtLeast: "{fieldName} must be at least {min}",
diff --git a/public/js/translations/id.js b/public/js/translations/id.js
index 6e53bb1..917bc06 100644
--- a/public/js/translations/id.js
+++ b/public/js/translations/id.js
@@ -270,6 +270,18 @@ window.translations.id = {
     defaultCooldownDesc: "Cooldown bawaan jika API tidak memberikan waktu reset.",
     maxWaitThreshold: "Batas Tunggu Maksimal",
     maxWaitDesc: "Jika semua akun terkena rate limit lebih lama dari ini, langsung gagal.",
+    // Error Handling Tuning
+    errorHandlingTuning: "Penyetelan Penanganan Error",
+    rateLimitDedupWindow: "Jendela Deduplikasi Rate Limit",
+    rateLimitDedupWindowDesc: "Mencegah badai retry ketika beberapa permintaan terkena rate limit bersamaan.",
+    maxConsecutiveFailures: "Maks. Kegagalan Berturut-turut",
+    maxConsecutiveFailuresDesc: "Jumlah kegagalan berturut-turut sebelum menerapkan cooldown diperpanjang.",
+    extendedCooldown: "Cooldown Diperpanjang",
+    extendedCooldownDesc: "Durasi cooldown setelah mencapai maks. kegagalan berturut-turut.",
+    capacityRetryDelay: "Jeda Retry Kapasitas",
+    capacityRetryDelayDesc: "Jeda sebelum retry saat kapasitas model habis (bukan kuota).",
+    maxCapacityRetries: "Maks. Retry Kapasitas",
+    maxCapacityRetriesDesc: "Maksimum retry untuk kehabisan kapasitas sebelum ganti akun.",
     saveConfigServer: "Simpan Konfigurasi",
     serverRestartAlert: "Tersimpan ke {path}. Restart server untuk menerapkan.",
 
@@ -368,4 +380,17 @@ window.translations.id = {
     mustBeAtMost: "{fieldName} maksimal {max}",
     cannotBeEmpty: "{fieldName} tidak boleh kosong",
     mustBeTrueOrFalse: "Nilai harus true atau false",
+
+    // Account Selection Strategy translations
+    accountSelectionStrategy: "Strategi Pemilihan Akun",
+    selectionStrategy: "Strategi Pemilihan",
+    strategyStickyLabel: "Tetap (Optimisasi Cache)",
+    strategyRoundRobinLabel: "Bergilir (Load Balanced)",
+    strategyHybridLabel: "Hibrida (Distribusi Cerdas)",
+    strategyStickyDesc: "Tetap di akun yang sama hingga terkena rate limit. Terbaik untuk cache prompt.",
+    strategyRoundRobinDesc: "Berputar ke akun berikutnya setiap permintaan. Throughput maksimum.",
+    strategyHybridDesc: "Pemilihan cerdas berdasarkan kesehatan, token, dan kesegaran.",
+    strategyUpdated: "Strategi diubah ke: {strategy}",
+    failedToUpdateStrategy: "Gagal memperbarui strategi",
+    invalidStrategy: "Strategi tidak valid dipilih",
 };
diff --git a/public/js/translations/pt.js b/public/js/translations/pt.js
index 7c6fb87..468d35c 100644
--- a/public/js/translations/pt.js
+++ b/public/js/translations/pt.js
@@ -212,8 +212,21 @@ window.translations.pt = {
     persistTokenDesc: "Salvar sessões OAuth no disco para reinicializações mais rápidas",
     rateLimiting: "Limitação de Taxa de Conta & Timeouts",
     defaultCooldown: "Tempo de Resfriamento Padrão",
+    defaultCooldownDesc: "Resfriamento de fallback quando a API não fornece tempo de reset.",
     maxWaitThreshold: "Limiar Máximo de Espera (Sticky)",
     maxWaitDesc: "Tempo máximo para aguardar uma conta sticky resetar antes de trocar.",
+    // Ajuste de Tratamento de Erros
+    errorHandlingTuning: "Ajuste de Tratamento de Erros",
+    rateLimitDedupWindow: "Janela de Deduplicação de Rate Limit",
+    rateLimitDedupWindowDesc: "Previne tempestades de retry quando múltiplas requisições atingem rate limits simultaneamente.",
+    maxConsecutiveFailures: "Máx. Falhas Consecutivas",
+    maxConsecutiveFailuresDesc: "Número de falhas consecutivas antes de aplicar resfriamento estendido.",
+    extendedCooldown: "Resfriamento Estendido",
+    extendedCooldownDesc: "Duração do resfriamento aplicado após atingir máx. de falhas consecutivas.",
+    capacityRetryDelay: "Atraso de Retry de Capacidade",
+    capacityRetryDelayDesc: "Atraso antes de tentar novamente quando capacidade do modelo está esgotada (não quota).",
+    maxCapacityRetries: "Máx. Retries de Capacidade",
+    maxCapacityRetriesDesc: "Máximo de retries para esgotamento de capacidade antes de trocar conta.",
     saveConfigServer: "Salvar Configuração",
     serverRestartAlert: "Alterações salvas em {path}. Reinicie o servidor para aplicar algumas configurações.",
     changePassword: "Alterar Senha da WebUI",
@@ -258,4 +271,17 @@ window.translations.pt = {
     gemini1mDesc: "Adiciona sufixo [1m] aos modelos Gemini para suporte a janela de contexto de 1M.",
     gemini1mWarning: "⚠ Contexto grande pode reduzir o desempenho do Gemini-3-Pro.",
     clickToSet: "Clique para configurar...",
+
+    // Account Selection Strategy translations
+    accountSelectionStrategy: "Estratégia de Seleção de Conta",
+    selectionStrategy: "Estratégia de Seleção",
+    strategyStickyLabel: "Fixo (Otimizado para Cache)",
+    strategyRoundRobinLabel: "Rodízio (Balanceamento de Carga)",
+    strategyHybridLabel: "Híbrido (Distribuição Inteligente)",
+    strategyStickyDesc: "Permanece na mesma conta até atingir limite. Melhor para cache de prompts.",
+    strategyRoundRobinDesc: "Alterna para próxima conta a cada requisição. Máximo throughput.",
+    strategyHybridDesc: "Seleção inteligente baseada em saúde, tokens e frescor.",
+    strategyUpdated: "Estratégia atualizada para: {strategy}",
+    failedToUpdateStrategy: "Falha ao atualizar estratégia",
+    invalidStrategy: "Estratégia inválida selecionada",
 };
diff --git a/public/js/translations/tr.js b/public/js/translations/tr.js
index 746b083..ca68977 100644
--- a/public/js/translations/tr.js
+++ b/public/js/translations/tr.js
@@ -216,8 +216,21 @@ window.translations.tr = {
     persistTokenDesc: "Daha hızlı yeniden başlatmalar için OAuth oturumlarını diske kaydet",
     rateLimiting: "Hesap Hız Sınırlama ve Zaman Aşımları",
     defaultCooldown: "Varsayılan Soğuma Süresi",
+    defaultCooldownDesc: "API sıfırlama zamanı sağlamadığında yedek soğuma süresi.",
     maxWaitThreshold: "Maksimum Bekleme Eşiği (Yapışkan)",
     maxWaitDesc: "Yapışkan bir hesabın değiştirmeden önce sıfırlanması için beklenecek maksimum süre.",
+    // Hata İşleme Ayarları
+    errorHandlingTuning: "Hata İşleme Ayarları",
+    rateLimitDedupWindow: "Hız Sınırı Tekilleştirme Penceresi",
+    rateLimitDedupWindowDesc: "Birden fazla istek aynı anda hız sınırına ulaştığında yeniden deneme fırtınasını önler.",
+    maxConsecutiveFailures: "Maks. Ardışık Başarısızlık",
+    maxConsecutiveFailuresDesc: "Uzatılmış soğuma uygulamadan önce ardışık başarısızlık sayısı.",
+    extendedCooldown: "Uzatılmış Soğuma",
+    extendedCooldownDesc: "Maks. ardışık başarısızlık sonrası uygulanan soğuma süresi.",
+    capacityRetryDelay: "Kapasite Yeniden Deneme Gecikmesi",
+    capacityRetryDelayDesc: "Model kapasitesi tükendiğinde (kota değil) yeniden denemeden önceki gecikme.",
+    maxCapacityRetries: "Maks. Kapasite Yeniden Denemesi",
+    maxCapacityRetriesDesc: "Hesap değiştirmeden önce kapasite tükenmesi için maksimum yeniden deneme.",
     saveConfigServer: "Yapılandırmayı Kaydet",
     serverRestartAlert: "Değişiklikler {path} konumuna kaydedildi. Bazı ayarları uygulamak için sunucuyu yeniden başlatın.",
     changePassword: "WebUI Parolasını Değiştir",
@@ -313,4 +326,17 @@ window.translations.tr = {
 
     // TODO: Missing translation - Server config (exists in EN but missing here)
     // defaultCooldownDesc: "Fallback cooldown when API doesn't provide a reset time.",
+
+    // Account Selection Strategy translations
+    accountSelectionStrategy: "Hesap Seçim Stratejisi",
+    selectionStrategy: "Seçim Stratejisi",
+    strategyStickyLabel: "Sabit (Önbellek Optimizasyonu)",
+    strategyRoundRobinLabel: "Döngüsel (Yük Dengeleme)",
+    strategyHybridLabel: "Hibrit (Akıllı Dağıtım)",
+    strategyStickyDesc: "Hız sınırına ulaşılana kadar aynı hesapta kalır. Önbellek için en iyisi.",
+    strategyRoundRobinDesc: "Her istekte bir sonraki hesaba geçer. Maksimum verimlilik.",
+    strategyHybridDesc: "Sağlık, token ve tazeliğe dayalı akıllı seçim.",
+    strategyUpdated: "Strateji şu şekilde güncellendi: {strategy}",
+    failedToUpdateStrategy: "Strateji güncellenemedi",
+    invalidStrategy: "Geçersiz strateji seçildi",
 };
diff --git a/public/js/translations/zh.js b/public/js/translations/zh.js
index e441c22..a1e7ef2 100644
--- a/public/js/translations/zh.js
+++ b/public/js/translations/zh.js
@@ -237,6 +237,18 @@ window.translations.zh = {
     defaultCooldownDesc: "当 API 未提供重置时间时的备用冷却时间。",
     maxWaitThreshold: "最大等待阈值",
     maxWaitDesc: "如果所有账号的限流时间超过此阈值，立即返回错误而非等待。",
+    // 错误处理调优
+    errorHandlingTuning: "错误处理调优",
+    rateLimitDedupWindow: "限流去重窗口",
+    rateLimitDedupWindowDesc: "当多个请求同时触发限流时，防止并发重试风暴。",
+    maxConsecutiveFailures: "最大连续失败次数",
+    maxConsecutiveFailuresDesc: "触发扩展冷却前允许的连续失败次数。",
+    extendedCooldown: "扩展冷却时间",
+    extendedCooldownDesc: "达到最大连续失败后应用的冷却时长。",
+    capacityRetryDelay: "容量重试延迟",
+    capacityRetryDelayDesc: "模型容量耗尽（非配额）时重试前的延迟。",
+    maxCapacityRetries: "最大容量重试次数",
+    maxCapacityRetriesDesc: "容量耗尽时在切换账号前的最大重试次数。",
     saveConfigServer: "保存配置",
     serverRestartAlert: "配置已保存至 {path}。部分更改可能需要重启服务器。",
     changePassword: "修改 WebUI 密码",
@@ -329,4 +341,17 @@ window.translations.zh = {
     // mustBeAtMost: "{fieldName} must be at most {max}",
     // cannotBeEmpty: "{fieldName} cannot be empty",
     // mustBeTrueOrFalse: "Value must be true or false",
+
+    // Account Selection Strategy translations
+    accountSelectionStrategy: "账户选择策略",
+    selectionStrategy: "选择策略",
+    strategyStickyLabel: "固定 (缓存优化)",
+    strategyRoundRobinLabel: "轮询 (负载均衡)",
+    strategyHybridLabel: "混合 (智能分配)",
+    strategyStickyDesc: "保持使用同一账户直到被限速。最适合提示词缓存。",
+    strategyRoundRobinDesc: "每次请求轮换到下一个账户。最大吞吐量。",
+    strategyHybridDesc: "基于健康度、令牌和新鲜度的智能选择。",
+    strategyUpdated: "策略已更新为: {strategy}",
+    failedToUpdateStrategy: "更新策略失败",
+    invalidStrategy: "选择了无效的策略",
 };
diff --git a/public/views/settings.html b/public/views/settings.html
index 169a1bf..c14f5de 100644
--- a/public/views/settings.html
+++ b/public/views/settings.html
@@ -936,6 +936,35 @@
                     </div>
                 </div>
 
