Use six settled Turns for product and benchmark replan defaults

apps/presentation/dashboard/src/features/personal-workspace/capability-localization.ts

@@ -159,7 +159,7 @@ const fieldCopy: Record<WorkspaceLocale, FieldCopy> = {
     executor_model: { label: "Model", description: "Optional model for the selected executor. Leave blank to keep the executor's own default." },
     executor_reasoning_effort: { label: "Reasoning effort", description: "Optional reasoning effort for the selected executor. Leave blank to keep the executor's own default." },
     count_unit: { label: "Review after", description: "Work Turns require accepted settlement; polls and retries do not count.", options: { completed_todos: "Completed Todos", effective_turns: "Settled work Turns" } },
-    count: { label: "Number between reviews", description: "From 1 to 5, using the selected unit. Goal overrides take precedence over device defaults." },
+    count: { label: "Number between reviews", description: "Settled work Turns: 1–6; completed Todos: 1–5. Goal overrides take precedence over device defaults." },
     allowed_domains: { label: "Allowed responsibility domains", description: "Enter one bounded, public-safe domain per line." },
     coordinator_agent_id: { label: "Coordinator Agent", description: "Use an already registered Agent id; leave blank to disable coordination." },
     enabled: { label: "Enabled" },
@@ -191,7 +191,7 @@ const fieldCopy: Record<WorkspaceLocale, FieldCopy> = {
     executor_model: { label: "模型", description: "所选执行器使用的模型，可留空；留空表示沿用执行器自身的默认模型。" },
     executor_reasoning_effort: { label: "推理档位", description: "所选执行器使用的推理档位，可留空；留空表示沿用执行器自身的默认档位。" },
     count_unit: { label: "复核计数依据", description: "有效工作 Turn 须完成结算；轮询和重复重试不计数。", options: { completed_todos: "已完成 Todo", effective_turns: "已结算工作 Turn" } },
-    count: { label: "两次复核间的数量", description: "按所选单位计数，范围 1–5；Goal 显式设置优先于设备默认值。" },
+    count: { label: "两次复核间的数量", description: "已结算工作 Turn 为 1–6，已完成 Todo 为 1–5；Goal 显式设置优先于设备默认值。" },
     allowed_domains: { label: "允许的职责域", description: "每行填写一个有边界、可公开的职责域。" },
     coordinator_agent_id: { label: "协调 Agent", description: "填写一个已经注册的 Agent ID；留空表示关闭协调。" },
     enabled: { label: "启用" },

benchmark/edgebench/README.md

@@ -368,9 +368,10 @@ and credentials belong outside the public repository.
 
 ## Default replan cadence
 
-New `heartbeat-resume` and `heartbeat-explore` trials replan after **3 settled
+New `heartbeat-resume` and `heartbeat-explore` trials replan after **6 settled
 effective work Turns** by default. Use `--replan-after-turns N` to set a different
-count (1–5), or `--replan-after-todos 3` for the previous completed-Todo cadence
+count (1–6); `--replan-after-turns 3` retains the previous benchmark default.
+Use `--replan-after-todos 3` for the previous completed-Todo cadence
 as an explicit ablation. The two options are mutually exclusive and require a
 heartbeat profile. This default is independent of `--task-entry`, feedback mode,
 and `--turn-envelope`. Official, single and native-goal profiles are unchanged.

benchmark/edgebench/run.py

@@ -109,8 +109,8 @@ def main(argv=None):
     parser.add_argument("--task-entry", choices=TASK_ENTRIES,
                         help="Heartbeat default: loopx-planned; seeded-todo is an explicit ablation")
     cadence = parser.add_mutually_exclusive_group()
-    cadence.add_argument("--replan-after-turns", type=int, choices=range(1, 6),
-                         help="Heartbeat default: 3 settled effective work Turns")
+    cadence.add_argument("--replan-after-turns", type=int, choices=range(1, 7),
+                         help="Heartbeat default: 6 settled effective work Turns")
     cadence.add_argument("--replan-after-todos", type=int, choices=range(1, 6),
                          help="Explicit completed-Todo cadence ablation for heartbeat profiles")
     parser.add_argument("--eval-interval", type=int,

