diff --git a/CLAUDE.md b/CLAUDE.md
index 57c981f..eb957a8 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -79,10 +79,12 @@ score = P(self) + Σ P(completed deps) + Σ P(tasks blocked by self)
 
 | Role | Easy | Medium | Hard |
 |---|---|---|---|
-| executor | claude-sonnet-4-6 | claude-opus-4-8 | claude-fable-5 |
-| reviewer | claude-sonnet-4-6 | claude-opus-4-8 | claude-opus-4-8 |
+| executor | claude-sonnet-4-6 | claude-opus-4-8 | claude-opus-4-8 |
+| planner | claude-sonnet-4-6 | claude-opus-4-8 | claude-fable-5 |
+| reviewer | claude-fable-5 | claude-fable-5 | claude-fable-5 |
+| conflict | claude-fable-5 | claude-fable-5 | claude-fable-5 |
 
-Project-level `model` setting overrides env vars; env vars override built-in defaults.
+Reviewer/conflict use the strongest tier (fable-5) and are NOT downgraded by project-level `model` (preserves self-review/conflict-resolution rigor). Project-level `model` overrides executor/planner env vars; env vars override built-in defaults. Model fallback chain on unavailability: `fable-5 → opus-4-8 → sonnet-4-6`.
 
 ### Data Directory
 
diff --git a/src/executor/models.ts b/src/executor/models.ts
index 8671a71..ffbfaef 100644
--- a/src/executor/models.ts
+++ b/src/executor/models.ts
@@ -4,8 +4,8 @@ import type { Complexity } from '../model/complexity.js';
 /** 角色：executor（执行改动）/ reviewer（执行后复审）/ planner（拆解 Hard / 写方案 Medium）/ conflict（冲突裁决） */
 export type ModelRole = 'executor' | 'reviewer' | 'planner' | 'conflict';
 
-/** 模型不可用时的回退链（按序取下一档，同一 run 内只重试一次） */
-export const MODEL_FALLBACK_CHAIN = ['claude-opus-4-8', 'claude-sonnet-4-6'] as const;
+/** 模型不可用时的回退链（按序取下一档，同一 run 内只重试一次；fable 置链首=优先用最强档） */
+export const MODEL_FALLBACK_CHAIN = ['claude-fable-5', 'claude-opus-4-8', 'claude-sonnet-4-6'] as const;
 
 /** 复杂度 → [env 覆盖变量名, 默认模型] */
 const MODEL_TABLE: Record<ModelRole, Record<Complexity, [env: string, fallback: string]>> = {
@@ -14,22 +14,23 @@ const MODEL_TABLE: Record<ModelRole, Record<Complexity, [env: string, fallback:
     medium: ['MAESTRO_MODEL_MEDIUM', 'claude-opus-4-8'],
     hard: ['MAESTRO_MODEL_HARD', 'claude-opus-4-8'],
   },
+  // reviewer：复审是质量关口，统一用最强 fable-5（不受 project.model 降档，保自审严格度）
   reviewer: {
-    easy: ['MAESTRO_MODEL_REVIEW_EASY', 'claude-opus-4-8'],
-    medium: ['MAESTRO_MODEL_REVIEW_MEDIUM', 'claude-opus-4-8'],
-    hard: ['MAESTRO_MODEL_REVIEW_HARD', 'claude-opus-4-8'],
+    easy: ['MAESTRO_MODEL_REVIEW_EASY', 'claude-fable-5'],
+    medium: ['MAESTRO_MODEL_REVIEW_MEDIUM', 'claude-fable-5'],
+    hard: ['MAESTRO_MODEL_REVIEW_HARD', 'claude-fable-5'],
   },
-  // planner：拆解/写方案按复杂度分化（不受 project.model 覆盖）
+  // planner：拆解/写方案按复杂度分化（不受 project.model 覆盖）；hard 拆解最烧脑→fable-5
   planner: {
     easy: ['MAESTRO_MODEL_PLAN_EASY', 'claude-sonnet-4-6'],
     medium: ['MAESTRO_MODEL_PLAN_MEDIUM', 'claude-opus-4-8'],
-    hard: ['MAESTRO_MODEL_PLAN_HARD', 'claude-opus-4-8'],
+    hard: ['MAESTRO_MODEL_PLAN_HARD', 'claude-fable-5'],
   },
-  // conflict：固定 opus-4-8，TABLE 条目仅占位（pickModel 对此角色直接 return 不走此处）
+  // conflict：固定 fable-5（解冲突最难，恒用最强档），TABLE 条目仅占位（pickModel 对此角色直接 return）
   conflict: {
-    easy: ['', 'claude-opus-4-8'],
-    medium: ['', 'claude-opus-4-8'],
-    hard: ['', 'claude-opus-4-8'],
+    easy: ['', 'claude-fable-5'],
+    medium: ['', 'claude-fable-5'],
+    hard: ['', 'claude-fable-5'],
   },
 };
 