+                <!-- 🔀 Account Selection Strategy -->
+                <div class="space-y-3">
+                    <div class="flex items-center gap-2 mb-1 px-1">
+                        <span class="text-[10px] uppercase text-gray-500 font-bold tracking-widest"
+                            x-text="$store.global.t('accountSelectionStrategy')">Account Selection Strategy</span>
+                        <div class="h-px flex-1 bg-space-border/30"></div>
+                    </div>
+
+                    <div class="form-control view-card border-space-border/50 hover:border-neon-cyan/50">
+                        <div class="flex items-center justify-between">
+                            <div class="flex flex-col gap-1 flex-1">
+                                <span class="text-sm font-medium text-gray-200"
+                                    x-text="$store.global.t('selectionStrategy')">Selection Strategy</span>
+                                <span class="text-[11px] text-gray-500"
+                                    x-text="currentStrategyDescription()">How accounts are selected for requests</span>
+                            </div>
+                            <select
+                                class="select bg-space-800 border-space-border text-gray-200 focus:border-neon-cyan focus:ring-neon-cyan/20 w-64"
+                                :value="serverConfig.accountSelection?.strategy || 'hybrid'"
+                                @change="toggleStrategy($el.value)"
+                                aria-label="Account selection strategy">
+                                <option value="hybrid" x-text="$store.global.t('strategyHybridLabel')">Hybrid (Smart Distribution)</option>
+                                <option value="sticky" x-text="$store.global.t('strategyStickyLabel')">Sticky (Cache Optimized)</option>
+                                <option value="round-robin" x-text="$store.global.t('strategyRoundRobinLabel')">Round Robin (Load Balanced)</option>
+                            </select>
+                        </div>
+                    </div>
+                </div>
+
                 <!-- ▼ Advanced Tuning (Fixed Logic) -->
                 <div class="view-card !p-0 border-space-border/50">
                     <div class="flex items-center justify-between p-4 cursor-pointer hover:bg-white/5 transition-colors"
@@ -1095,6 +1124,134 @@
                                     x-text="$store.global.t('maxWaitDesc')">If all accounts are rate-limited longer than this, error immediately.</p>
                             </div>
                         </div>
+
+                        <!-- Error Handling Tuning -->
+                        <div class="space-y-4 pt-2 border-t border-space-border/10">
+                            <div class="flex items-center gap-2 mb-2">
+                                <span class="text-[10px] text-gray-500 font-bold uppercase tracking-widest"
+                                    x-text="$store.global.t('errorHandlingTuning')">Error Handling Tuning</span>
+                            </div>
+
+                            <div class="form-control">
+                                <label class="label pt-0">
+                                    <span class="label-text text-gray-400 text-xs"
+                                        x-text="$store.global.t('rateLimitDedupWindow')">Rate Limit Dedup Window</span>
+                                    <span class="label-text-alt font-mono text-neon-cyan text-xs font-semibold"
+                                        x-text="Math.round((serverConfig.rateLimitDedupWindowMs || 5000) / 1000) + 's'"></span>
+                                </label>
+                                <div class="flex gap-3 items-center">
+                                    <input type="range" min="1000" max="30000" step="1000"
+                                        class="custom-range custom-range-cyan flex-1"
+                                        :value="serverConfig.rateLimitDedupWindowMs || 5000"
+                                        :style="`background-size: ${((serverConfig.rateLimitDedupWindowMs || 5000) - 1000) / 290}% 100%`"
+                                        @input="toggleRateLimitDedupWindowMs($event.target.value)"
+                                        aria-label="Rate limit dedup window slider">
+                                    <input type="number" min="1000" max="30000" step="1000"
+                                        class="input input-xs input-bordered w-20 bg-space-800 border-space-border text-white font-mono text-center"
+                                        :value="serverConfig.rateLimitDedupWindowMs || 5000"
+                                        @change="toggleRateLimitDedupWindowMs($event.target.value)"
+                                        aria-label="Rate limit dedup window value">
+                                </div>
+                                <p class="text-[9px] text-gray-600 mt-1 leading-tight"
+                                    x-text="$store.global.t('rateLimitDedupWindowDesc')">Prevents concurrent retry storms.</p>
+                            </div>
+
+                            <div class="form-control">
+                                <label class="label pt-0">
+                                    <span class="label-text text-gray-400 text-xs"
+                                        x-text="$store.global.t('maxConsecutiveFailures')">Max Consecutive Failures</span>
+                                    <span class="label-text-alt font-mono text-neon-cyan text-xs font-semibold"
+                                        x-text="serverConfig.maxConsecutiveFailures || 3"></span>
+                                </label>
+                                <div class="flex gap-3 items-center">
+                                    <input type="range" min="1" max="10" step="1"
+                                        class="custom-range custom-range-cyan flex-1"
+                                        :value="serverConfig.maxConsecutiveFailures || 3"
+                                        :style="`background-size: ${((serverConfig.maxConsecutiveFailures || 3) - 1) / 0.09}% 100%`"
+                                        @input="toggleMaxConsecutiveFailures($event.target.value)"
+                                        aria-label="Max consecutive failures slider">
+                                    <input type="number" min="1" max="10" step="1"
+                                        class="input input-xs input-bordered w-16 bg-space-800 border-space-border text-white font-mono text-center"
+                                        :value="serverConfig.maxConsecutiveFailures || 3"
+                                        @change="toggleMaxConsecutiveFailures($event.target.value)"
+                                        aria-label="Max consecutive failures value">
+                                </div>
+                                <p class="text-[9px] text-gray-600 mt-1 leading-tight"
+                                    x-text="$store.global.t('maxConsecutiveFailuresDesc')">Failures before extended cooldown.</p>
+                            </div>
+
+                            <div class="form-control">
+                                <label class="label pt-0">
+                                    <span class="label-text text-gray-400 text-xs"
+                                        x-text="$store.global.t('extendedCooldown')">Extended Cooldown</span>
+                                    <span class="label-text-alt font-mono text-neon-cyan text-xs font-semibold"
+                                        x-text="((serverConfig.extendedCooldownMs || 60000) >= 60000 ? Math.round((serverConfig.extendedCooldownMs || 60000) / 60000) + 'm' : Math.round((serverConfig.extendedCooldownMs || 60000) / 1000) + 's')"></span>
+                                </label>
+                                <div class="flex gap-3 items-center">
+                                    <input type="range" min="10000" max="300000" step="10000"
+                                        class="custom-range custom-range-cyan flex-1"
+                                        :value="serverConfig.extendedCooldownMs || 60000"
+                                        :style="`background-size: ${((serverConfig.extendedCooldownMs || 60000) - 10000) / 2900}% 100%`"
+                                        @input="toggleExtendedCooldownMs($event.target.value)"
+                                        aria-label="Extended cooldown slider">
+                                    <input type="number" min="10000" max="300000" step="10000"
+                                        class="input input-xs input-bordered w-24 bg-space-800 border-space-border text-white font-mono text-center"
+                                        :value="serverConfig.extendedCooldownMs || 60000"
+                                        @change="toggleExtendedCooldownMs($event.target.value)"
+                                        aria-label="Extended cooldown value">
+                                </div>
+                                <p class="text-[9px] text-gray-600 mt-1 leading-tight"
+                                    x-text="$store.global.t('extendedCooldownDesc')">Applied after max consecutive failures.</p>
+                            </div>
+
+                            <div class="form-control">
+                                <label class="label pt-0">
+                                    <span class="label-text text-gray-400 text-xs"
+                                        x-text="$store.global.t('capacityRetryDelay')">Capacity Retry Delay</span>
+                                    <span class="label-text-alt font-mono text-neon-cyan text-xs font-semibold"
+                                        x-text="Math.round((serverConfig.capacityRetryDelayMs || 2000) / 1000) + 's'"></span>
+                                </label>
+                                <div class="flex gap-3 items-center">
+                                    <input type="range" min="500" max="10000" step="500"
+                                        class="custom-range custom-range-cyan flex-1"
+                                        :value="serverConfig.capacityRetryDelayMs || 2000"
+                                        :style="`background-size: ${((serverConfig.capacityRetryDelayMs || 2000) - 500) / 95}% 100%`"
+                                        @input="toggleCapacityRetryDelayMs($event.target.value)"
+                                        aria-label="Capacity retry delay slider">
+                                    <input type="number" min="500" max="10000" step="500"
+                                        class="input input-xs input-bordered w-20 bg-space-800 border-space-border text-white font-mono text-center"
+                                        :value="serverConfig.capacityRetryDelayMs || 2000"
+                                        @change="toggleCapacityRetryDelayMs($event.target.value)"
+                                        aria-label="Capacity retry delay value">
+                                </div>
+                                <p class="text-[9px] text-gray-600 mt-1 leading-tight"
+                                    x-text="$store.global.t('capacityRetryDelayDesc')">Delay for capacity (not quota) issues.</p>
+                            </div>
+
+                            <div class="form-control">
+                                <label class="label pt-0">
+                                    <span class="label-text text-gray-400 text-xs"
+                                        x-text="$store.global.t('maxCapacityRetries')">Max Capacity Retries</span>
+                                    <span class="label-text-alt font-mono text-neon-cyan text-xs font-semibold"
+                                        x-text="serverConfig.maxCapacityRetries || 3"></span>
+                                </label>
+                                <div class="flex gap-3 items-center">
+                                    <input type="range" min="1" max="10" step="1"
+                                        class="custom-range custom-range-cyan flex-1"
+                                        :value="serverConfig.maxCapacityRetries || 3"
+                                        :style="`background-size: ${((serverConfig.maxCapacityRetries || 3) - 1) / 0.09}% 100%`"
+                                        @input="toggleMaxCapacityRetries($event.target.value)"
+                                        aria-label="Max capacity retries slider">
+                                    <input type="number" min="1" max="10" step="1"
+                                        class="input input-xs input-bordered w-16 bg-space-800 border-space-border text-white font-mono text-center"
+                                        :value="serverConfig.maxCapacityRetries || 3"
+                                        @change="toggleMaxCapacityRetries($event.target.value)"
+                                        aria-label="Max capacity retries value">
+                                </div>
+                                <p class="text-[9px] text-gray-600 mt-1 leading-tight"
+                                    x-text="$store.global.t('maxCapacityRetriesDesc')">Retries before switching accounts.</p>
+                            </div>
+                        </div>
                     </div>
                 </div>
 
diff --git a/src/account-manager/index.js b/src/account-manager/index.js
index 8db296d..8231344 100644
--- a/src/account-manager/index.js
+++ b/src/account-manager/index.js
@@ -1,6 +1,6 @@
 /**
  * Account Manager
- * Manages multiple Antigravity accounts with sticky selection,
+ * Manages multiple Antigravity accounts with configurable selection strategies,
  * automatic failover, and smart cooldown for rate-limited accounts.
  */
 
@@ -23,13 +23,9 @@ import {
     clearProjectCache as clearProject,
     clearTokenCache as clearToken
 } from './credentials.js';
-import {
-    pickNext as selectNext,
-    getCurrentStickyAccount as getSticky,
-    shouldWaitForCurrentAccount as shouldWait,
-    pickStickyAccount as selectSticky
-} from './selection.js';
+import { createStrategy, getStrategyLabel, DEFAULT_STRATEGY } from './strategies/index.js';
 import { logger } from '../utils/logger.js';
+import { config } from '../config.js';
 
 export class AccountManager {
     #accounts = [];
@@ -37,19 +33,26 @@ export class AccountManager {
     #configPath;
     #settings = {};
     #initialized = false;
+    #strategy = null;
+    #strategyName = DEFAULT_STRATEGY;
 
     // Per-account caches
     #tokenCache = new Map(); // email -> { token, extractedAt }
     #projectCache = new Map(); // email -> projectId
 
-    constructor(configPath = ACCOUNT_CONFIG_PATH) {
+    constructor(configPath = ACCOUNT_CONFIG_PATH, strategyName = null) {
         this.#configPath = configPath;
+        // Strategy name can be set at construction or later via initialize
+        if (strategyName) {
+            this.#strategyName = strategyName;
+        }
     }
 
     /**
      * Initialize the account manager by loading config
+     * @param {string} [strategyOverride] - Override strategy name (from CLI flag or env var)
      */
-    async initialize() {
+    async initialize(strategyOverride = null) {
         if (this.#initialized) return;
 
         const { accounts, settings, activeIndex } = await loadAccounts(this.#configPath);
@@ -66,6 +69,16 @@ export class AccountManager {
             this.#tokenCache = tokenCache;
         }
 
+        // Determine strategy: CLI override > env var > config file > default
+        const configStrategy = config?.accountSelection?.strategy;
+        const envStrategy = process.env.ACCOUNT_STRATEGY;
+        this.#strategyName = strategyOverride || envStrategy || configStrategy || this.#strategyName;
+
+        // Create the strategy instance
+        const strategyConfig = config?.accountSelection || {};
+        this.#strategy = createStrategy(this.#strategyName, strategyConfig);
+        logger.info(`[AccountManager] Using ${getStrategyLabel(this.#strategyName)} selection strategy`);
+
         // Clear any expired rate limits
         this.clearExpiredLimits();
 
@@ -138,51 +151,88 @@ export class AccountManager {
     }
 
     /**
-     * Pick the next available account (fallback when current is unavailable).
-     * Sets activeIndex to the selected account's index.
-     * @param {string} [modelId] - Optional model ID
-     * @returns {Object|null} The next available account or null if none available
-     */
-    pickNext(modelId = null) {
-        const { account, newIndex } = selectNext(this.#accounts, this.#currentIndex, () => this.saveToDisk(), modelId);
-        this.#currentIndex = newIndex;
-        return account;
-    }
-
-    /**
-     * Get the current account without advancing the index (sticky selection).
-     * Used for cache continuity - sticks to the same account until rate-limited.
-     * @param {string} [modelId] - Optional model ID
-     * @returns {Object|null} The current account or null if unavailable/rate-limited
-     */
-    getCurrentStickyAccount(modelId = null) {
-        const { account, newIndex } = getSticky(this.#accounts, this.#currentIndex, () => this.saveToDisk(), modelId);
-        this.#currentIndex = newIndex;
-        return account;
-    }
-
-    /**
-     * Check if we should wait for the current account's rate limit to reset.
-     * Used for sticky account selection - wait if rate limit is short (≤ threshold).
-     * @param {string} [modelId] - Optional model ID
-     * @returns {{shouldWait: boolean, waitMs: number, account: Object|null}}
-     */
-    shouldWaitForCurrentAccount(modelId = null) {
-        return shouldWait(this.#accounts, this.#currentIndex, modelId);
-    }
-
-    /**
-     * Pick an account with sticky selection preference.
-     * Prefers the current account for cache continuity, only switches when:
-     * - Current account is rate-limited for > 2 minutes
-     * - Current account is invalid
-     * @param {string} [modelId] - Optional model ID
+     * Select an account using the configured strategy.
+     * This is the main method to use for account selection.
+     * @param {string} [modelId] - Model ID for the request
+     * @param {Object} [options] - Additional options
+     * @param {string} [options.sessionId] - Session ID for cache continuity
      * @returns {{account: Object|null, waitMs: number}} Account to use and optional wait time
      */
-    pickStickyAccount(modelId = null) {
-        const { account, waitMs, newIndex } = selectSticky(this.#accounts, this.#currentIndex, () => this.saveToDisk(), modelId);
-        this.#currentIndex = newIndex;
-        return { account, waitMs };
+    selectAccount(modelId = null, options = {}) {
+        if (!this.#strategy) {
+            throw new Error('AccountManager not initialized. Call initialize() first.');
+        }
+
+        const result = this.#strategy.selectAccount(this.#accounts, modelId, {
+            currentIndex: this.#currentIndex,
+            onSave: () => this.saveToDisk(),
+            ...options
+        });
+
+        this.#currentIndex = result.index;
+        return { account: result.account, waitMs: result.waitMs || 0 };
+    }
+
+    /**
+     * Notify the strategy of a successful request
+     * @param {Object} account - The account that was used
+     * @param {string} modelId - The model ID that was used
+     */
+    notifySuccess(account, modelId) {
+        if (this.#strategy) {
+            this.#strategy.onSuccess(account, modelId);
+        }
+    }
+
+    /**
+     * Notify the strategy of a rate limit
+     * @param {Object} account - The account that was rate-limited
+     * @param {string} modelId - The model ID that was rate-limited
+     */
+    notifyRateLimit(account, modelId) {
+        if (this.#strategy) {
+            this.#strategy.onRateLimit(account, modelId);
+        }
+    }
+
+    /**
+     * Notify the strategy of a failure
+     * @param {Object} account - The account that failed
+     * @param {string} modelId - The model ID that failed
+     */
+    notifyFailure(account, modelId) {
+        if (this.#strategy) {
+            this.#strategy.onFailure(account, modelId);
+        }
+    }
+
+    /**
+     * Get the current strategy name
+     * @returns {string} Strategy name
+     */
+    getStrategyName() {
+        return this.#strategyName;
+    }
+
+    /**
+     * Get the strategy display label
+     * @returns {string} Strategy display label
+     */
+    getStrategyLabel() {
+        return getStrategyLabel(this.#strategyName);
+    }
+
+    /**
+     * Get the health tracker from the current strategy (if available)
+     * Used by handlers for consecutive failure tracking
+     * Only available when using hybrid strategy
+     * @returns {Object|null} Health tracker instance or null if not available
+     */
+    getHealthTracker() {
+        if (this.#strategy && typeof this.#strategy.getHealthTracker === 'function') {
+            return this.#strategy.getHealthTracker();
+        }
+        return null;
     }
 