benchmark/runtime/RUNTIME.md

@@ -30,7 +30,7 @@ agents:
       codex_sandbox: danger-full-access
       turn_timeout_sec: null
       scheduler_timeout_sec: 5080
-      replan_after_turns: 3
+      replan_after_turns: 6
 ```
 
 For an explicitly selected **heartbeat-only** context experiment, add
@@ -266,20 +266,25 @@ by selecting this option for a new run.
 ### Default effective-Turn cadence
 
 Harbor LoopX modes (`heartbeat`, `turn`, `loopx-goal`) default to
-`replan_after_turns: 3`; native EdgeBench `heartbeat-resume` and
-`heartbeat-explore` default to `--replan-after-turns 3`. This passes the existing Goal option
+`replan_after_turns: 6`; native EdgeBench `heartbeat-resume` and
+`heartbeat-explore` default to `--replan-after-turns 6`. This passes the existing Goal option
 `--execution-replan-after-turns` and verifies the persisted
 `replan_after_effective_turns` value before execution. The shared TypeScript
 control plane still owns which settled work Turns count; adapters do not count
 records or completed Todos themselves.
 
-Omitting both cadence options now selects three effective work Turns instead
-of three completed Todos. Idle wakes, tool calls and the planning checkpoint do
+Omitting both cadence options selects six effective work Turns, matching the
+product default. The previous benchmark default was three effective work Turns;
+the previous product default was five. Explicit settings retain their values.
+Idle wakes, tool calls and the planning checkpoint do
 not count as effective work Turns; this is a deterministic threshold rather than
 a per-wake probability, and other replan triggers can act sooner.
-To retain the old cadence in a new trial, pass `replan_after_todos: 3` in Harbor
-or `--replan-after-todos 3` in EdgeBench. Explicit Turn/Todo settings are mutually
-exclusive and accept counts from one through five. Resolved runtime and worker
+To retain the previous benchmark default in a new trial, pass
+`replan_after_turns: 3` in Harbor or `--replan-after-turns 3` in EdgeBench.
+For completed-Todo cadence, use `replan_after_todos` / `--replan-after-todos`.
+Explicit Turn/Todo settings are mutually
+exclusive; effective-Turn counts accept one through six and completed-Todo counts
+accept one through five. Resolved runtime and worker
 receipts name the selected unit even when no flag was supplied. Non-LoopX
 profiles retain their existing behavior. No active attempt, task, scoring, feedback,
 spawn permission, or total-budget change is implied.

benchmark/runtime/SETTINGS.md

@@ -1,8 +1,7 @@
 # Runner defaults and controlled ablations
 
 New LoopX executions default to **loopx-planned** task entry and replanning
-after **3 settled effective work Turns**, deliberately earlier than the product
-default of five. This applies to
+after **6 settled effective work Turns**, matching the product default. This applies to
 Harbor heartbeat, Turn and LoopX Goal modes, and EdgeBench heartbeat-resume and
 heartbeat-explore profiles. Planned entry replaces the seeded-todo default; pass
 `task_entry: seeded-todo` / `--task-entry seeded-todo` to retain the previous
@@ -22,16 +21,20 @@ use matched repetitions to test its effect.
 | Task entry | LoopX modes: loopx-planned | Record planned or seeded for every arm |
 | Explore | Off in heartbeat-resume; on in heartbeat-explore | Keep resume as the reference |
 | Turn envelope | Off | Enable only in its ablation |
-| Replan cadence | 3 settled effective work Turns | Use explicit completed-Todo cadence or another Turn count for ablation |
+| Replan cadence | 6 settled effective work Turns | Use explicit completed-Todo cadence or another Turn count for ablation |
 | Iteration context | Harbor: fresh; EdgeBench heartbeat: resume | Freeze the provider and context within a comparison |
 | Evaluator feedback | EdgeBench: best-only (strict new-best snapshot notifications) | Native and blind remain explicit controls; freeze feedback mode for every comparison |
 | Model and effort | Caller-selected | Pin both; never infer them from a profile name |
 | Time and sampling | EdgeBench task defaults in [task settings](../edgebench/README.md#trial-timeouts); explicit flags override | Pin resolved seconds in the study manifest |
 
 `replan_after_turns` counts settled effective work turns through the shared
-control-plane contract, not tool calls or idle heartbeat wakes. The threshold is 3 by
+control-plane contract, not tool calls or idle heartbeat wakes. The threshold is 6 by
 default, independent of task entry and TurnEnvelope. This is a deterministic
 threshold, not a random per-wake probability; other replan triggers may act sooner.
+An accepted replan resets the same Agent's periodic and repeated-progress history
+windows. A failed or unaccepted replan does not; current acceptance gaps remain
+independent triggers. Six reduces periodic interruptions relative to the previous
+default of three, but does not promise lower total replan time or better scores.
 Task timeouts bound attempts, not a requirement to consume every second.
 
 ## Recommended small study
@@ -41,11 +44,11 @@ These are recommendations, not automatically launched experiments.
 
 | Arm | EdgeBench flags relative to the reference | Question |
 | --- | --- | --- |
-| Reference | `--worker heartbeat-resume --task-entry loopx-planned --replan-after-turns 3` | Planned entry with effective-turn replanning |
+| Reference | `--worker heartbeat-resume --task-entry loopx-planned --replan-after-turns 6` | Planned entry with effective-turn replanning |
 | Seeded | Replace only `--task-entry` with `seeded-todo` | Does initial task decomposition help? |
 | Explore | Replace only `--worker` with `heartbeat-explore` | Are recorded evidence and subsequent route choices useful? |
 | Short envelope | Add `--turn-envelope` | Does progressive context loading reduce overhead without losing decisions? |
-| Todo cadence | Replace `--replan-after-turns 3` with `--replan-after-todos 3` | Does effective-turn cadence avoid postponing replans on long Todos? |
+| Todo cadence | Replace `--replan-after-turns 6` with `--replan-after-todos 3` | How do units and thresholds change interruptions? This is a combined ablation. |
 | No LoopX | `--worker official`; omit LoopX-specific flags | What is the net effect of the whole LoopX treatment? |
 
 The no-LoopX comparison changes several mechanisms; do not attribute its delta
@@ -64,7 +67,7 @@ python -m benchmark.edgebench.run \
   --task portfolio_risk_calibration --tasks-dir "$TASKS_DIR" \
   --log-dir "$RUNS_DIR" --run-id "$NEW_ATTEMPT_ID" \
   --worker heartbeat-resume --task-entry loopx-planned \
-  --replan-after-turns 3 --feedback best-only \
+  --replan-after-turns 6 --feedback best-only \
   --model "$MODEL" --effort xhigh --timeout 43200 --eval-interval 300 \
   --judge-url "$JUDGE_URL" --api-proxy-url "$API_PROXY_URL"
 ```
@@ -78,8 +81,10 @@ recorded service setting, not inherited from another task.
 