@@ -44,7 +45,7 @@ const MODEL_TABLE: Record<ModelRole, Record<Complexity, [env: string, fallback:
 export function pickModel(task: Task, project: Project, role: ModelRole): string {
   // reviewer/conflict 用最强档，不受 project.model 覆盖（保自审独立性）
   if (role !== 'reviewer' && role !== 'conflict' && project.model) return project.model;
-  if (role === 'conflict') return 'claude-opus-4-8';
+  if (role === 'conflict') return 'claude-fable-5';
   const [env, fallback] = MODEL_TABLE[role][task.complexity];
   const override = process.env[env]?.trim();
   return override || fallback;
diff --git a/test/models.test.ts b/test/models.test.ts
index b754e03..f33266b 100644
--- a/test/models.test.ts
+++ b/test/models.test.ts
@@ -40,10 +40,14 @@ test('pickModel：复杂度映射默认值（executor / reviewer）', () => {
     assert.equal(pickModel(fakeTask('easy'), p, 'executor'), 'claude-sonnet-4-6');
     assert.equal(pickModel(fakeTask('medium'), p, 'executor'), 'claude-opus-4-8');
     assert.equal(pickModel(fakeTask('hard'), p, 'executor'), 'claude-opus-4-8');
-    // reviewer 三档统一 opus-4-8（保复审独立性）
-    assert.equal(pickModel(fakeTask('easy'), p, 'reviewer'), 'claude-opus-4-8');
-    assert.equal(pickModel(fakeTask('medium'), p, 'reviewer'), 'claude-opus-4-8');
-    assert.equal(pickModel(fakeTask('hard'), p, 'reviewer'), 'claude-opus-4-8');
+    // reviewer 三档统一 fable-5（复审是质量关口，用最强档保严格度）
+    assert.equal(pickModel(fakeTask('easy'), p, 'reviewer'), 'claude-fable-5');
+    assert.equal(pickModel(fakeTask('medium'), p, 'reviewer'), 'claude-fable-5');
+    assert.equal(pickModel(fakeTask('hard'), p, 'reviewer'), 'claude-fable-5');
+    // planner hard 拆解最烧脑 → fable-5；medium=opus；easy=sonnet
+    assert.equal(pickModel(fakeTask('hard'), p, 'planner'), 'claude-fable-5');
+    assert.equal(pickModel(fakeTask('medium'), p, 'planner'), 'claude-opus-4-8');
+    assert.equal(pickModel(fakeTask('easy'), p, 'planner'), 'claude-sonnet-4-6');
   });
 });
 
@@ -53,10 +57,10 @@ test('pickModel：project.model executor 最优先，但 reviewer/conflict 不
     for (const c of ['easy', 'medium', 'hard'] as const) {
       // executor 受 project.model 覆盖
       assert.equal(pickModel(fakeTask(c), p, 'executor'), 'claude-opus-4-6');
-      // reviewer 不受 project.model 覆盖，始终用 opus-4-8（保自审独立性）
-      assert.equal(pickModel(fakeTask(c), p, 'reviewer'), 'claude-opus-4-8');
-      // conflict 固定 opus-4-8
-      assert.equal(pickModel(fakeTask(c), p, 'conflict'), 'claude-opus-4-8');
+      // reviewer 不受 project.model 覆盖，始终用 fable-5（保自审独立性）
+      assert.equal(pickModel(fakeTask(c), p, 'reviewer'), 'claude-fable-5');
+      // conflict 固定 fable-5（解冲突恒用最强档）
+      assert.equal(pickModel(fakeTask(c), p, 'conflict'), 'claude-fable-5');
     }
   });
 });
@@ -70,8 +74,8 @@ test('pickModel：env 覆盖默认值，executor/reviewer 各用各的 env', ()
     assert.equal(pickModel(fakeTask('easy'), p, 'executor'), 'claude-haiku-4-5');
     // reviewer easy 被 MAESTRO_MODEL_REVIEW_EASY 覆盖
     assert.equal(pickModel(fakeTask('easy'), p, 'reviewer'), 'claude-sonnet-4-6');
-    // reviewer hard 无 env → 默认 opus-4-8
-    assert.equal(pickModel(fakeTask('hard'), p, 'reviewer'), 'claude-opus-4-8');
+    // reviewer hard 无 env → 默认 fable-5
+    assert.equal(pickModel(fakeTask('hard'), p, 'reviewer'), 'claude-fable-5');
     assert.equal(pickModel(fakeTask('hard'), p, 'executor'), 'claude-opus-4-8');
   });
 });
@@ -98,8 +102,9 @@ test('isModelError：模型不可用类错误才触发回退', () => {
 });
 
 test('pickFallbackModel：回退链取下一档（循环）', () => {
-  // 链：opus → sonnet → opus（循环）
-  assert.deepEqual([...MODEL_FALLBACK_CHAIN], ['claude-opus-4-8', 'claude-sonnet-4-6']);
+  // 链：fable → opus → sonnet → fable（循环，fable 置链首=优先最强档）
+  assert.deepEqual([...MODEL_FALLBACK_CHAIN], ['claude-fable-5', 'claude-opus-4-8', 'claude-sonnet-4-6']);
+  assert.equal(pickFallbackModel('claude-fable-5'), 'claude-opus-4-8');
   assert.equal(pickFallbackModel('claude-opus-4-8'), 'claude-sonnet-4-6');
-  assert.equal(pickFallbackModel('claude-sonnet-4-6'), 'claude-opus-4-8');
+  assert.equal(pickFallbackModel('claude-sonnet-4-6'), 'claude-fable-5');
 });