     /**
diff --git a/src/account-manager/selection.js b/src/account-manager/selection.js
deleted file mode 100644
index ad41307..0000000
--- a/src/account-manager/selection.js
+++ /dev/null
@@ -1,201 +0,0 @@
-/**
- * Account Selection
- *
- * Handles account picking logic (round-robin, sticky) for cache continuity.
- * All rate limit checks are model-specific.
- */
-
-import { MAX_WAIT_BEFORE_ERROR_MS } from '../constants.js';
-import { formatDuration } from '../utils/helpers.js';
-import { logger } from '../utils/logger.js';
-import { clearExpiredLimits, getAvailableAccounts } from './rate-limits.js';
-
-/**
- * Check if an account is usable for a specific model
- * @param {Object} account - Account object
- * @param {string} modelId - Model ID to check
- * @returns {boolean} True if account is usable
- */
-function isAccountUsable(account, modelId) {
-    if (!account || account.isInvalid) return false;
-
-    // WebUI: Skip disabled accounts
-    if (account.enabled === false) return false;
-
-    if (modelId && account.modelRateLimits && account.modelRateLimits[modelId]) {
-        const limit = account.modelRateLimits[modelId];
-        if (limit.isRateLimited && limit.resetTime > Date.now()) {
-            return false;
-        }
-    }
-
-    return true;
-}
-
-/**
- * Pick the next available account (fallback when current is unavailable).
- *
- * @param {Array} accounts - Array of account objects
- * @param {number} currentIndex - Current account index
- * @param {Function} onSave - Callback to save changes
- * @param {string} [modelId] - Model ID to check rate limits for
- * @returns {{account: Object|null, newIndex: number}} The next available account and new index
- */
-export function pickNext(accounts, currentIndex, onSave, modelId = null) {
-    clearExpiredLimits(accounts);
-
-    const available = getAvailableAccounts(accounts, modelId);
-    if (available.length === 0) {
-        return { account: null, newIndex: currentIndex };
-    }
-
-    // Clamp index to valid range
-    let index = currentIndex;
-    if (index >= accounts.length) {
-        index = 0;
-    }
-
-    // Find next available account starting from index AFTER current
-    for (let i = 1; i <= accounts.length; i++) {
-        const idx = (index + i) % accounts.length;
-        const account = accounts[idx];
-
-        if (isAccountUsable(account, modelId)) {
-            account.lastUsed = Date.now();
-
-            const position = idx + 1;
-            const total = accounts.length;
-            logger.info(`[AccountManager] Using account: ${account.email} (${position}/${total})`);
-
-            // Trigger save (don't await to avoid blocking)
-            if (onSave) onSave();
-
-            return { account, newIndex: idx };
-        }
-    }
-
-    return { account: null, newIndex: currentIndex };
-}
-
-/**
- * Get the current account without advancing the index (sticky selection).
- *
- * @param {Array} accounts - Array of account objects
- * @param {number} currentIndex - Current account index
- * @param {Function} onSave - Callback to save changes
- * @param {string} [modelId] - Model ID to check rate limits for
- * @returns {{account: Object|null, newIndex: number}} The current account and index
- */
-export function getCurrentStickyAccount(accounts, currentIndex, onSave, modelId = null) {
-    clearExpiredLimits(accounts);
-
-    if (accounts.length === 0) {
-        return { account: null, newIndex: currentIndex };
-    }
-
-    // Clamp index to valid range
-    let index = currentIndex;
-    if (index >= accounts.length) {
-        index = 0;
-    }
-
-    // Get current account directly (activeIndex = current account)
-    const account = accounts[index];
-
-    if (isAccountUsable(account, modelId)) {
-        account.lastUsed = Date.now();
-        // Trigger save (don't await to avoid blocking)
-        if (onSave) onSave();
-        return { account, newIndex: index };
-    }
-
-    return { account: null, newIndex: index };
-}
-
-/**
- * Check if we should wait for the current account's rate limit to reset.
- *
- * @param {Array} accounts - Array of account objects
- * @param {number} currentIndex - Current account index
- * @param {string} [modelId] - Model ID to check rate limits for
- * @returns {{shouldWait: boolean, waitMs: number, account: Object|null}}
- */
-export function shouldWaitForCurrentAccount(accounts, currentIndex, modelId = null) {
-    if (accounts.length === 0) {
-        return { shouldWait: false, waitMs: 0, account: null };
-    }
-
-    // Clamp index to valid range
-    let index = currentIndex;
-    if (index >= accounts.length) {
-        index = 0;
-    }
-
-    // Get current account directly (activeIndex = current account)
-    const account = accounts[index];
-
-    if (!account || account.isInvalid) {
-        return { shouldWait: false, waitMs: 0, account: null };
-    }
-
-    let waitMs = 0;
-
-    // Check model-specific limit
-    if (modelId && account.modelRateLimits && account.modelRateLimits[modelId]) {
-        const limit = account.modelRateLimits[modelId];
-        if (limit.isRateLimited && limit.resetTime) {
-            waitMs = limit.resetTime - Date.now();
-        }
-    }
-
-    // If wait time is within threshold, recommend waiting
-    if (waitMs > 0 && waitMs <= MAX_WAIT_BEFORE_ERROR_MS) {
-        return { shouldWait: true, waitMs, account };
-    }
-
-    return { shouldWait: false, waitMs: 0, account };
-}
-
-/**
- * Pick an account with sticky selection preference.
- * Prefers the current account for cache continuity.
- *
- * @param {Array} accounts - Array of account objects
- * @param {number} currentIndex - Current account index
- * @param {Function} onSave - Callback to save changes
- * @param {string} [modelId] - Model ID to check rate limits for
- * @returns {{account: Object|null, waitMs: number, newIndex: number}}
- */
-export function pickStickyAccount(accounts, currentIndex, onSave, modelId = null) {
-    // First try to get the current sticky account
-    const { account: stickyAccount, newIndex: stickyIndex } = getCurrentStickyAccount(accounts, currentIndex, onSave, modelId);
-    if (stickyAccount) {
-        return { account: stickyAccount, waitMs: 0, newIndex: stickyIndex };
-    }
-
-    // Current account is rate-limited or invalid.
-    // CHECK IF OTHERS ARE AVAILABLE before deciding to wait.
-    const available = getAvailableAccounts(accounts, modelId);
-    if (available.length > 0) {
-        // Found a free account! Switch immediately.
-        const { account: nextAccount, newIndex } = pickNext(accounts, currentIndex, onSave, modelId);
-        if (nextAccount) {
-            logger.info(`[AccountManager] Switched to new account (failover): ${nextAccount.email}`);
-            return { account: nextAccount, waitMs: 0, newIndex };
-        }
-    }
-
-    // No other accounts available. Now checking if we should wait for current account.
-    const waitInfo = shouldWaitForCurrentAccount(accounts, currentIndex, modelId);
-    if (waitInfo.shouldWait) {
-        logger.info(`[AccountManager] Waiting ${formatDuration(waitInfo.waitMs)} for sticky account: ${waitInfo.account.email}`);
-        return { account: null, waitMs: waitInfo.waitMs, newIndex: currentIndex };
-    }
-
-    // Current account unavailable for too long/invalid, and no others available?
-    const { account: nextAccount, newIndex } = pickNext(accounts, currentIndex, onSave, modelId);
-    if (nextAccount) {
-        logger.info(`[AccountManager] Switched to new account for cache: ${nextAccount.email}`);
-    }
-    return { account: nextAccount, waitMs: 0, newIndex };
-}
diff --git a/src/account-manager/strategies/base-strategy.js b/src/account-manager/strategies/base-strategy.js
new file mode 100644
index 0000000..57007cc
--- /dev/null
+++ b/src/account-manager/strategies/base-strategy.js
@@ -0,0 +1,104 @@
+/**
+ * Base Strategy
+ *
+ * Abstract base class defining the interface for account selection strategies.
+ * All strategies must implement the selectAccount method.
+ */
+
+/**
+ * @typedef {Object} SelectionResult
+ * @property {Object|null} account - The selected account or null if none available
+ * @property {number} index - The index of the selected account
+ * @property {number} [waitMs] - Optional wait time before account becomes available
+ */
+
+export class BaseStrategy {
+    /**
+     * Create a new BaseStrategy
+     * @param {Object} config - Strategy configuration
+     */
+    constructor(config = {}) {
+        if (new.target === BaseStrategy) {
+            throw new Error('BaseStrategy is abstract and cannot be instantiated directly');
+        }
+        this.config = config;
+    }
+
+    /**
+     * Select an account for a request
+     * @param {Array} accounts - Array of account objects
+     * @param {string} modelId - The model ID for the request
+     * @param {Object} options - Additional options
+     * @param {number} options.currentIndex - Current account index
+     * @param {string} [options.sessionId] - Session ID for cache continuity
+     * @param {Function} [options.onSave] - Callback to save changes
+     * @returns {SelectionResult} The selected account and index
+     */
+    selectAccount(accounts, modelId, options = {}) {
+        throw new Error('selectAccount must be implemented by subclass');
+    }
+
+    /**
+     * Called after a successful request
+     * @param {Object} account - The account that was used
+     * @param {string} modelId - The model ID that was used
+     */
+    onSuccess(account, modelId) {
+        // Default: no-op, override in subclass if needed
+    }
+
+    /**
+     * Called when a request is rate-limited
+     * @param {Object} account - The account that was rate-limited
+     * @param {string} modelId - The model ID that was rate-limited
+     */
+    onRateLimit(account, modelId) {
+        // Default: no-op, override in subclass if needed
+    }
+
+    /**
+     * Called when a request fails (non-rate-limit error)
+     * @param {Object} account - The account that failed
+     * @param {string} modelId - The model ID that failed
+     */
+    onFailure(account, modelId) {
+        // Default: no-op, override in subclass if needed
+    }
+
+    /**
+     * Check if an account is usable for a specific model
+     * @param {Object} account - Account object
+     * @param {string} modelId - Model ID to check
+     * @returns {boolean} True if account is usable
+     */
+    isAccountUsable(account, modelId) {
+        if (!account || account.isInvalid) return false;
+
+        // Skip disabled accounts
+        if (account.enabled === false) return false;
+
+        // Check model-specific rate limit
+        if (modelId && account.modelRateLimits && account.modelRateLimits[modelId]) {
+            const limit = account.modelRateLimits[modelId];
+            if (limit.isRateLimited && limit.resetTime > Date.now()) {
+                return false;
+            }
+        }
+
+        return true;
+    }
+
+    /**
+     * Get all usable accounts for a model
+     * @param {Array} accounts - Array of account objects
+     * @param {string} modelId - Model ID to check
+     * @returns {Array} Array of usable accounts with their original indices
+     */
+    getUsableAccounts(accounts, modelId) {
+        return accounts
+            .map((account, index) => ({ account, index }))
+            .filter(({ account }) => this.isAccountUsable(account, modelId));
+    }
+}
+
+export default BaseStrategy;
diff --git a/src/account-manager/strategies/hybrid-strategy.js b/src/account-manager/strategies/hybrid-strategy.js
new file mode 100644
index 0000000..6814d7f
--- /dev/null
+++ b/src/account-manager/strategies/hybrid-strategy.js
@@ -0,0 +1,195 @@
+/**
+ * Hybrid Strategy
+ *
+ * Smart selection based on health score, token bucket, and LRU freshness.
+ * Combines multiple signals for optimal account distribution.
+ *
+ * Scoring formula:
+ *   score = (Health × 2) + ((Tokens / MaxTokens × 100) × 5) + (LRU × 0.1)
+ *
+ * Filters accounts that are:
+ * - Not rate-limited
+ * - Not invalid or disabled
+ * - Health score >= minUsable
+ * - Has tokens available
+ */
+
+import { BaseStrategy } from './base-strategy.js';
+import { HealthTracker, TokenBucketTracker } from './trackers/index.js';
+import { logger } from '../../utils/logger.js';
+
+// Default weights for scoring
+const DEFAULT_WEIGHTS = {
+    health: 2,
+    tokens: 5,
+    lru: 0.1
+};
+
+export class HybridStrategy extends BaseStrategy {
+    #healthTracker;
+    #tokenBucketTracker;
+    #weights;
+
+    /**
+     * Create a new HybridStrategy
+     * @param {Object} config - Strategy configuration
+     * @param {Object} [config.healthScore] - Health tracker configuration
+     * @param {Object} [config.tokenBucket] - Token bucket configuration
+     * @param {Object} [config.weights] - Scoring weights
+     */
+    constructor(config = {}) {
+        super(config);
+        this.#healthTracker = new HealthTracker(config.healthScore || {});
+        this.#tokenBucketTracker = new TokenBucketTracker(config.tokenBucket || {});
+        this.#weights = { ...DEFAULT_WEIGHTS, ...config.weights };
+    }
+
+    /**
+     * Select an account based on combined health, tokens, and LRU score
+     *
+     * @param {Array} accounts - Array of account objects
+     * @param {string} modelId - The model ID for the request
+     * @param {Object} options - Additional options
+     * @returns {SelectionResult} The selected account and index
+     */
+    selectAccount(accounts, modelId, options = {}) {
+        const { onSave } = options;
+
+        if (accounts.length === 0) {
+            return { account: null, index: 0, waitMs: 0 };
+        }
+
+        // Get candidates that pass all filters
+        const candidates = this.#getCandidates(accounts, modelId);
+
+        if (candidates.length === 0) {
+            logger.debug('[HybridStrategy] No candidates available');
+            return { account: null, index: 0, waitMs: 0 };
+        }
+
+        // Score and sort candidates
+        const scored = candidates.map(({ account, index }) => ({
+            account,
+            index,
+            score: this.#calculateScore(account)
+        }));
+
+        scored.sort((a, b) => b.score - a.score);
+
+        // Select the best candidate
+        const best = scored[0];
+        best.account.lastUsed = Date.now();
+
+        // Consume a token from the bucket
+        this.#tokenBucketTracker.consume(best.account.email);
+
+        if (onSave) onSave();
+
+        const position = best.index + 1;
+        const total = accounts.length;
+        logger.info(`[HybridStrategy] Using account: ${best.account.email} (${position}/${total}, score: ${best.score.toFixed(1)})`);
+
+        return { account: best.account, index: best.index, waitMs: 0 };
+    }
+
+    /**
+     * Called after a successful request
+     */
+    onSuccess(account, modelId) {
+        if (account && account.email) {
+            this.#healthTracker.recordSuccess(account.email);
+        }
+    }
+
+    /**
+     * Called when a request is rate-limited
+     */
+    onRateLimit(account, modelId) {
+        if (account && account.email) {
+            this.#healthTracker.recordRateLimit(account.email);
+        }
+    }
+
+    /**
+     * Called when a request fails
+     */
+    onFailure(account, modelId) {
+        if (account && account.email) {
+            this.#healthTracker.recordFailure(account.email);
+            // Refund the token since the request didn't complete
+            this.#tokenBucketTracker.refund(account.email);
+        }
+    }
+
+    /**
+     * Get candidates that pass all filters
+     * @private
+     */
+    #getCandidates(accounts, modelId) {
+        return accounts
+            .map((account, index) => ({ account, index }))
+            .filter(({ account }) => {
+                // Basic usability check
+                if (!this.isAccountUsable(account, modelId)) {
+                    return false;
+                }
+
+                // Health score check
+                if (!this.#healthTracker.isUsable(account.email)) {
+                    return false;
+                }
+
+                // Token availability check
+                if (!this.#tokenBucketTracker.hasTokens(account.email)) {
+                    return false;
+                }
+
+                return true;
+            });
+    }
+
+    /**
+     * Calculate the combined score for an account
+     * @private
+     */
+    #calculateScore(account) {
+        const email = account.email;
+
+        // Health component (0-100 scaled by weight)
+        const health = this.#healthTracker.getScore(email);
+        const healthComponent = health * this.#weights.health;
+
+        // Token component (0-100 scaled by weight)
+        const tokens = this.#tokenBucketTracker.getTokens(email);
+        const maxTokens = this.#tokenBucketTracker.getMaxTokens();
+        const tokenRatio = tokens / maxTokens;
+        const tokenComponent = (tokenRatio * 100) * this.#weights.tokens;
+
+        // LRU component (older = higher score)
+        // Use time since last use, capped at 1 hour for scoring
+        const lastUsed = account.lastUsed || 0;
+        const timeSinceLastUse = Math.min(Date.now() - lastUsed, 3600000); // Cap at 1 hour
+        const lruMinutes = timeSinceLastUse / 60000;
+        const lruComponent = lruMinutes * this.#weights.lru;
+
+        return healthComponent + tokenComponent + lruComponent;
+    }
+
+    /**
+     * Get the health tracker (for testing/debugging)
+     * @returns {HealthTracker} The health tracker instance
+     */
+    getHealthTracker() {
+        return this.#healthTracker;
+    }
+
+    /**
+     * Get the token bucket tracker (for testing/debugging)
+     * @returns {TokenBucketTracker} The token bucket tracker instance
+     */
+    getTokenBucketTracker() {
+        return this.#tokenBucketTracker;
+    }
+}
+
+export default HybridStrategy;
diff --git a/src/account-manager/strategies/index.js b/src/account-manager/strategies/index.js
new file mode 100644
index 0000000..239170a
--- /dev/null
+++ b/src/account-manager/strategies/index.js
@@ -0,0 +1,85 @@
+/**
+ * Strategy Factory
+ *
+ * Creates and exports account selection strategy instances.
+ */
+
+import { StickyStrategy } from './sticky-strategy.js';
+import { RoundRobinStrategy } from './round-robin-strategy.js';
+import { HybridStrategy } from './hybrid-strategy.js';
+import { logger } from '../../utils/logger.js';
+import {
+    SELECTION_STRATEGIES,
+    DEFAULT_SELECTION_STRATEGY
+} from '../../constants.js';
+
+// Re-export strategy constants for convenience
+export const STRATEGY_NAMES = SELECTION_STRATEGIES;
+export const DEFAULT_STRATEGY = DEFAULT_SELECTION_STRATEGY;
+
+// Strategy display labels
+export const STRATEGY_LABELS = {
+    'sticky': 'Sticky (Cache Optimized)',
+    'round-robin': 'Round Robin (Load Balanced)',
+    'hybrid': 'Hybrid (Smart Distribution)'
+};
+
+/**
+ * Create a strategy instance
+ * @param {string} strategyName - Name of the strategy ('sticky', 'round-robin', 'hybrid')
+ * @param {Object} config - Strategy configuration
+ * @returns {BaseStrategy} The strategy instance
+ */
+export function createStrategy(strategyName, config = {}) {
+    const name = (strategyName || DEFAULT_STRATEGY).toLowerCase();
+
+    switch (name) {
+        case 'sticky':
+            logger.debug('[Strategy] Creating StickyStrategy');
+            return new StickyStrategy(config);
+
+        case 'round-robin':
+        case 'roundrobin':
+            logger.debug('[Strategy] Creating RoundRobinStrategy');
+            return new RoundRobinStrategy(config);
+
+        case 'hybrid':
+            logger.debug('[Strategy] Creating HybridStrategy');
+            return new HybridStrategy(config);
+
+        default:
+            logger.warn(`[Strategy] Unknown strategy "${strategyName}", falling back to ${DEFAULT_STRATEGY}`);
+            return new HybridStrategy(config);
+    }
+}
+
+/**
+ * Check if a strategy name is valid
+ * @param {string} name - Strategy name to check
+ * @returns {boolean} True if valid
+ */
+export function isValidStrategy(name) {
+    if (!name) return false;
+    const lower = name.toLowerCase();
+    return STRATEGY_NAMES.includes(lower) || lower === 'roundrobin';
+}
+
+/**
+ * Get the display label for a strategy
+ * @param {string} name - Strategy name
+ * @returns {string} Display label
+ */
+export function getStrategyLabel(name) {
+    const lower = (name || DEFAULT_STRATEGY).toLowerCase();
+    if (lower === 'roundrobin') return STRATEGY_LABELS['round-robin'];
+    return STRATEGY_LABELS[lower] || STRATEGY_LABELS[DEFAULT_STRATEGY];
+}
+
+// Re-export strategies for direct use
+export { StickyStrategy } from './sticky-strategy.js';
+export { RoundRobinStrategy } from './round-robin-strategy.js';
+export { HybridStrategy } from './hybrid-strategy.js';
+export { BaseStrategy } from './base-strategy.js';
+
+// Re-export trackers
+export { HealthTracker, TokenBucketTracker } from './trackers/index.js';
diff --git a/src/account-manager/strategies/round-robin-strategy.js b/src/account-manager/strategies/round-robin-strategy.js
new file mode 100644
index 0000000..1b79bd6
--- /dev/null
+++ b/src/account-manager/strategies/round-robin-strategy.js
@@ -0,0 +1,76 @@
+/**
+ * Round-Robin Strategy
+ *
+ * Rotates to the next account on every request for maximum throughput.
+ * Does not maintain cache continuity but maximizes concurrent requests.
+ */
+
+import { BaseStrategy } from './base-strategy.js';
+import { logger } from '../../utils/logger.js';
+
+export class RoundRobinStrategy extends BaseStrategy {
+    #cursor = 0; // Tracks current position in rotation
+
+    /**
+     * Create a new RoundRobinStrategy
+     * @param {Object} config - Strategy configuration
+     */
+    constructor(config = {}) {
+        super(config);
+    }
+
+    /**
+     * Select the next available account in rotation
+     *
+     * @param {Array} accounts - Array of account objects
+     * @param {string} modelId - The model ID for the request
+     * @param {Object} options - Additional options
+     * @returns {SelectionResult} The selected account and index
+     */
+    selectAccount(accounts, modelId, options = {}) {
+        const { onSave } = options;
+
+        if (accounts.length === 0) {
+            return { account: null, index: 0, waitMs: 0 };
+        }
+
+        // Clamp cursor to valid range
+        if (this.#cursor >= accounts.length) {
+            this.#cursor = 0;
+        }
+
+        // Start from the next position after the cursor
+        const startIndex = (this.#cursor + 1) % accounts.length;
+
+        // Try each account starting from startIndex
+        for (let i = 0; i < accounts.length; i++) {
+            const idx = (startIndex + i) % accounts.length;
+            const account = accounts[idx];
+
+            if (this.isAccountUsable(account, modelId)) {
+                account.lastUsed = Date.now();
+                this.#cursor = idx;
+
+                if (onSave) onSave();
+
+                const position = idx + 1;
+                const total = accounts.length;
+                logger.info(`[RoundRobinStrategy] Using account: ${account.email} (${position}/${total})`);
+
+                return { account, index: idx, waitMs: 0 };
+            }
+        }
+
+        // No usable accounts found
+        return { account: null, index: this.#cursor, waitMs: 0 };
+    }
+
+    /**
+     * Reset the cursor position
+     */
+    resetCursor() {
+        this.#cursor = 0;
+    }
+}
+
+export default RoundRobinStrategy;
diff --git a/src/account-manager/strategies/sticky-strategy.js b/src/account-manager/strategies/sticky-strategy.js
new file mode 100644
index 0000000..988f370
--- /dev/null
+++ b/src/account-manager/strategies/sticky-strategy.js
@@ -0,0 +1,138 @@
+/**
+ * Sticky Strategy
+ *
+ * Keeps using the same account until it becomes unavailable (rate-limited or invalid).
+ * Best for prompt caching as it maintains cache continuity across requests.
+ */
+
+import { BaseStrategy } from './base-strategy.js';
+import { logger } from '../../utils/logger.js';
+import { formatDuration } from '../../utils/helpers.js';
+import { MAX_WAIT_BEFORE_ERROR_MS } from '../../constants.js';
+
+export class StickyStrategy extends BaseStrategy {
+    /**
+     * Create a new StickyStrategy
+     * @param {Object} config - Strategy configuration
+     */
+    constructor(config = {}) {
+        super(config);
+    }
+
+    /**
+     * Select an account with sticky preference
+     * Prefers the current account for cache continuity, only switches when:
+     * - Current account is rate-limited for > 2 minutes
+     * - Current account is invalid
+     * - Current account is disabled
+     *
+     * @param {Array} accounts - Array of account objects
+     * @param {string} modelId - The model ID for the request
+     * @param {Object} options - Additional options
+     * @returns {SelectionResult} The selected account and index
+     */
+    selectAccount(accounts, modelId, options = {}) {
+        const { currentIndex = 0, onSave } = options;
+
+        if (accounts.length === 0) {
+            return { account: null, index: currentIndex, waitMs: 0 };
+        }
+
+        // Clamp index to valid range
+        let index = currentIndex >= accounts.length ? 0 : currentIndex;
+        const currentAccount = accounts[index];
+
+        // Check if current account is usable
+        if (this.isAccountUsable(currentAccount, modelId)) {
+            currentAccount.lastUsed = Date.now();
+            if (onSave) onSave();
+            return { account: currentAccount, index, waitMs: 0 };
+        }
+
+        // Current account is not usable - check if others are available
+        const usableAccounts = this.getUsableAccounts(accounts, modelId);
+
+        if (usableAccounts.length > 0) {
+            // Found a free account - switch immediately
+            const { account: nextAccount, index: nextIndex } = this.#pickNext(
+                accounts,
+                index,
+                modelId,
+                onSave
+            );
+            if (nextAccount) {
+                logger.info(`[StickyStrategy] Switched to new account (failover): ${nextAccount.email}`);
+                return { account: nextAccount, index: nextIndex, waitMs: 0 };
+            }
+        }
+
+        // No other accounts available - check if we should wait for current
+        const waitInfo = this.#shouldWaitForAccount(currentAccount, modelId);
+        if (waitInfo.shouldWait) {
+            logger.info(`[StickyStrategy] Waiting ${formatDuration(waitInfo.waitMs)} for sticky account: ${currentAccount.email}`);
+            return { account: null, index, waitMs: waitInfo.waitMs };
+        }
+
+        // Current account unavailable for too long, try to find any other
+        const { account: nextAccount, index: nextIndex } = this.#pickNext(
+            accounts,
+            index,
+            modelId,
+            onSave
+        );
+
+        return { account: nextAccount, index: nextIndex, waitMs: 0 };
+    }
+
+    /**
+     * Pick the next available account starting from after the current index
+     * @private
+     */
+    #pickNext(accounts, currentIndex, modelId, onSave) {
+        for (let i = 1; i <= accounts.length; i++) {
+            const idx = (currentIndex + i) % accounts.length;
+            const account = accounts[idx];
+
+            if (this.isAccountUsable(account, modelId)) {
+                account.lastUsed = Date.now();
+                if (onSave) onSave();
+
+                const position = idx + 1;
+                const total = accounts.length;
+                logger.info(`[StickyStrategy] Using account: ${account.email} (${position}/${total})`);
+
+                return { account, index: idx };
+            }
+        }
+
+        return { account: null, index: currentIndex };
+    }
+
+    /**
+     * Check if we should wait for an account's rate limit to reset
+     * @private
+     */
+    #shouldWaitForAccount(account, modelId) {
+        if (!account || account.isInvalid || account.enabled === false) {
+            return { shouldWait: false, waitMs: 0 };
+        }
+
+        let waitMs = 0;
+
+        if (modelId && account.modelRateLimits && account.modelRateLimits[modelId]) {
+            const limit = account.modelRateLimits[modelId];
+            if (limit.isRateLimited && limit.resetTime) {
+                waitMs = limit.resetTime - Date.now();
+            }
+        }
+
+        // Wait if within threshold
+        if (waitMs > 0 && waitMs <= MAX_WAIT_BEFORE_ERROR_MS) {
+            return { shouldWait: true, waitMs };
+        }
+
+        return { shouldWait: false, waitMs: 0 };
+    }
+}
+
+export default StickyStrategy;
diff --git a/src/account-manager/strategies/trackers/health-tracker.js b/src/account-manager/strategies/trackers/health-tracker.js
new file mode 100644
index 0000000..a53274f
--- /dev/null
+++ b/src/account-manager/strategies/trackers/health-tracker.js
@@ -0,0 +1,162 @@
+/**
+ * Health Tracker
+ *
+ * Tracks per-account health scores to prioritize healthy accounts.
+ * Scores increase on success and decrease on failures/rate limits.
+ * Passive recovery over time helps accounts recover from temporary issues.
+ */
+
+// Default configuration (matches opencode-antigravity-auth)
+const DEFAULT_CONFIG = {
+    initial: 70,           // Starting score for new accounts
+    successReward: 1,      // Points on successful request
+    rateLimitPenalty: -10, // Points on rate limit
+    failurePenalty: -20,   // Points on other failures
+    recoveryPerHour: 2,    // Passive recovery rate
+    minUsable: 50,         // Minimum score to be selected
+    maxScore: 100          // Maximum score cap
+};
+
+export class HealthTracker {
+    #scores = new Map(); // email -> { score, lastUpdated, consecutiveFailures }
+    #config;
+
+    /**
+     * Create a new HealthTracker
+     * @param {Object} config - Health score configuration
+     */
+    constructor(config = {}) {
+        this.#config = { ...DEFAULT_CONFIG, ...config };
+    }
+
+    /**
+     * Get the health score for an account
+     * @param {string} email - Account email
+     * @returns {number} Current health score (with passive recovery applied)
+     */
+    getScore(email) {
+        const record = this.#scores.get(email);
+        if (!record) {
+            return this.#config.initial;
+        }
+
+        // Apply passive recovery based on time elapsed
+        const now = Date.now();
+        const hoursElapsed = (now - record.lastUpdated) / (1000 * 60 * 60);
+        const recovery = hoursElapsed * this.#config.recoveryPerHour;
+        const recoveredScore = Math.min(
+            this.#config.maxScore,
+            record.score + recovery
+        );
+
+        return recoveredScore;
+    }
+
+    /**
+     * Record a successful request for an account
+     * @param {string} email - Account email
+     */
+    recordSuccess(email) {
+        const currentScore = this.getScore(email);
+        const newScore = Math.min(
+            this.#config.maxScore,
+            currentScore + this.#config.successReward
+        );
+        this.#scores.set(email, {
+            score: newScore,
+            lastUpdated: Date.now(),
+            consecutiveFailures: 0 // Reset on success
+        });
+    }
+
+    /**
+     * Record a rate limit for an account
+     * @param {string} email - Account email
+     */
+    recordRateLimit(email) {
+        const record = this.#scores.get(email);
+        const currentScore = this.getScore(email);
+        const newScore = Math.max(
+            0,
+            currentScore + this.#config.rateLimitPenalty
+        );
+        this.#scores.set(email, {
+            score: newScore,
+            lastUpdated: Date.now(),
+            consecutiveFailures: (record?.consecutiveFailures ?? 0) + 1
+        });
+    }
+
+    /**
+     * Record a failure for an account
+     * @param {string} email - Account email
+     */
+    recordFailure(email) {
+        const record = this.#scores.get(email);
+        const currentScore = this.getScore(email);
+        const newScore = Math.max(
+            0,
+            currentScore + this.#config.failurePenalty
+        );
+        this.#scores.set(email, {
+            score: newScore,
+            lastUpdated: Date.now(),
+            consecutiveFailures: (record?.consecutiveFailures ?? 0) + 1
+        });
+    }
+
+    /**
+     * Check if an account is usable based on health score
+     * @param {string} email - Account email
+     * @returns {boolean} True if account health score is above minimum threshold
+     */
+    isUsable(email) {
+        return this.getScore(email) >= this.#config.minUsable;
+    }
+
+    /**
+     * Get the minimum usable score threshold
+     * @returns {number} Minimum score for an account to be usable
+     */
+    getMinUsable() {
+        return this.#config.minUsable;
+    }
+
+    /**
+     * Get the maximum score cap
+     * @returns {number} Maximum health score
+     */
+    getMaxScore() {
+        return this.#config.maxScore;
+    }
+
+    /**
+     * Reset the score for an account (e.g., after re-authentication)
+     * @param {string} email - Account email
+     */
+    reset(email) {
+        this.#scores.set(email, {
+            score: this.#config.initial,
+            lastUpdated: Date.now(),
+            consecutiveFailures: 0
+        });
+    }
+
+    /**
+     * Get the consecutive failure count for an account
+     * @param {string} email - Account email
+     * @returns {number} Number of consecutive failures
+     */
+    getConsecutiveFailures(email) {
+        return this.#scores.get(email)?.consecutiveFailures ?? 0;
+    }
+
+    /**
+     * Clear all tracked scores
+     */
+    clear() {
+        this.#scores.clear();
+    }
+}
+
+export default HealthTracker;
diff --git a/src/account-manager/strategies/trackers/index.js b/src/account-manager/strategies/trackers/index.js
new file mode 100644
index 0000000..974f527
--- /dev/null
+++ b/src/account-manager/strategies/trackers/index.js
@@ -0,0 +1,8 @@
+/**
+ * Trackers Index
+ *
+ * Exports all tracker classes for account selection strategies.
+ */
+
+export { HealthTracker } from './health-tracker.js';
+export { TokenBucketTracker } from './token-bucket-tracker.js';
diff --git a/src/account-manager/strategies/trackers/token-bucket-tracker.js b/src/account-manager/strategies/trackers/token-bucket-tracker.js
new file mode 100644
index 0000000..33d548c
--- /dev/null
+++ b/src/account-manager/strategies/trackers/token-bucket-tracker.js
@@ -0,0 +1,121 @@
+/**
+ * Token Bucket Tracker
+ *
+ * Client-side rate limiting using the token bucket algorithm.
+ * Each account has a bucket of tokens that regenerate over time.
+ * Requests consume tokens; accounts without tokens are deprioritized.
+ */
+
+// Default configuration (matches opencode-antigravity-auth)
+const DEFAULT_CONFIG = {
+    maxTokens: 50,        // Maximum token capacity
+    tokensPerMinute: 6,   // Regeneration rate
+    initialTokens: 50     // Starting tokens
+};
+
+export class TokenBucketTracker {
+    #buckets = new Map(); // email -> { tokens, lastUpdated }
+    #config;
+
+    /**
+     * Create a new TokenBucketTracker
+     * @param {Object} config - Token bucket configuration
+     */
+    constructor(config = {}) {
+        this.#config = { ...DEFAULT_CONFIG, ...config };
+    }
+
+    /**
+     * Get the current token count for an account
+     * @param {string} email - Account email
+     * @returns {number} Current token count (with regeneration applied)
+     */
+    getTokens(email) {
+        const bucket = this.#buckets.get(email);
+        if (!bucket) {
+            return this.#config.initialTokens;
+        }
+
+        // Apply token regeneration based on time elapsed
+        const now = Date.now();
+        const minutesElapsed = (now - bucket.lastUpdated) / (1000 * 60);
+        const regenerated = minutesElapsed * this.#config.tokensPerMinute;
+        const currentTokens = Math.min(
+            this.#config.maxTokens,
+            bucket.tokens + regenerated
+        );
+
+        return currentTokens;
+    }
+
+    /**
+     * Check if an account has tokens available
+     * @param {string} email - Account email
+     * @returns {boolean} True if account has at least 1 token
+     */
+    hasTokens(email) {
+        return this.getTokens(email) >= 1;
+    }
+
+    /**
+     * Consume a token from an account's bucket
+     * @param {string} email - Account email
+     * @returns {boolean} True if token was consumed, false if no tokens available
+     */
+    consume(email) {
+        const currentTokens = this.getTokens(email);
+        if (currentTokens < 1) {
+            return false;
+        }
+
+        this.#buckets.set(email, {
+            tokens: currentTokens - 1,
+            lastUpdated: Date.now()
+        });
+        return true;
+    }
+
+    /**
+     * Refund a token to an account's bucket (e.g., on request failure before processing)
+     * @param {string} email - Account email
+     */
+    refund(email) {
+        const currentTokens = this.getTokens(email);
+        const newTokens = Math.min(
+            this.#config.maxTokens,
+            currentTokens + 1
+        );
+        this.#buckets.set(email, {
+            tokens: newTokens,
+            lastUpdated: Date.now()
+        });
+    }
+
+    /**
+     * Get the maximum token capacity
+     * @returns {number} Maximum tokens per bucket
+     */
+    getMaxTokens() {
+        return this.#config.maxTokens;
+    }
+
+    /**
+     * Reset the bucket for an account
+     * @param {string} email - Account email
+     */
+    reset(email) {
+        this.#buckets.set(email, {
+            tokens: this.#config.initialTokens,
+            lastUpdated: Date.now()
+        });
+    }
+
+    /**
+     * Clear all tracked buckets
+     */
+    clear() {
+        this.#buckets.clear();
+    }
+}
+
+export default TokenBucketTracker;
diff --git a/src/cloudcode/message-handler.js b/src/cloudcode/message-handler.js
index 970e834..586ee16 100644
--- a/src/cloudcode/message-handler.js
+++ b/src/cloudcode/message-handler.js
@@ -10,6 +10,11 @@ import {
     MAX_RETRIES,
     MAX_WAIT_BEFORE_ERROR_MS,
     DEFAULT_COOLDOWN_MS,
+    RATE_LIMIT_DEDUP_WINDOW_MS,
+    MAX_CONSECUTIVE_FAILURES,
+    EXTENDED_COOLDOWN_MS,
+    CAPACITY_RETRY_DELAY_MS,
+    MAX_CAPACITY_RETRIES,
     isThinkingModel
 } from '../constants.js';
 import { convertGoogleToAnthropic } from '../format/index.js';
@@ -21,6 +26,85 @@ import { buildCloudCodeRequest, buildHeaders } from './request-builder.js';
 import { parseThinkingSSEResponse } from './sse-parser.js';
 import { getFallbackModel } from '../fallback-config.js';
 
