diff --git a/apps/api/src/modules/activity-registry/__tests__/acceptance-predicate.test.ts b/apps/api/src/modules/activity-registry/__tests__/acceptance-predicate.test.ts
index 494c2fd..ed2597a 100644
--- a/apps/api/src/modules/activity-registry/__tests__/acceptance-predicate.test.ts
+++ b/apps/api/src/modules/activity-registry/__tests__/acceptance-predicate.test.ts
@@ -517,4 +517,26 @@ describe('配方静态审:写错地址 = 每一单都错,装的时候就拒', ()
     const findings = autoRecipeFindings(withRecipe({ do: [{ kind: 'submit', rationale: 'x' }] }));
     expect(findings.some((f) => f.severity === 'error' && f.message.includes('验收清单'))).toBe(true);
   });
+
+  /*
+   * **card 后面还有笔 → warn**(2026-08-29 终审 I3 附带):卡真发出去那一拍到此为止,
+   * 排在后面的笔要等人交回之后的那一拍才跑。写配方的人常按顺序读成「铺格 → 发卡 → 收口整理」
+   * 一拍连做 —— 那是读错的,装的时候提前说一声。
+   */
+  it('card 后面还有笔 → warn(它们要等人交回之后的那一拍才跑)', () => {
+    const findings = autoRecipeFindings(
+      withRecipe({
+        do: [
+          { kind: 'card', as: 'execution', title: 'x' },
+          { kind: 'project', into: 'demo.record', from: { once: true }, fields: { case_ref: 'a' } },
+        ],
+      }),
+    );
+    expect(findings.some((f) => f.severity === 'warn' && f.message.includes('card 后面还有 1 笔'))).toBe(true);
+  });
+
+  it('card 是最后一笔:不出这条 warn', () => {
+    const findings = autoRecipeFindings(withRecipe({ do: [{ kind: 'card', as: 'execution', title: 'x' }] }));
+    expect(findings.some((f) => f.message.includes('card 后面还有'))).toBe(false);
+  });
 });
diff --git a/apps/api/src/modules/activity-registry/internal/domain/static-audit.ts b/apps/api/src/modules/activity-registry/internal/domain/static-audit.ts
index b648521..30e23cb 100644
--- a/apps/api/src/modules/activity-registry/internal/domain/static-audit.ts
+++ b/apps/api/src/modules/activity-registry/internal/domain/static-audit.ts
@@ -471,6 +471,22 @@ export function autoRecipeFindings(manifest: ActivityManifest): StaticAuditFindi
         if (action.ask !== undefined) walkExpr(action.ask, where);
         if (action.standard !== undefined) walkExpr(action.standard, where);
         for (const b of action.basis ?? []) walkExpr(b.v, where);
+        /*
+         * **card 后面还有笔:说清它们什么时候跑**(2026-08-29 终审 I3 附带的一条 warn)。
+         * 卡真发出去那一拍到此为止(`stop`);排在后面的笔要等**人交回之后**的那一拍才跑
+         * (那一拍这一笔 `card` 因为 `carded` 被跳过)。写配方的人常按顺序读成
+         * 「铺格 → 发卡 → 收口整理」一拍连做 —— 那是读错的,提前说一声。
+         */
+        if (index < recipe.do.length - 1) {
+          out.push({
+            code: 'schema-invalid',
+            severity: 'warn',
+            message:
+              `card 后面还有 ${recipe.do.length - index - 1} 笔:卡发出去那一拍到此为止,` +
+              '这几笔要等人交回之后的那一拍才跑(那一拍 card 这一笔被跳过)—— 确认这就是你要的次序',
+            at: where,
+          });
+        }
       }
       /* `say.text` 同样是 AutoExpr,同一个理由要核(2026-08-29 评审补) */
       if (action.kind === 'say') {
diff --git a/apps/api/src/modules/loop-engine/__tests__/auto-runner.test.ts b/apps/api/src/modules/loop-engine/__tests__/auto-runner.test.ts
index a2b0311..d52e9b5 100644
--- a/apps/api/src/modules/loop-engine/__tests__/auto-runner.test.ts
+++ b/apps/api/src/modules/loop-engine/__tests__/auto-runner.test.ts
@@ -581,3 +581,45 @@ describe('createOnly + 配方:已在库的当「不用建」跳过,不抛(I2)',
     expect([...drafts.values()].find((d) => d.key['internal_pn'] === 'IN_LIB')?.landed).toBe(true);
   });