+/**
+ * Gap 1: Rate limit deduplication - prevents thundering herd on concurrent rate limits
+ * Tracks last rate limit timestamp per model to skip duplicate retries
+ */
+const lastRateLimitTimestamps = new Map(); // modelId -> timestamp
+
+/**
+ * Check if we should skip retry due to recent rate limit on this model
+ * @param {string} model - Model ID
+ * @returns {boolean} True if retry should be skipped (within dedup window)
+ */
+function shouldSkipRetryDueToDedup(model) {
+    const lastTimestamp = lastRateLimitTimestamps.get(model);
+    if (!lastTimestamp) return false;
+
+    const elapsed = Date.now() - lastTimestamp;
+    if (elapsed < RATE_LIMIT_DEDUP_WINDOW_MS) {
+        logger.debug(`[CloudCode] Rate limit on ${model} within dedup window (${elapsed}ms ago), skipping retry`);
+        return true;
+    }
+    return false;
+}
+
+/**
+ * Record rate limit timestamp for deduplication
+ * @param {string} model - Model ID
+ */
+function recordRateLimitTimestamp(model) {
+    lastRateLimitTimestamps.set(model, Date.now());
+}
+
+/**
+ * Clear rate limit timestamp after successful retry
+ * @param {string} model - Model ID
+ */
+function clearRateLimitTimestamp(model) {
+    lastRateLimitTimestamps.delete(model);
+}
+
+/**
+ * Gap 3: Detect permanent authentication failures that require re-authentication
+ * These should mark the account as invalid rather than just clearing cache
+ * @param {string} errorText - Error message from API
+ * @returns {boolean} True if permanent auth failure
+ */
+function isPermanentAuthFailure(errorText) {
+    const lower = (errorText || '').toLowerCase();
+    return lower.includes('invalid_grant') ||
+        lower.includes('token revoked') ||
+        lower.includes('token has been expired or revoked') ||
+        lower.includes('token_revoked') ||
+        lower.includes('invalid_client') ||
+        lower.includes('credentials are invalid');
+}
+
+/**
+ * Gap 4: Detect if 429 error is due to model capacity (not user quota)
+ * Capacity issues should retry on same account with shorter delay
+ * @param {string} errorText - Error message from API
+ * @returns {boolean} True if capacity exhausted (not quota)
+ */
+function isModelCapacityExhausted(errorText) {
+    const lower = (errorText || '').toLowerCase();
+    return lower.includes('model_capacity_exhausted') ||
+        lower.includes('capacity_exhausted') ||
+        lower.includes('model is currently overloaded') ||
+        lower.includes('service temporarily unavailable');
+}
+
+// Periodically clean up stale dedup timestamps (every 60 seconds)
+setInterval(() => {
+    const cutoff = Date.now() - 60000; // 1 minute
+    for (const [model, timestamp] of lastRateLimitTimestamps.entries()) {
+        if (timestamp < cutoff) {
+            lastRateLimitTimestamps.delete(model);
+        }
+    }
+}, 60000);
+
 /**
  * Send a non-streaming request to Cloud Code with multi-account support
  * Uses SSE endpoint for thinking models (non-streaming doesn't return thinking blocks)
@@ -83,10 +167,14 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
             throw new Error('No accounts available');
         }
 
-        // Pick sticky account (prefers current for cache continuity)
-        let account = accountManager.getCurrentStickyAccount(model);
-        if (!account) {
-            account = accountManager.pickNext(model);
+        // Select account using configured strategy
+        const { account, waitMs } = accountManager.selectAccount(model);
+
+        // If strategy returns a wait time, sleep and retry
+        if (!account && waitMs > 0) {
+            logger.info(`[CloudCode] Waiting ${formatDuration(waitMs)} for account...`);
+            await sleep(waitMs + 500);
+            continue;
         }
 
         if (!account) {
@@ -101,11 +189,14 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
 
             logger.debug(`[CloudCode] Sending request for model: ${model}`);
 
-            // Try each endpoint
+            // Try each endpoint with index-based loop for capacity retry support
             let lastError = null;
             let retriedOnce = false; // Track if we've already retried for short rate limit
+            let capacityRetryCount = 0; // Gap 4: Track capacity exhaustion retries
+            let endpointIndex = 0;
 
-            for (const endpoint of ANTIGRAVITY_ENDPOINT_FALLBACKS) {
+            while (endpointIndex < ANTIGRAVITY_ENDPOINT_FALLBACKS.length) {
+                const endpoint = ANTIGRAVITY_ENDPOINT_FALLBACKS[endpointIndex];
                 try {
                     const url = isThinking
                         ? `${endpoint}/v1internal:streamGenerateContent?alt=sse`
@@ -122,16 +213,45 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
                         logger.warn(`[CloudCode] Error at ${endpoint}: ${response.status} - ${errorText}`);
 
                         if (response.status === 401) {
-                            // Auth error - clear caches and retry with fresh token
-                            logger.warn('[CloudCode] Auth error, refreshing token...');
+                            // Gap 3: Check for permanent auth failures
+                            if (isPermanentAuthFailure(errorText)) {
+                                logger.error(`[CloudCode] Permanent auth failure for ${account.email}: ${errorText.substring(0, 100)}`);
+                                accountManager.markInvalid(account.email, 'Token revoked - re-authentication required');
+                                throw new Error(`AUTH_INVALID_PERMANENT: ${errorText}`);
+                            }
+
+                            // Transient auth error - clear caches and retry with fresh token
+                            logger.warn('[CloudCode] Transient auth error, refreshing token...');
                             accountManager.clearTokenCache(account.email);
                             accountManager.clearProjectCache(account.email);
+                            endpointIndex++;
                             continue;
                         }
 
                         if (response.status === 429) {
                             const resetMs = parseResetTime(response, errorText);
 
+                            // Gap 4: Check if capacity issue (NOT quota) - retry SAME endpoint
+                            if (isModelCapacityExhausted(errorText)) {
+                                if (capacityRetryCount < MAX_CAPACITY_RETRIES) {
+                                    capacityRetryCount++;
+                                    const waitMs = resetMs || CAPACITY_RETRY_DELAY_MS;
+                                    logger.info(`[CloudCode] Model capacity exhausted, retry ${capacityRetryCount}/${MAX_CAPACITY_RETRIES} after ${formatDuration(waitMs)}...`);
+                                    await sleep(waitMs);
+                                    // Don't increment endpointIndex - retry same endpoint
+                                    continue;
+                                }
+                                // Max capacity retries exceeded - treat as quota exhaustion
+                                logger.warn(`[CloudCode] Max capacity retries (${MAX_CAPACITY_RETRIES}) exceeded, switching account`);
+                            }
+
+                            // Gap 1: Check deduplication window to prevent thundering herd
+                            if (shouldSkipRetryDueToDedup(model)) {
+                                logger.info(`[CloudCode] Skipping retry due to recent rate limit, switching account...`);
+                                accountManager.markRateLimited(account.email, resetMs || DEFAULT_COOLDOWN_MS, model);
+                                throw new Error(`RATE_LIMITED_DEDUP: ${errorText}`);
+                            }
+
                             // Decision: wait and retry OR switch account
                             if (resetMs && resetMs > DEFAULT_COOLDOWN_MS) {
                                 // Long-term quota exhaustion (> 10s) - switch to next account
@@ -144,31 +264,11 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
 
                                 if (!retriedOnce) {
                                     retriedOnce = true;
+                                    recordRateLimitTimestamp(model); // Gap 1: Record before retry
                                     logger.info(`[CloudCode] Short rate limit (${formatDuration(waitMs)}), waiting and retrying...`);
                                     await sleep(waitMs);
-                                    // Retry same endpoint
-                                    const retryResponse = await fetch(url, {
-                                        method: 'POST',
-                                        headers: buildHeaders(token, model, isThinking ? 'text/event-stream' : 'application/json'),
-                                        body: JSON.stringify(payload)
-                                    });
-
-                                    if (retryResponse.ok) {
-                                        // Process retry response
-                                        if (isThinking) {
-                                            return await parseThinkingSSEResponse(retryResponse, anthropicRequest.model);
-                                        }
-                                        const data = await retryResponse.json();
-                                        logger.debug('[CloudCode] Response received after retry');
-                                        return convertGoogleToAnthropic(data, anthropicRequest.model);
-                                    }
-
-                                    // Retry also failed - parse new reset time
-                                    const retryErrorText = await retryResponse.text();
-                                    const retryResetMs = parseResetTime(retryResponse, retryErrorText);
-                                    logger.warn(`[CloudCode] Retry also failed, marking and switching...`);
-                                    accountManager.markRateLimited(account.email, retryResetMs || waitMs, model);
-                                    throw new Error(`RATE_LIMITED_AFTER_RETRY: ${retryErrorText}`);
+                                    // Don't increment endpointIndex - retry same endpoint
+                                    continue;
                                 } else {
                                     // Already retried once, mark and switch
                                     accountManager.markRateLimited(account.email, waitMs, model);
@@ -184,18 +284,26 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
                                 logger.warn(`[CloudCode] ${response.status} error, waiting 1s before retry...`);
                                 await sleep(1000);
                             }
+                            endpointIndex++;
                             continue;
                         }
                     }
 
                     // For thinking models, parse SSE and accumulate all parts
                     if (isThinking) {
-                        return await parseThinkingSSEResponse(response, anthropicRequest.model);
+                        const result = await parseThinkingSSEResponse(response, anthropicRequest.model);
+                        // Gap 1: Clear timestamp on success
+                        clearRateLimitTimestamp(model);
+                        accountManager.notifySuccess(account, model);
+                        return result;
                     }
 
                     // Non-thinking models use regular JSON
                     const data = await response.json();
                     logger.debug('[CloudCode] Response received');
+                    // Gap 1: Clear timestamp on success
+                    clearRateLimitTimestamp(model);
+                    accountManager.notifySuccess(account, model);
                     return convertGoogleToAnthropic(data, anthropicRequest.model);
 
                 } catch (endpointError) {
@@ -204,6 +312,7 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
                     }
                     logger.warn(`[CloudCode] Error at ${endpoint}:`, endpointError.message);
                     lastError = endpointError;
+                    endpointIndex++;
                 }
             }
 
@@ -219,7 +328,8 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
 
         } catch (error) {
             if (isRateLimitError(error)) {
-                // Rate limited - already marked, continue to next account
+                // Rate limited - already marked, notify strategy and continue to next account
+                accountManager.notifyRateLimit(account, model);
                 logger.info(`[CloudCode] Account ${account.email} rate-limited, trying next...`);
                 continue;
             }
@@ -230,15 +340,31 @@ export async function sendMessage(anthropicRequest, accountManager, fallbackEnab
             }
             // Handle 5xx errors
             if (error.message.includes('API error 5') || error.message.includes('500') || error.message.includes('503')) {
-                logger.warn(`[CloudCode] Account ${account.email} failed with 5xx error, trying next...`);
-                accountManager.pickNext(model);
+                accountManager.notifyFailure(account, model);
+
+                // Gap 2: Check consecutive failures for extended cooldown
+                const consecutiveFailures = accountManager.getHealthTracker()?.getConsecutiveFailures(account.email) || 0;
+                if (consecutiveFailures >= MAX_CONSECUTIVE_FAILURES) {
+                    logger.warn(`[CloudCode] Account ${account.email} has ${consecutiveFailures} consecutive failures, applying extended cooldown (${formatDuration(EXTENDED_COOLDOWN_MS)})`);
+                    accountManager.markRateLimited(account.email, EXTENDED_COOLDOWN_MS, model);
+                } else {
+                    logger.warn(`[CloudCode] Account ${account.email} failed with 5xx error, trying next...`);
+                }
                 continue;
             }
 
             if (isNetworkError(error)) {
-                logger.warn(`[CloudCode] Network error for ${account.email}, trying next account... (${error.message})`);
+                accountManager.notifyFailure(account, model);
+
+                // Gap 2: Check consecutive failures for extended cooldown
+                const consecutiveFailures = accountManager.getHealthTracker()?.getConsecutiveFailures(account.email) || 0;
+                if (consecutiveFailures >= MAX_CONSECUTIVE_FAILURES) {
+                    logger.warn(`[CloudCode] Account ${account.email} has ${consecutiveFailures} consecutive network failures, applying extended cooldown (${formatDuration(EXTENDED_COOLDOWN_MS)})`);
+                    accountManager.markRateLimited(account.email, EXTENDED_COOLDOWN_MS, model);
+                } else {
+                    logger.warn(`[CloudCode] Network error for ${account.email}, trying next account... (${error.message})`);
+                }
                 await sleep(1000);
-                accountManager.pickNext(model);
                 continue;
             }
 
diff --git a/src/cloudcode/streaming-handler.js b/src/cloudcode/streaming-handler.js
index f5d76f0..df74a68 100644
--- a/src/cloudcode/streaming-handler.js
+++ b/src/cloudcode/streaming-handler.js
@@ -10,7 +10,12 @@ import {
     MAX_RETRIES,
     MAX_EMPTY_RESPONSE_RETRIES,
     MAX_WAIT_BEFORE_ERROR_MS,
-    DEFAULT_COOLDOWN_MS
+    DEFAULT_COOLDOWN_MS,
+    RATE_LIMIT_DEDUP_WINDOW_MS,
+    MAX_CONSECUTIVE_FAILURES,
+    EXTENDED_COOLDOWN_MS,
+    CAPACITY_RETRY_DELAY_MS,
+    MAX_CAPACITY_RETRIES
 } from '../constants.js';
 import { isRateLimitError, isAuthError, isEmptyResponseError } from '../errors.js';
 import { formatDuration, sleep, isNetworkError } from '../utils/helpers.js';
@@ -21,6 +26,83 @@ import { streamSSEResponse } from './sse-streamer.js';
 import { getFallbackModel } from '../fallback-config.js';
 import crypto from 'crypto';
 
+/**
+ * Gap 1: Rate limit deduplication - prevents thundering herd on concurrent rate limits
+ * Tracks last rate limit timestamp per model to skip duplicate retries
+ */
+const lastRateLimitTimestamps = new Map(); // modelId -> timestamp
+
+/**
+ * Check if we should skip retry due to recent rate limit on this model
+ * @param {string} model - Model ID
+ * @returns {boolean} True if retry should be skipped (within dedup window)
+ */
+function shouldSkipRetryDueToDedup(model) {
+    const lastTimestamp = lastRateLimitTimestamps.get(model);
+    if (!lastTimestamp) return false;
+
+    const elapsed = Date.now() - lastTimestamp;
+    if (elapsed < RATE_LIMIT_DEDUP_WINDOW_MS) {
+        logger.debug(`[CloudCode] Rate limit on ${model} within dedup window (${elapsed}ms ago), skipping retry`);
+        return true;
+    }
+    return false;
+}
+
+/**
+ * Record rate limit timestamp for deduplication
+ * @param {string} model - Model ID
+ */
+function recordRateLimitTimestamp(model) {
+    lastRateLimitTimestamps.set(model, Date.now());
+}
+
+/**
+ * Clear rate limit timestamp after successful retry
+ * @param {string} model - Model ID
+ */
+function clearRateLimitTimestamp(model) {
+    lastRateLimitTimestamps.delete(model);
+}
+
+/**
+ * Gap 3: Detect permanent authentication failures that require re-authentication
+ * @param {string} errorText - Error message from API
+ * @returns {boolean} True if permanent auth failure
+ */
+function isPermanentAuthFailure(errorText) {
+    const lower = (errorText || '').toLowerCase();
+    return lower.includes('invalid_grant') ||
+        lower.includes('token revoked') ||
+        lower.includes('token has been expired or revoked') ||
+        lower.includes('token_revoked') ||
+        lower.includes('invalid_client') ||
+        lower.includes('credentials are invalid');
+}
+
+/**
+ * Gap 4: Detect if 429 error is due to model capacity (not user quota)
+ * @param {string} errorText - Error message from API
+ * @returns {boolean} True if capacity exhausted (not quota)
+ */
+function isModelCapacityExhausted(errorText) {
+    const lower = (errorText || '').toLowerCase();
+    return lower.includes('model_capacity_exhausted') ||
+        lower.includes('capacity_exhausted') ||
+        lower.includes('model is currently overloaded') ||
+        lower.includes('service temporarily unavailable');
+}
+
+// Periodically clean up stale dedup timestamps (every 60 seconds)
+setInterval(() => {
+    const cutoff = Date.now() - 60000; // 1 minute
+    for (const [model, timestamp] of lastRateLimitTimestamps.entries()) {
+        if (timestamp < cutoff) {
+            lastRateLimitTimestamps.delete(model);
+        }
+    }
+}, 60000);
+
 /**
  * Send a streaming request to Cloud Code with multi-account support
  * Streams events in real-time as they arrive from the server
@@ -83,10 +165,14 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
             throw new Error('No accounts available');
         }
 
-        // Pick sticky account (prefers current for cache continuity)
-        let account = accountManager.getCurrentStickyAccount(model);
-        if (!account) {
-            account = accountManager.pickNext(model);
+        // Select account using configured strategy
+        const { account, waitMs } = accountManager.selectAccount(model);
+
+        // If strategy returns a wait time, sleep and retry
+        if (!account && waitMs > 0) {
+            logger.info(`[CloudCode] Waiting ${formatDuration(waitMs)} for account...`);
+            await sleep(waitMs + 500);
+            continue;
         }
 
         if (!account) {
@@ -101,11 +187,14 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
 
             logger.debug(`[CloudCode] Starting stream for model: ${model}`);
 
-            // Try each endpoint for streaming
+            // Try each endpoint with index-based loop for capacity retry support
             let lastError = null;
             let retriedOnce = false; // Track if we've already retried for short rate limit
+            let capacityRetryCount = 0; // Gap 4: Track capacity exhaustion retries
+            let endpointIndex = 0;
 
-            for (const endpoint of ANTIGRAVITY_ENDPOINT_FALLBACKS) {
+            while (endpointIndex < ANTIGRAVITY_ENDPOINT_FALLBACKS.length) {
+                const endpoint = ANTIGRAVITY_ENDPOINT_FALLBACKS[endpointIndex];
                 try {
                     const url = `${endpoint}/v1internal:streamGenerateContent?alt=sse`;
 
@@ -120,15 +209,44 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
                         logger.warn(`[CloudCode] Stream error at ${endpoint}: ${response.status} - ${errorText}`);
 
                         if (response.status === 401) {
-                            // Auth error - clear caches and retry
+                            // Gap 3: Check for permanent auth failures
+                            if (isPermanentAuthFailure(errorText)) {
+                                logger.error(`[CloudCode] Permanent auth failure for ${account.email}: ${errorText.substring(0, 100)}`);
+                                accountManager.markInvalid(account.email, 'Token revoked - re-authentication required');
+                                throw new Error(`AUTH_INVALID_PERMANENT: ${errorText}`);
+                            }
+
+                            // Transient auth error - clear caches and retry
                             accountManager.clearTokenCache(account.email);
                             accountManager.clearProjectCache(account.email);
+                            endpointIndex++;
                             continue;
                         }
 
                         if (response.status === 429) {
                             const resetMs = parseResetTime(response, errorText);
 