 });
+
+/**
+ * **`carded` 只跳过这一笔,不收拍**(2026-08-29 终审 I3)。
+ *
+ * 从前 carded 分支 `stop: true` → break 掉整条 `do`。同一份配方 `[project, card, project]`
+ * 在两种局面下行为相反:人还在填(quiet)时 card 被 quiet 过滤 `continue` 掉,**第三笔照跑**;
+ * 人交回之后(非 quiet、carded)走到 carded 分支 break,**第三笔不跑** ——
+ * 唯一一个「人的答案已经在草稿里」的时刻,配方的后半截偏偏不执行,而且静默、没有痕。
+ */
+describe('carded:跳这一笔,不收拍(I3)', () => {
+  const THREE: AutoRecipe = {
+    do: [
+      { kind: 'project', into: 'sub_eval.part', from: { once: true }, fields: { eval_ref: 'W05-1', internal_pn: 'A' } },
+      { kind: 'card', as: 'execution', title: '请补齐', basis: [{ k: 'a', v: 'b' }] },
+      { kind: 'project', into: 'parts.parameter', from: { once: true }, fields: { internal_pn: 'A', param_name: '收口', param_value: '1' } },
+    ],
+  };
+
+  it('人交回之后那一拍:卡不再发,**卡后面那一笔照跑**', async () => {
+    const { runner, drafts, cards, ctx } = harness({});
+    const out = await runner.run(THREE, { ...ctx, carded: true });
+    expect(cards.issued).toHaveLength(0);
+    expect(out.issuedCardId).toBeUndefined();
+    expect([...drafts.values()].map((d) => d.object).sort()).toEqual(['parameter', 'part']);
+  });
+
+  it('卡真发出去那一拍仍然收拍 —— 「收拍」留给发得出去的那一次', async () => {
+    const { runner, drafts, cards, ctx } = harness({});
+    const out = await runner.run(THREE, ctx);
+    expect(cards.issued).toHaveLength(1);
+    expect(out.issuedCardId).toBe('card-1');
+    // 第一笔写了,第三笔没跑
+    expect([...drafts.values()].map((d) => d.object)).toEqual(['part']);
+  });
+
+  it('人还在填(quiet)那一拍:card 被跳过,前后两笔都跑 —— 两种局面现在一致了', async () => {
+    const { runner, drafts, cards, ctx } = harness({});
+    await runner.run(THREE, { ...ctx, quiet: true, carded: true });
+    expect(cards.issued).toHaveLength(0);
+    expect([...drafts.values()].map((d) => d.object).sort()).toEqual(['parameter', 'part']);
+  });
+});
diff --git a/apps/api/src/modules/loop-engine/internal/usecase/auto-runner.ts b/apps/api/src/modules/loop-engine/internal/usecase/auto-runner.ts
index 25302db..7e8b337 100644
--- a/apps/api/src/modules/loop-engine/internal/usecase/auto-runner.ts
+++ b/apps/api/src/modules/loop-engine/internal/usecase/auto-runner.ts
@@ -211,13 +211,22 @@ export function createAutoRunner(deps: AutoRunnerDeps): {
            * 不必再有一个「先 start」的礼仪:学不稳的礼仪就是死法(2026-08-02 已拆过一次)。
            */
           if (context.carded) {
+            /*
+             * **只跳过这一笔,不收拍**(2026-08-29 终审 I3)。从前这里 `stop: true`,
+             * 于是同一份配方 `[project, card, project]` 在两种局面下行为相反:
+             * 人还在填(quiet)时 card 被上面那道 quiet 过滤 `continue` 掉 → **第三笔照跑**;
+             * 人交回之后(非 quiet、carded)→ 走到这儿 break,**第三笔不跑**。
+             * 也就是说唯一一个「人的答案已经在草稿里」的时刻,配方的后半截偏偏不执行 ——
+             * 写配方的人会很自然地写「铺格 → 发卡 → 收口整理」,然后发现第三笔从来没生效过,
+             * 而且静默、没有痕。**「收拍」该留给真发出去了卡那一次**(下面两条 return 带 stop)。
+             */
             await deps.memory.append({
               taskId: context.taskId,
               kind: 'observation',
               title: `这一步已经发过卡,不再发第二张`,
               detail: { auto: `${context.stepId}#${index + 1}`, internal: true },
             });
-            return { wrote: 0, dropped: 0, stop: true };
+            return { wrote: 0, dropped: 0 };
           }
           const basis = (action.basis ?? []).map((b) => ({ k: b.k, v: evalText(b.v, base) }));
           if (action.as === 'decide') {