+                            // Gap 4: Check if capacity issue (NOT quota) - retry SAME endpoint
+                            if (isModelCapacityExhausted(errorText)) {
+                                if (capacityRetryCount < MAX_CAPACITY_RETRIES) {
+                                    capacityRetryCount++;
+                                    const waitMs = resetMs || CAPACITY_RETRY_DELAY_MS;
+                                    logger.info(`[CloudCode] Model capacity exhausted, retry ${capacityRetryCount}/${MAX_CAPACITY_RETRIES} after ${formatDuration(waitMs)}...`);
+                                    await sleep(waitMs);
+                                    // Don't increment endpointIndex - retry same endpoint
+                                    continue;
+                                }
+                                // Max capacity retries exceeded - treat as quota exhaustion
+                                logger.warn(`[CloudCode] Max capacity retries (${MAX_CAPACITY_RETRIES}) exceeded, switching account`);
+                            }
+
+                            // Gap 1: Check deduplication window to prevent thundering herd
+                            if (shouldSkipRetryDueToDedup(model)) {
+                                logger.info(`[CloudCode] Skipping retry due to recent rate limit, switching account...`);
+                                accountManager.markRateLimited(account.email, resetMs || DEFAULT_COOLDOWN_MS, model);
+                                throw new Error(`RATE_LIMITED_DEDUP: ${errorText}`);
+                            }
+
                             // Decision: wait and retry OR switch account
                             if (resetMs && resetMs > DEFAULT_COOLDOWN_MS) {
                                 // Long-term quota exhaustion (> 10s) - switch to next account
@@ -141,28 +259,11 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
 
                                 if (!retriedOnce) {
                                     retriedOnce = true;
+                                    recordRateLimitTimestamp(model); // Gap 1: Record before retry
                                     logger.info(`[CloudCode] Short rate limit (${formatDuration(waitMs)}), waiting and retrying...`);
                                     await sleep(waitMs);
-                                    // Retry same endpoint
-                                    const retryResponse = await fetch(url, {
-                                        method: 'POST',
-                                        headers: buildHeaders(token, model, 'text/event-stream'),
-                                        body: JSON.stringify(payload)
-                                    });
-
-                                    if (retryResponse.ok) {
-                                        // Stream the retry response
-                                        yield* streamSSEResponse(retryResponse, anthropicRequest.model);
-                                        logger.debug('[CloudCode] Stream completed after retry');
-                                        return;
-                                    }
-
-                                    // Retry also failed - parse new reset time
-                                    const retryErrorText = await retryResponse.text();
-                                    const retryResetMs = parseResetTime(retryResponse, retryErrorText);
-                                    logger.warn(`[CloudCode] Retry also failed, marking and switching...`);
-                                    accountManager.markRateLimited(account.email, retryResetMs || waitMs, model);
-                                    throw new Error(`RATE_LIMITED_AFTER_RETRY: ${retryErrorText}`);
+                                    // Don't increment endpointIndex - retry same endpoint
+                                    continue;
                                 } else {
                                     // Already retried once, mark and switch
                                     accountManager.markRateLimited(account.email, waitMs, model);
@@ -179,6 +280,7 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
                             await sleep(1000);
                         }
 
+                        endpointIndex++;
                         continue;
                     }
 
@@ -189,6 +291,9 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
                         try {
                             yield* streamSSEResponse(currentResponse, anthropicRequest.model);
                             logger.debug('[CloudCode] Stream completed');
+                            // Gap 1: Clear timestamp on success
+                            clearRateLimitTimestamp(model);
+                            accountManager.notifySuccess(account, model);
                             return;
                         } catch (streamError) {
                             // Only retry on EmptyResponseError
@@ -226,8 +331,13 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
                                     throw new Error(`429 RESOURCE_EXHAUSTED during retry: ${retryErrorText}`);
                                 }
 
-                                // Auth error - clear caches and throw with recognizable message
+                                // Auth error - check for permanent failure
                                 if (currentResponse.status === 401) {
+                                    if (isPermanentAuthFailure(retryErrorText)) {
+                                        logger.error(`[CloudCode] Permanent auth failure during retry for ${account.email}`);
+                                        accountManager.markInvalid(account.email, 'Token revoked - re-authentication required');
+                                        throw new Error(`AUTH_INVALID_PERMANENT: ${retryErrorText}`);
+                                    }
                                     accountManager.clearTokenCache(account.email);
                                     accountManager.clearProjectCache(account.email);
                                     throw new Error(`401 AUTH_INVALID during retry: ${retryErrorText}`);
@@ -261,6 +371,7 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
                     }
                     logger.warn(`[CloudCode] Stream error at ${endpoint}:`, endpointError.message);
                     lastError = endpointError;
+                    endpointIndex++;
                 }
             }
 
@@ -276,7 +387,8 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
 
         } catch (error) {
             if (isRateLimitError(error)) {
-                // Rate limited - already marked, continue to next account
+                // Rate limited - already marked, notify strategy and continue to next account
+                accountManager.notifyRateLimit(account, model);
                 logger.info(`[CloudCode] Account ${account.email} rate-limited, trying next...`);
                 continue;
             }
@@ -287,15 +399,31 @@ export async function* sendMessageStream(anthropicRequest, accountManager, fallb
             }
             // Handle 5xx errors
             if (error.message.includes('API error 5') || error.message.includes('500') || error.message.includes('503')) {
-                logger.warn(`[CloudCode] Account ${account.email} failed with 5xx stream error, trying next...`);
-                accountManager.pickNext(model);
+                accountManager.notifyFailure(account, model);
+
+                // Gap 2: Check consecutive failures for extended cooldown
+                const consecutiveFailures = accountManager.getHealthTracker()?.getConsecutiveFailures(account.email) || 0;
+                if (consecutiveFailures >= MAX_CONSECUTIVE_FAILURES) {
+                    logger.warn(`[CloudCode] Account ${account.email} has ${consecutiveFailures} consecutive failures, applying extended cooldown (${formatDuration(EXTENDED_COOLDOWN_MS)})`);
+                    accountManager.markRateLimited(account.email, EXTENDED_COOLDOWN_MS, model);
+                } else {
+                    logger.warn(`[CloudCode] Account ${account.email} failed with 5xx stream error, trying next...`);
+                }
                 continue;
             }
 
             if (isNetworkError(error)) {
-                logger.warn(`[CloudCode] Network error for ${account.email} (stream), trying next account... (${error.message})`);
+                accountManager.notifyFailure(account, model);
+
+                // Gap 2: Check consecutive failures for extended cooldown
+                const consecutiveFailures = accountManager.getHealthTracker()?.getConsecutiveFailures(account.email) || 0;
+                if (consecutiveFailures >= MAX_CONSECUTIVE_FAILURES) {
+                    logger.warn(`[CloudCode] Account ${account.email} has ${consecutiveFailures} consecutive network failures, applying extended cooldown (${formatDuration(EXTENDED_COOLDOWN_MS)})`);
+                    accountManager.markRateLimited(account.email, EXTENDED_COOLDOWN_MS, model);
+                } else {
+                    logger.warn(`[CloudCode] Network error for ${account.email} (stream), trying next account... (${error.message})`);
+                }
                 await sleep(1000);
-                accountManager.pickNext(model);
                 continue;
             }
 
diff --git a/src/config.js b/src/config.js
index 3aa3109..a05b3c2 100644
--- a/src/config.js
+++ b/src/config.js
@@ -15,7 +15,26 @@ const DEFAULT_CONFIG = {
     persistTokenCache: false,
     defaultCooldownMs: 10000,  // 10 seconds
     maxWaitBeforeErrorMs: 120000, // 2 minutes
-    modelMapping: {}
+    modelMapping: {},
+    // Account selection strategy configuration
+    accountSelection: {
+        strategy: 'hybrid',           // 'sticky' | 'round-robin' | 'hybrid'
+        // Hybrid strategy tuning (optional - sensible defaults)
+        healthScore: {
+            initial: 70,              // Starting score for new accounts
+            successReward: 1,         // Points on successful request
+            rateLimitPenalty: -10,    // Points on rate limit
+            failurePenalty: -20,      // Points on other failures
+            recoveryPerHour: 2,       // Passive recovery rate
+            minUsable: 50,            // Minimum score to be selected
+            maxScore: 100             // Maximum score cap
+        },
+        tokenBucket: {
+            maxTokens: 50,            // Maximum token capacity
+            tokensPerMinute: 6,       // Regeneration rate
+            initialTokens: 50         // Starting tokens
+        }
+    }
 };
 
 // Config locations
diff --git a/src/constants.js b/src/constants.js
index 4eb7a30..e8136e3 100644
--- a/src/constants.js
+++ b/src/constants.js
@@ -103,9 +103,24 @@ export const MAX_ACCOUNTS = config?.maxAccounts || 10; // From config or 10
 // Rate limit wait thresholds
 export const MAX_WAIT_BEFORE_ERROR_MS = config?.maxWaitBeforeErrorMs || 120000; // From config or 2 minutes
 
+// Gap 1: Retry deduplication - prevents thundering herd on concurrent rate limits
+export const RATE_LIMIT_DEDUP_WINDOW_MS = config?.rateLimitDedupWindowMs || 5000; // 5 seconds
+
+// Gap 2: Consecutive failure tracking - extended cooldown after repeated failures
+export const MAX_CONSECUTIVE_FAILURES = config?.maxConsecutiveFailures || 3;
+export const EXTENDED_COOLDOWN_MS = config?.extendedCooldownMs || 60000; // 1 minute
+
+// Gap 4: Capacity exhaustion - shorter retry for model capacity issues (not quota)
+export const CAPACITY_RETRY_DELAY_MS = config?.capacityRetryDelayMs || 2000; // 2 seconds
+export const MAX_CAPACITY_RETRIES = config?.maxCapacityRetries || 3;
+
 // Thinking model constants
 export const MIN_SIGNATURE_LENGTH = 50; // Minimum valid thinking signature length
 
+// Account selection strategies
+export const SELECTION_STRATEGIES = ['sticky', 'round-robin', 'hybrid'];
+export const DEFAULT_SELECTION_STRATEGY = 'hybrid';
+
 // Gemini-specific limits
 export const GEMINI_MAX_OUTPUT_TOKENS = 16384;
 
@@ -235,6 +250,11 @@ export default {
     MAX_EMPTY_RESPONSE_RETRIES,
     MAX_ACCOUNTS,
     MAX_WAIT_BEFORE_ERROR_MS,
+    RATE_LIMIT_DEDUP_WINDOW_MS,
+    MAX_CONSECUTIVE_FAILURES,
+    EXTENDED_COOLDOWN_MS,
+    CAPACITY_RETRY_DELAY_MS,
+    MAX_CAPACITY_RETRIES,
     MIN_SIGNATURE_LENGTH,
     GEMINI_MAX_OUTPUT_TOKENS,
     GEMINI_SKIP_SIGNATURE,
diff --git a/src/errors.js b/src/errors.js
index fac9410..c5a14f2 100644
--- a/src/errors.js
+++ b/src/errors.js
@@ -149,6 +149,23 @@ export class EmptyResponseError extends AntigravityError {
     }
 }
 
+/**
+ * Capacity exhausted error - Google's model is at capacity (not user quota)
+ * Should retry on same account with shorter delay, not switch accounts immediately
+ * Different from QUOTA_EXHAUSTED which indicates user's daily/hourly limit
+ */
+export class CapacityExhaustedError extends AntigravityError {
+    /**
+     * @param {string} message - Error message
+     * @param {number|null} retryAfterMs - Suggested retry delay in ms
+     */
+    constructor(message = 'Model capacity exhausted', retryAfterMs = null) {
+        super(message, 'CAPACITY_EXHAUSTED', true, { retryAfterMs });
+        this.name = 'CapacityExhaustedError';
+        this.retryAfterMs = retryAfterMs;
+    }
+}
+
 /**
  * Check if an error is a rate limit error
  * Works with both custom error classes and legacy string-based errors
@@ -188,6 +205,22 @@ export function isEmptyResponseError(error) {
         error?.name === 'EmptyResponseError';
 }
 
+/**
+ * Check if an error is a capacity exhausted error (model overload, not user quota)
+ * This is different from quota exhaustion - capacity issues are temporary infrastructure
+ * limits that should be retried on the SAME account with shorter delays
+ * @param {Error} error - Error to check
+ * @returns {boolean}
+ */
+export function isCapacityExhaustedError(error) {
+    if (error instanceof CapacityExhaustedError) return true;
+    const msg = (error.message || '').toLowerCase();
+    return msg.includes('model_capacity_exhausted') ||
+        msg.includes('capacity_exhausted') ||
+        msg.includes('model is currently overloaded') ||
+        msg.includes('service temporarily unavailable');
+}
+
 export default {
     AntigravityError,
     RateLimitError,
@@ -197,7 +230,9 @@ export default {
     ApiError,
     NativeModuleError,
     EmptyResponseError,
+    CapacityExhaustedError,
     isRateLimitError,
     isAuthError,
-    isEmptyResponseError
+    isEmptyResponseError,
+    isCapacityExhaustedError
 };
diff --git a/src/index.js b/src/index.js
index f17e9f6..8864766 100644
--- a/src/index.js
+++ b/src/index.js
@@ -3,9 +3,10 @@
  * Entry point - starts the proxy server
  */
 
-import app from './server.js';
+import app, { accountManager } from './server.js';
 import { DEFAULT_PORT } from './constants.js';
 import { logger } from './utils/logger.js';
+import { getStrategyLabel, STRATEGY_NAMES, DEFAULT_STRATEGY } from './account-manager/strategies/index.js';
 import path from 'path';
 import os from 'os';
 
@@ -14,6 +15,21 @@ const args = process.argv.slice(2);
 const isDebug = args.includes('--debug') || process.env.DEBUG === 'true';
 const isFallbackEnabled = args.includes('--fallback') || process.env.FALLBACK === 'true';
 
+// Parse --strategy flag (format: --strategy=sticky or --strategy sticky)
+let strategyOverride = null;
+for (let i = 0; i < args.length; i++) {
+    if (args[i].startsWith('--strategy=')) {
+        strategyOverride = args[i].split('=')[1];
+    } else if (args[i] === '--strategy' && args[i + 1]) {
+        strategyOverride = args[i + 1];
+    }
+}
+// Validate strategy
+if (strategyOverride && !STRATEGY_NAMES.includes(strategyOverride.toLowerCase())) {
+    logger.warn(`[Startup] Invalid strategy "${strategyOverride}". Valid options: ${STRATEGY_NAMES.join(', ')}. Using default.`);
+    strategyOverride = null;
+}
+
 // Initialize logger
 logger.setDebug(isDebug);
 
@@ -45,6 +61,7 @@ const server = app.listen(PORT, () => {
     
     // Build Control section dynamically
     let controlSection = '║  Control:                                                    ║\n';
+    controlSection += '║    --strategy=<s>     Set selection strategy (sticky/hybrid) ║\n';
     if (!isDebug) {
         controlSection += '║    --debug            Enable debug logging                   ║\n';
     }
@@ -53,17 +70,18 @@ const server = app.listen(PORT, () => {
     }
     controlSection += '║    Ctrl+C             Stop server                            ║';
 
-    // Build status section if any modes are active
-    let statusSection = '';
-    if (isDebug || isFallbackEnabled) {
-        statusSection = '║                                                              ║\n';
-        statusSection += '║  Active Modes:                                               ║\n';
-        if (isDebug) {
-            statusSection += '║    ✓ Debug mode enabled                                      ║\n';
-        }
-        if (isFallbackEnabled) {
-            statusSection += '║    ✓ Model fallback enabled                                  ║\n';
-        }
+    // Get the strategy label (accountManager will be initialized by now)
+    const strategyLabel = accountManager.getStrategyLabel();
+
+    // Build status section - always show strategy, plus any active modes
+    let statusSection = '║                                                              ║\n';
+    statusSection += '║  Active Modes:                                               ║\n';
+    statusSection += `${border}    ${align4(`✓ Strategy: ${strategyLabel}`)}${border}\n`;
+    if (isDebug) {
+        statusSection += '║    ✓ Debug mode enabled                                      ║\n';
+    }
+    if (isFallbackEnabled) {
+        statusSection += '║    ✓ Model fallback enabled                                  ║\n';
     }
 
     logger.log(`
diff --git a/src/server.js b/src/server.js
index 827eb58..ef5eb6c 100644
--- a/src/server.js
+++ b/src/server.js
@@ -26,13 +26,23 @@ import usageStats from './modules/usage-stats.js';
 const args = process.argv.slice(2);
 const FALLBACK_ENABLED = args.includes('--fallback') || process.env.FALLBACK === 'true';
 
+// Parse --strategy flag (format: --strategy=sticky or --strategy sticky)
+let STRATEGY_OVERRIDE = null;
+for (let i = 0; i < args.length; i++) {
+    if (args[i].startsWith('--strategy=')) {
+        STRATEGY_OVERRIDE = args[i].split('=')[1];
+    } else if (args[i] === '--strategy' && args[i + 1]) {
+        STRATEGY_OVERRIDE = args[i + 1];
+    }
+}
+
 const app = express();
 
 // Disable x-powered-by header for security
 app.disable('x-powered-by');
 
 // Initialize account manager (will be fully initialized on first request or startup)
-const accountManager = new AccountManager();
+export const accountManager = new AccountManager();
 
 // Track initialization status
 let isInitialized = false;
@@ -50,7 +60,7 @@ async function ensureInitialized() {
 
     initPromise = (async () => {
         try {
-            await accountManager.initialize();
+            await accountManager.initialize(STRATEGY_OVERRIDE);
             isInitialized = true;
             const status = accountManager.getStatus();
             logger.success(`[Server] Account pool initialized: ${status.summary}`);
diff --git a/src/webui/index.js b/src/webui/index.js
index 0386550..d449297 100644
--- a/src/webui/index.js
+++ b/src/webui/index.js
@@ -282,7 +282,7 @@ export function mountWebUI(app, dirname, accountManager) {
      */
     app.post('/api/config', (req, res) => {
         try {
-            const { debug, logLevel, maxRetries, retryBaseMs, retryMaxMs, persistTokenCache, defaultCooldownMs, maxWaitBeforeErrorMs } = req.body;
+            const { debug, logLevel, maxRetries, retryBaseMs, retryMaxMs, persistTokenCache, defaultCooldownMs, maxWaitBeforeErrorMs, accountSelection } = req.body;
 
             // Only allow updating specific fields (security)
             const updates = {};
@@ -308,6 +308,16 @@ export function mountWebUI(app, dirname, accountManager) {
             if (typeof maxWaitBeforeErrorMs === 'number' && maxWaitBeforeErrorMs >= 0 && maxWaitBeforeErrorMs <= 600000) {
                 updates.maxWaitBeforeErrorMs = maxWaitBeforeErrorMs;
             }
+            // Account selection strategy validation
+            if (accountSelection && typeof accountSelection === 'object') {
+                const validStrategies = ['sticky', 'round-robin', 'hybrid'];
+                if (accountSelection.strategy && validStrategies.includes(accountSelection.strategy)) {
+                    updates.accountSelection = {
+                        ...(config.accountSelection || {}),
+                        strategy: accountSelection.strategy
+                    };
+                }
+            }
 
             if (Object.keys(updates).length === 0) {
                 return res.status(400).json({
diff --git a/tests/run-all.cjs b/tests/run-all.cjs
index 01256cd..b8bec8f 100644
--- a/tests/run-all.cjs
+++ b/tests/run-all.cjs
@@ -9,6 +9,7 @@ const { spawn } = require('child_process');
 const path = require('path');
 
 const tests = [
+    { name: 'Account Selection Strategies', file: 'test-strategies.cjs' },
     { name: 'Thinking Signatures', file: 'test-thinking-signatures.cjs' },
     { name: 'Multi-turn Tools (Non-Streaming)', file: 'test-multiturn-thinking-tools.cjs' },
     { name: 'Multi-turn Tools (Streaming)', file: 'test-multiturn-thinking-tools-streaming.cjs' },
diff --git a/tests/test-strategies.cjs b/tests/test-strategies.cjs
new file mode 100644
index 0000000..0be8162
--- /dev/null
+++ b/tests/test-strategies.cjs
@@ -0,0 +1,795 @@
+/**
+ * Test Account Selection Strategies - Unit Tests
+ *
+ * Tests the strategy pattern implementation for account selection:
+ * - HealthTracker: health score tracking with passive recovery
+ * - TokenBucketTracker: token bucket rate limiting
+ * - StickyStrategy: cache-optimized sticky selection
+ * - RoundRobinStrategy: load-balanced rotation
+ * - HybridStrategy: smart multi-signal distribution
+ * - Strategy Factory: createStrategy, isValidStrategy, getStrategyLabel
+ */
+
+// Since we're in CommonJS and the module is ESM, we need to use dynamic import
+async function runTests() {
+    console.log('╔══════════════════════════════════════════════════════════════╗');
+    console.log('║           ACCOUNT SELECTION STRATEGY TEST SUITE              ║');
+    console.log('╚══════════════════════════════════════════════════════════════╝\n');
+
+    // Dynamic imports for ESM modules
+    const { HealthTracker } = await import('../src/account-manager/strategies/trackers/health-tracker.js');
+    const { TokenBucketTracker } = await import('../src/account-manager/strategies/trackers/token-bucket-tracker.js');
+    const { StickyStrategy } = await import('../src/account-manager/strategies/sticky-strategy.js');
+    const { RoundRobinStrategy } = await import('../src/account-manager/strategies/round-robin-strategy.js');
+    const { HybridStrategy } = await import('../src/account-manager/strategies/hybrid-strategy.js');
+    const { BaseStrategy } = await import('../src/account-manager/strategies/base-strategy.js');
+    const {
+        createStrategy,
+        isValidStrategy,
+        getStrategyLabel,
+        STRATEGY_NAMES,
+        DEFAULT_STRATEGY
+    } = await import('../src/account-manager/strategies/index.js');
+
+    let passed = 0;
+    let failed = 0;
+
+    function test(name, fn) {
+        try {
+            fn();
+            console.log(`✓ ${name}`);
+            passed++;
+        } catch (e) {
+            console.log(`✗ ${name}`);
+            console.log(`  Error: ${e.message}`);
+            failed++;
+        }
+    }
+
+    function assertEqual(actual, expected, message = '') {
+        if (actual !== expected) {
+            throw new Error(`${message}\nExpected: ${expected}\nActual: ${actual}`);
+        }
+    }
+
+    function assertDeepEqual(actual, expected, message = '') {
+        if (JSON.stringify(actual) !== JSON.stringify(expected)) {
+            throw new Error(`${message}\nExpected: ${JSON.stringify(expected, null, 2)}\nActual: ${JSON.stringify(actual, null, 2)}`);
+        }
+    }
+
+    function assertTrue(value, message = '') {
+        if (!value) {
+            throw new Error(message || 'Expected true but got false');
+        }
+    }
+
+    function assertFalse(value, message = '') {
+        if (value) {
+            throw new Error(message || 'Expected false but got true');
+        }
+    }
+
+    function assertNull(value, message = '') {
+        if (value !== null) {
+            throw new Error(`${message}\nExpected null but got: ${value}`);
+        }
+    }
+
+    function assertWithin(actual, min, max, message = '') {
+        if (actual < min || actual > max) {
+            throw new Error(`${message}\nExpected value between ${min} and ${max}, got: ${actual}`);
+        }
+    }
+
+    // Helper to create mock accounts
+    function createMockAccounts(count = 3, options = {}) {
+        return Array.from({ length: count }, (_, i) => ({
+            email: `account${i + 1}@example.com`,
+            enabled: true,
+            isInvalid: false,
+            lastUsed: Date.now() - (i * 60000), // Stagger by 1 minute
+            modelRateLimits: {},
+            ...options
+        }));
+    }
+
+    // ==========================================================================
+    // HEALTH TRACKER TESTS
+    // ==========================================================================
+    console.log('\n─── HealthTracker Tests ───');
+
+    test('HealthTracker: initial score is 70 by default', () => {
+        const tracker = new HealthTracker();
+        const score = tracker.getScore('new@example.com');
+        assertEqual(score, 70, 'Default initial score should be 70');
+    });
+
+    test('HealthTracker: custom initial score', () => {
+        const tracker = new HealthTracker({ initial: 80 });
+        const score = tracker.getScore('new@example.com');
+        assertEqual(score, 80, 'Custom initial score should be 80');
+    });
+
+    test('HealthTracker: recordSuccess increases score', () => {
+        const tracker = new HealthTracker({ initial: 70, successReward: 1 });
+        tracker.recordSuccess('test@example.com');
+        const score = tracker.getScore('test@example.com');
+        assertEqual(score, 71, 'Score should increase by 1 on success');
+    });
+
+    test('HealthTracker: recordRateLimit decreases score', () => {
+        const tracker = new HealthTracker({ initial: 70, rateLimitPenalty: -10 });
+        tracker.recordRateLimit('test@example.com');
+        const score = tracker.getScore('test@example.com');
+        assertEqual(score, 60, 'Score should decrease by 10 on rate limit');
+    });
+
+    test('HealthTracker: recordFailure decreases score', () => {
+        const tracker = new HealthTracker({ initial: 70, failurePenalty: -20 });
+        tracker.recordFailure('test@example.com');
+        const score = tracker.getScore('test@example.com');
+        assertEqual(score, 50, 'Score should decrease by 20 on failure');
+    });
+
+    test('HealthTracker: score cannot exceed maxScore', () => {
+        const tracker = new HealthTracker({ initial: 99, maxScore: 100, successReward: 5 });
+        tracker.recordSuccess('test@example.com');
+        const score = tracker.getScore('test@example.com');
+        assertEqual(score, 100, 'Score should be capped at maxScore');
+    });
+
+    test('HealthTracker: score cannot go below 0', () => {
+        const tracker = new HealthTracker({ initial: 10, failurePenalty: -50 });
+        tracker.recordFailure('test@example.com');
+        const score = tracker.getScore('test@example.com');
+        assertEqual(score, 0, 'Score should not go below 0');
+    });
+
+    test('HealthTracker: isUsable returns true when score >= minUsable', () => {
+        const tracker = new HealthTracker({ initial: 50, minUsable: 50 });
+        assertTrue(tracker.isUsable('test@example.com'), 'Should be usable at minUsable');
+    });
+
+    test('HealthTracker: isUsable returns false when score < minUsable', () => {
+        const tracker = new HealthTracker({ initial: 49, minUsable: 50 });
+        assertFalse(tracker.isUsable('test@example.com'), 'Should not be usable below minUsable');
+    });
+
+    test('HealthTracker: reset restores initial score', () => {
+        const tracker = new HealthTracker({ initial: 70 });
+        tracker.recordFailure('test@example.com'); // Score drops
+        tracker.reset('test@example.com');
+        const score = tracker.getScore('test@example.com');
+        assertEqual(score, 70, 'Reset should restore initial score');
+    });
+
+    test('HealthTracker: clear removes all scores', () => {
+        const tracker = new HealthTracker({ initial: 70 });
+        tracker.recordSuccess('a@example.com');
+        tracker.recordSuccess('b@example.com');
+        tracker.clear();
+        // After clear, new accounts should get initial score
+        assertEqual(tracker.getScore('a@example.com'), 70);
+        assertEqual(tracker.getScore('b@example.com'), 70);
+    });
+
+    test('HealthTracker: getConsecutiveFailures returns 0 for new account', () => {
+        const tracker = new HealthTracker();
+        assertEqual(tracker.getConsecutiveFailures('new@example.com'), 0);
+    });
+
+    test('HealthTracker: recordRateLimit increments consecutiveFailures', () => {
+        const tracker = new HealthTracker();
+        tracker.recordRateLimit('test@example.com');
+        assertEqual(tracker.getConsecutiveFailures('test@example.com'), 1);
+        tracker.recordRateLimit('test@example.com');
+        assertEqual(tracker.getConsecutiveFailures('test@example.com'), 2);
+    });
+
+    test('HealthTracker: recordFailure increments consecutiveFailures', () => {
+        const tracker = new HealthTracker();
+        tracker.recordFailure('test@example.com');
+        assertEqual(tracker.getConsecutiveFailures('test@example.com'), 1);
+    });
+
+    test('HealthTracker: recordSuccess resets consecutiveFailures', () => {
+        const tracker = new HealthTracker();
+        tracker.recordRateLimit('test@example.com');
+        tracker.recordRateLimit('test@example.com');
+        assertEqual(tracker.getConsecutiveFailures('test@example.com'), 2);
+        tracker.recordSuccess('test@example.com');
+        assertEqual(tracker.getConsecutiveFailures('test@example.com'), 0);
+    });
+
+    test('HealthTracker: reset clears consecutiveFailures', () => {
+        const tracker = new HealthTracker();
+        tracker.recordFailure('test@example.com');
+        tracker.recordFailure('test@example.com');
+        assertEqual(tracker.getConsecutiveFailures('test@example.com'), 2);
+        tracker.reset('test@example.com');
+        assertEqual(tracker.getConsecutiveFailures('test@example.com'), 0);
+    });
+
+    // ==========================================================================
+    // TOKEN BUCKET TRACKER TESTS
+    // ==========================================================================
+    console.log('\n─── TokenBucketTracker Tests ───');
+
+    test('TokenBucketTracker: initial tokens is 50 by default', () => {
+        const tracker = new TokenBucketTracker();
+        const tokens = tracker.getTokens('new@example.com');
+        assertEqual(tokens, 50, 'Default initial tokens should be 50');
+    });
+
+    test('TokenBucketTracker: custom initial tokens', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 30 });
+        const tokens = tracker.getTokens('new@example.com');
+        assertEqual(tokens, 30, 'Custom initial tokens should be 30');
+    });
+
+    test('TokenBucketTracker: consume decreases tokens', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 10, maxTokens: 10 });
+        const consumed = tracker.consume('test@example.com');
+        assertTrue(consumed, 'Consume should return true');
+        assertEqual(tracker.getTokens('test@example.com'), 9, 'Tokens should decrease by 1');
+    });
+
+    test('TokenBucketTracker: consume fails when no tokens', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 0, maxTokens: 10 });
+        const consumed = tracker.consume('test@example.com');
+        assertFalse(consumed, 'Consume should return false when no tokens');
+    });
+
+    test('TokenBucketTracker: hasTokens returns true when tokens > 0', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 1 });
+        assertTrue(tracker.hasTokens('test@example.com'), 'Should have tokens');
+    });
+
+    test('TokenBucketTracker: hasTokens returns false when tokens < 1', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 0 });
+        assertFalse(tracker.hasTokens('test@example.com'), 'Should not have tokens');
+    });
+
+    test('TokenBucketTracker: refund increases tokens', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 5, maxTokens: 10 });
+        tracker.consume('test@example.com'); // 5 -> 4
+        tracker.refund('test@example.com');  // 4 -> 5
+        assertEqual(tracker.getTokens('test@example.com'), 5, 'Refund should restore token');
+    });
+
+    test('TokenBucketTracker: refund cannot exceed maxTokens', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 10, maxTokens: 10 });
+        tracker.refund('test@example.com');
+        assertEqual(tracker.getTokens('test@example.com'), 10, 'Refund should not exceed max');
+    });
+
+    test('TokenBucketTracker: getMaxTokens returns configured max', () => {
+        const tracker = new TokenBucketTracker({ maxTokens: 100 });
+        assertEqual(tracker.getMaxTokens(), 100, 'getMaxTokens should return 100');
+    });
+
+    test('TokenBucketTracker: reset restores initial tokens', () => {
+        const tracker = new TokenBucketTracker({ initialTokens: 50, maxTokens: 50 });
+        tracker.consume('test@example.com');
+        tracker.consume('test@example.com');
+        tracker.reset('test@example.com');
+        assertEqual(tracker.getTokens('test@example.com'), 50, 'Reset should restore initial');
+    });
+
+    // ==========================================================================
+    // BASE STRATEGY TESTS
+    // ==========================================================================
+    console.log('\n─── BaseStrategy Tests ───');
+
+    test('BaseStrategy: cannot be instantiated directly', () => {
+        try {
+            new BaseStrategy();
+            throw new Error('Should have thrown');
+        } catch (e) {
+            assertTrue(e.message.includes('abstract'), 'Should throw abstract error');
+        }
+    });
+
+    test('BaseStrategy: isAccountUsable returns false for null account', () => {
+        // Create a minimal subclass to test
+        class TestStrategy extends BaseStrategy {
+            selectAccount() { return { account: null, index: 0 }; }
+        }
+        const strategy = new TestStrategy();
+        assertFalse(strategy.isAccountUsable(null, 'model'), 'Null account should not be usable');
+    });
+
+    test('BaseStrategy: isAccountUsable returns false for invalid account', () => {
+        class TestStrategy extends BaseStrategy {
+            selectAccount() { return { account: null, index: 0 }; }
+        }
+        const strategy = new TestStrategy();
+        const account = { email: 'test@example.com', isInvalid: true };
+        assertFalse(strategy.isAccountUsable(account, 'model'), 'Invalid account should not be usable');
+    });
+
+    test('BaseStrategy: isAccountUsable returns false for disabled account', () => {
+        class TestStrategy extends BaseStrategy {
+            selectAccount() { return { account: null, index: 0 }; }
+        }
+        const strategy = new TestStrategy();
+        const account = { email: 'test@example.com', enabled: false };
+        assertFalse(strategy.isAccountUsable(account, 'model'), 'Disabled account should not be usable');
+    });
+
+    test('BaseStrategy: isAccountUsable returns false for rate-limited model', () => {
+        class TestStrategy extends BaseStrategy {
+            selectAccount() { return { account: null, index: 0 }; }
+        }
+        const strategy = new TestStrategy();
+        const account = {
+            email: 'test@example.com',
+            modelRateLimits: {
+                'claude-sonnet': {
+                    isRateLimited: true,
+                    resetTime: Date.now() + 60000 // 1 minute in future
+                }
+            }
+        };
+        assertFalse(strategy.isAccountUsable(account, 'claude-sonnet'), 'Rate-limited model should not be usable');
+    });
+
+    test('BaseStrategy: isAccountUsable returns true for expired rate limit', () => {
+        class TestStrategy extends BaseStrategy {
+            selectAccount() { return { account: null, index: 0 }; }
+        }
+        const strategy = new TestStrategy();
+        const account = {
+            email: 'test@example.com',
+            modelRateLimits: {
+                'claude-sonnet': {
+                    isRateLimited: true,
+                    resetTime: Date.now() - 1000 // 1 second in past
+                }
+            }
+        };
+        assertTrue(strategy.isAccountUsable(account, 'claude-sonnet'), 'Expired rate limit should be usable');
+    });
+
+    test('BaseStrategy: getUsableAccounts filters correctly', () => {
+        class TestStrategy extends BaseStrategy {
+            selectAccount() { return { account: null, index: 0 }; }
+        }
+        const strategy = new TestStrategy();
+        const accounts = [
+            { email: 'a@example.com', enabled: true },
+            { email: 'b@example.com', enabled: false },
+            { email: 'c@example.com', enabled: true, isInvalid: true },
+            { email: 'd@example.com', enabled: true }
+        ];
+        const usable = strategy.getUsableAccounts(accounts, 'model');
+        assertEqual(usable.length, 2, 'Should have 2 usable accounts');
+        assertEqual(usable[0].account.email, 'a@example.com');
+        assertEqual(usable[1].account.email, 'd@example.com');
+    });
+
+    // ==========================================================================
+    // STICKY STRATEGY TESTS
+    // ==========================================================================
+    console.log('\n─── StickyStrategy Tests ───');
+
+    test('StickyStrategy: returns null for empty accounts', () => {
+        const strategy = new StickyStrategy();
+        const result = strategy.selectAccount([], 'model', { currentIndex: 0 });
+        assertNull(result.account, 'Should return null for empty accounts');
+    });
+
+    test('StickyStrategy: keeps using current account when available', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(3);
+
+        const result1 = strategy.selectAccount(accounts, 'model', { currentIndex: 0 });
+        assertEqual(result1.account.email, 'account1@example.com');
+        assertEqual(result1.index, 0);
+
+        const result2 = strategy.selectAccount(accounts, 'model', { currentIndex: 0 });
+        assertEqual(result2.account.email, 'account1@example.com', 'Should stick to same account');
+        assertEqual(result2.index, 0);
+    });
+
+    test('StickyStrategy: switches when current account is rate-limited', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(3);
+        // Rate-limit account1 for 5 minutes (longer than MAX_WAIT)
+        accounts[0].modelRateLimits = {
+            'model': { isRateLimited: true, resetTime: Date.now() + 300000 }
+        };
+
+        const result = strategy.selectAccount(accounts, 'model', { currentIndex: 0 });
+        assertEqual(result.account.email, 'account2@example.com', 'Should switch to next available');
+        assertEqual(result.index, 1);
+    });
+
+    test('StickyStrategy: returns waitMs when current account has short rate limit', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(1); // Only one account
+        // Rate-limit for 30 seconds (less than MAX_WAIT of 2 minutes)
+        accounts[0].modelRateLimits = {
+            'model': { isRateLimited: true, resetTime: Date.now() + 30000 }
+        };
+
+        const result = strategy.selectAccount(accounts, 'model', { currentIndex: 0 });
+        assertNull(result.account, 'Should return null when waiting');
+        assertWithin(result.waitMs, 29000, 31000, 'Should return ~30s wait time');
+    });
+
+    test('StickyStrategy: switches when current account is disabled', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(3);
+        accounts[0].enabled = false;
+
+        const result = strategy.selectAccount(accounts, 'model', { currentIndex: 0 });
+        assertEqual(result.account.email, 'account2@example.com', 'Should switch to next');
+    });
+
+    test('StickyStrategy: switches when current account is invalid', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(3);
+        accounts[0].isInvalid = true;
+
+        const result = strategy.selectAccount(accounts, 'model', { currentIndex: 0 });
+        assertEqual(result.account.email, 'account2@example.com', 'Should switch to next');
+    });
+
+    test('StickyStrategy: wraps around when at end of list', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(3);
+        accounts[2].isInvalid = true; // Last account invalid
+
+        const result = strategy.selectAccount(accounts, 'model', { currentIndex: 2 });
+        assertEqual(result.account.email, 'account1@example.com', 'Should wrap to first');
+        assertEqual(result.index, 0);
+    });
+
+    test('StickyStrategy: clamps invalid currentIndex', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(3);
+
+        const result = strategy.selectAccount(accounts, 'model', { currentIndex: 10 });
+        assertEqual(result.account.email, 'account1@example.com', 'Should clamp to valid index');
+        assertEqual(result.index, 0);
+    });
+
+    // ==========================================================================
+    // ROUND-ROBIN STRATEGY TESTS
+    // ==========================================================================
+    console.log('\n─── RoundRobinStrategy Tests ───');
+
+    test('RoundRobinStrategy: returns null for empty accounts', () => {
+        const strategy = new RoundRobinStrategy();
+        const result = strategy.selectAccount([], 'model');
+        assertNull(result.account, 'Should return null for empty accounts');
+    });
+
+    test('RoundRobinStrategy: rotates through accounts', () => {
+        const strategy = new RoundRobinStrategy();
+        const accounts = createMockAccounts(3);
+
+        const r1 = strategy.selectAccount(accounts, 'model');
+        const r2 = strategy.selectAccount(accounts, 'model');
+        const r3 = strategy.selectAccount(accounts, 'model');
+        const r4 = strategy.selectAccount(accounts, 'model');
+
+        // First call starts at cursor 0, looks at (0+1)%3 = 1
+        // Then cursor becomes 1, next looks at (1+1)%3 = 2
+        // Then cursor becomes 2, next looks at (2+1)%3 = 0
+        // Then cursor becomes 0, next looks at (0+1)%3 = 1
+        assertEqual(r1.account.email, 'account2@example.com', 'First should be account2');
+        assertEqual(r2.account.email, 'account3@example.com', 'Second should be account3');
+        assertEqual(r3.account.email, 'account1@example.com', 'Third should wrap to account1');
+        assertEqual(r4.account.email, 'account2@example.com', 'Fourth should continue rotation');
+    });
+
+    test('RoundRobinStrategy: skips unavailable accounts', () => {
+        const strategy = new RoundRobinStrategy();
+        const accounts = createMockAccounts(3);
+        accounts[1].enabled = false; // Disable account2
+
+        const r1 = strategy.selectAccount(accounts, 'model');
+        const r2 = strategy.selectAccount(accounts, 'model');
+        const r3 = strategy.selectAccount(accounts, 'model');
+
+        // account2 is skipped
+        assertEqual(r1.account.email, 'account3@example.com');
+        assertEqual(r2.account.email, 'account1@example.com');
+        assertEqual(r3.account.email, 'account3@example.com');
+    });
+
+    test('RoundRobinStrategy: returns null when all accounts unavailable', () => {
+        const strategy = new RoundRobinStrategy();
+        const accounts = createMockAccounts(3);
+        accounts.forEach(a => a.enabled = false);
+
+        const result = strategy.selectAccount(accounts, 'model');
+        assertNull(result.account, 'Should return null when all unavailable');
+    });
+
+    test('RoundRobinStrategy: resetCursor resets position', () => {
+        const strategy = new RoundRobinStrategy();
+        const accounts = createMockAccounts(3);
+
+        strategy.selectAccount(accounts, 'model'); // Moves cursor
+        strategy.selectAccount(accounts, 'model'); // Moves cursor
+        strategy.resetCursor();
+
+        const result = strategy.selectAccount(accounts, 'model');
+        assertEqual(result.account.email, 'account2@example.com', 'Should start from beginning after reset');
+    });
+
+    // ==========================================================================
+    // HYBRID STRATEGY TESTS
+    // ==========================================================================
+    console.log('\n─── HybridStrategy Tests ───');
+
+    test('HybridStrategy: returns null for empty accounts', () => {
+        const strategy = new HybridStrategy();
+        const result = strategy.selectAccount([], 'model');
+        assertNull(result.account, 'Should return null for empty accounts');
+    });
+
+    test('HybridStrategy: selects best scored account', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 70 },
+            tokenBucket: { initialTokens: 50, maxTokens: 50 }
+        });
+        const accounts = createMockAccounts(3);
+        // Make account3 older (higher LRU score)
+        accounts[2].lastUsed = Date.now() - 3600000; // 1 hour ago
+
+        const result = strategy.selectAccount(accounts, 'model');
+        // account3 should win due to higher LRU score
+        assertEqual(result.account.email, 'account3@example.com', 'Oldest account should be selected');
+    });
+
+    test('HybridStrategy: filters out unhealthy accounts', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 40, minUsable: 50 },
+            tokenBucket: { initialTokens: 50, maxTokens: 50 }
+        });
+        const accounts = createMockAccounts(3);
+
+        // All accounts start with health 40, which is below minUsable 50
+        const result = strategy.selectAccount(accounts, 'model');
+        assertNull(result.account, 'Should filter all accounts with low health');
+    });
+
+    test('HybridStrategy: filters out accounts without tokens', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 70 },
+            tokenBucket: { initialTokens: 0, maxTokens: 50 }
+        });
+        const accounts = createMockAccounts(3);
+
+        const result = strategy.selectAccount(accounts, 'model');
+        assertNull(result.account, 'Should filter all accounts without tokens');
+    });
+
+    test('HybridStrategy: consumes token on selection', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 70 },
+            tokenBucket: { initialTokens: 10, maxTokens: 50 }
+        });
+        const accounts = createMockAccounts(1);
+
+        strategy.selectAccount(accounts, 'model');
+        const tracker = strategy.getTokenBucketTracker();
+        assertEqual(tracker.getTokens(accounts[0].email), 9, 'Token should be consumed');
+    });
+
+    test('HybridStrategy: onSuccess increases health', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 70, successReward: 5 }
+        });
+        const account = { email: 'test@example.com' };
+
+        strategy.onSuccess(account, 'model');
+        const tracker = strategy.getHealthTracker();
+        assertEqual(tracker.getScore('test@example.com'), 75, 'Health should increase');
+    });
+
+    test('HybridStrategy: onRateLimit decreases health', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 70, rateLimitPenalty: -10 }
+        });
+        const account = { email: 'test@example.com' };
+
+        strategy.onRateLimit(account, 'model');
+        const tracker = strategy.getHealthTracker();
+        assertEqual(tracker.getScore('test@example.com'), 60, 'Health should decrease');
+    });
+
+    test('HybridStrategy: onFailure decreases health and refunds token', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 70, failurePenalty: -20 },
+            tokenBucket: { initialTokens: 10, maxTokens: 50 }
+        });
+        const accounts = createMockAccounts(1);
+
+        // First consume a token
+        strategy.selectAccount(accounts, 'model');
+        const tokensBefore = strategy.getTokenBucketTracker().getTokens(accounts[0].email);
+
+        // Then fail
+        strategy.onFailure(accounts[0], 'model');
+
+        const healthTracker = strategy.getHealthTracker();
+        const tokenTracker = strategy.getTokenBucketTracker();
+
+        assertEqual(healthTracker.getScore(accounts[0].email), 50, 'Health should decrease by 20');
+        assertEqual(tokenTracker.getTokens(accounts[0].email), tokensBefore + 1, 'Token should be refunded');
+    });
+
+    test('HybridStrategy: scoring formula weights work correctly', () => {
+        // Test that health, tokens, and LRU all contribute to score
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 100 },
+            tokenBucket: { initialTokens: 50, maxTokens: 50 },
+            weights: { health: 2, tokens: 5, lru: 0.1 }
+        });
+
+        const accounts = [
+            { email: 'high-health@example.com', enabled: true, lastUsed: Date.now() },
+            { email: 'old-account@example.com', enabled: true, lastUsed: Date.now() - 3600000 }
+        ];
+
+        // Both have same health and tokens, but old-account has higher LRU
+        const result = strategy.selectAccount(accounts, 'model');
+        assertEqual(result.account.email, 'old-account@example.com', 'Older account should win with LRU weight');
+    });
+
+    // ==========================================================================
+    // STRATEGY FACTORY TESTS
+    // ==========================================================================
+    console.log('\n─── Strategy Factory Tests ───');
+
+    test('createStrategy: creates StickyStrategy for "sticky"', () => {
+        const strategy = createStrategy('sticky');
+        assertTrue(strategy instanceof StickyStrategy, 'Should create StickyStrategy');
+    });
+
+    test('createStrategy: creates RoundRobinStrategy for "round-robin"', () => {
+        const strategy = createStrategy('round-robin');
+        assertTrue(strategy instanceof RoundRobinStrategy, 'Should create RoundRobinStrategy');
+    });
+
+    test('createStrategy: creates RoundRobinStrategy for "roundrobin"', () => {
+        const strategy = createStrategy('roundrobin');
+        assertTrue(strategy instanceof RoundRobinStrategy, 'Should accept roundrobin alias');
+    });
+
+    test('createStrategy: creates HybridStrategy for "hybrid"', () => {
+        const strategy = createStrategy('hybrid');
+        assertTrue(strategy instanceof HybridStrategy, 'Should create HybridStrategy');
+    });
+
+    test('createStrategy: falls back to HybridStrategy for unknown strategy', () => {
+        const strategy = createStrategy('unknown');
+        assertTrue(strategy instanceof HybridStrategy, 'Should fall back to HybridStrategy');
+    });
+
+    test('createStrategy: uses default when null', () => {
+        const strategy = createStrategy(null);
+        assertTrue(strategy instanceof HybridStrategy, 'Null should use default HybridStrategy');
+    });
+
+    test('createStrategy: is case-insensitive', () => {
+        const s1 = createStrategy('STICKY');
+        const s2 = createStrategy('Hybrid');
+        const s3 = createStrategy('ROUND-ROBIN');
+        assertTrue(s1 instanceof StickyStrategy);
+        assertTrue(s2 instanceof HybridStrategy);
+        assertTrue(s3 instanceof RoundRobinStrategy);
+    });
+
+    test('isValidStrategy: returns true for valid strategies', () => {
+        assertTrue(isValidStrategy('sticky'));
+        assertTrue(isValidStrategy('round-robin'));
+        assertTrue(isValidStrategy('hybrid'));
+        assertTrue(isValidStrategy('roundrobin'));
+    });
+
+    test('isValidStrategy: returns false for invalid strategies', () => {
+        assertFalse(isValidStrategy('invalid'));
+        assertFalse(isValidStrategy(''));
+        assertFalse(isValidStrategy(null));
+        assertFalse(isValidStrategy(undefined));
+    });
+
+    test('getStrategyLabel: returns correct labels', () => {
+        assertEqual(getStrategyLabel('sticky'), 'Sticky (Cache Optimized)');
+        assertEqual(getStrategyLabel('round-robin'), 'Round Robin (Load Balanced)');
+        assertEqual(getStrategyLabel('roundrobin'), 'Round Robin (Load Balanced)');
+        assertEqual(getStrategyLabel('hybrid'), 'Hybrid (Smart Distribution)');
+    });
+
+    test('getStrategyLabel: returns default label for unknown', () => {
+        assertEqual(getStrategyLabel('unknown'), 'Hybrid (Smart Distribution)');
+        assertEqual(getStrategyLabel(null), 'Hybrid (Smart Distribution)');
+    });
+
+    test('STRATEGY_NAMES contains all valid strategies', () => {
+        assertDeepEqual(STRATEGY_NAMES, ['sticky', 'round-robin', 'hybrid']);
+    });
+
+    test('DEFAULT_STRATEGY is hybrid', () => {
+        assertEqual(DEFAULT_STRATEGY, 'hybrid');
+    });
+
+    // ==========================================================================
+    // INTEGRATION TESTS
+    // ==========================================================================
+    console.log('\n─── Integration Tests ───');
+
+    test('Integration: Hybrid strategy recovers from rate limits', () => {
+        const strategy = new HybridStrategy({
+            healthScore: { initial: 70, rateLimitPenalty: -10, minUsable: 50 }
+        });
+        const accounts = createMockAccounts(2);
+
+        // Rate limit first account multiple times
+        for (let i = 0; i < 3; i++) {
+            strategy.onRateLimit(accounts[0], 'model');
+        }
+
+        // Health of first account should be 40 (below minUsable 50)
+        const healthTracker = strategy.getHealthTracker();
+        assertEqual(healthTracker.getScore(accounts[0].email), 40);
+        assertFalse(healthTracker.isUsable(accounts[0].email));
+
+        // Selection should prefer second account
+        const result = strategy.selectAccount(accounts, 'model');
+        assertEqual(result.account.email, 'account2@example.com');
+    });
+
+    test('Integration: Token consumption limits requests', () => {
+        const strategy = new HybridStrategy({
+            tokenBucket: { initialTokens: 2, maxTokens: 10 }
+        });
+        const accounts = createMockAccounts(1);
+
+        // Consume all tokens
+        strategy.selectAccount(accounts, 'model'); // 2 -> 1
+        strategy.selectAccount(accounts, 'model'); // 1 -> 0
+
+        // Third request should fail (no tokens)
+        const result = strategy.selectAccount(accounts, 'model');
+        assertNull(result.account, 'Should return null when tokens exhausted');
+    });
+
+    test('Integration: Multi-model rate limiting is independent', () => {
+        const strategy = new StickyStrategy();
+        const accounts = createMockAccounts(2);
+
+        // Rate limit account1 for model-a only
+        accounts[0].modelRateLimits = {
+            'model-a': { isRateLimited: true, resetTime: Date.now() + 300000 }
+        };
+
+        // model-a should switch to account2
+        const resultA = strategy.selectAccount(accounts, 'model-a', { currentIndex: 0 });
+        assertEqual(resultA.account.email, 'account2@example.com');
+
+        // model-b should still use account1
+        const resultB = strategy.selectAccount(accounts, 'model-b', { currentIndex: 0 });
+        assertEqual(resultB.account.email, 'account1@example.com');
+    });
+
+    // Summary
+    console.log('\n' + '═'.repeat(60));
+    console.log(`Tests completed: ${passed} passed, ${failed} failed`);
+
+    if (failed > 0) {
+        process.exit(1);
+    }
+}
+
+runTests().catch(err => {
+    console.error('Test suite failed:', err);
+    process.exit(1);
+});