 Harbor uses the same task-entry owner. In the agent kwargs, set
 `execution_mode: heartbeat`, `task_entry: loopx-planned`,
-`replan_after_turns: 3`, `iteration_context: resume` and `turn_envelope: false`
-for an equivalent mechanism reference. For Todo cadence, remove
+`replan_after_turns: 6`, `iteration_context: resume` and `turn_envelope: false`
+for an equivalent mechanism reference. For the previous benchmark default, set
+`replan_after_turns: 3`.
+For Todo cadence, remove
 `replan_after_turns` and set `replan_after_todos: 3`; the two cadence fields are
 mutually exclusive. The named Explore profile in this matrix is EdgeBench-specific; do not assume
 an equivalent Harbor kwargs switch. Keep the dataset, provider and validation
@@ -98,6 +103,6 @@ cannot isolate an individual fix.
 Old attempts are immutable references. A rerun after multiple fixes measures
 the combined revision change unless each fix has a matched control. Roll back
 the entry default through explicit seeded entry and the cadence default through
-`--replan-after-todos 3` / `replan_after_todos: 3` in a new attempt. Do not rewrite
+`--replan-after-turns 3` / `replan_after_turns: 3` in a new attempt. Do not rewrite
 old receipts or change an active worker. None of these settings grants model,
 submission, credential or launch authority.

benchmark/runtime/codex.py

@@ -9,8 +9,8 @@
 from pathlib import Path
 
 
-# Benchmarks deliberately review earlier than the product default of five.
-DEFAULT_REPLAN_AFTER_TURNS = 3
+# New benchmark runs and the product both default to six settled work Turns.
+DEFAULT_REPLAN_AFTER_TURNS = 6
 
 
 MODES = ("plain", "native-goal", "heartbeat", "turn", "loopx-goal")

benchmark/runtime/harbor.py

@@ -84,8 +84,8 @@ def __init__(
             raise ValueError("Choose replan_after_turns or replan_after_todos, not both")
         if replan_after_turns is not None:
             if (type(replan_after_turns) is not int or
-                    not 1 <= replan_after_turns <= 5):
-                raise ValueError("replan_after_turns must be an integer between 1 and 5")
+                    not 1 <= replan_after_turns <= 6):
+                raise ValueError("replan_after_turns must be an integer between 1 and 6")
             if not self.execution.uses_loopx:
                 raise ValueError("replan_after_turns requires a LoopX execution mode")
         if replan_after_turns is None and replan_after_todos is None and self.execution.uses_loopx:

benchmark/runtime/sforge.py

@@ -118,8 +118,8 @@ def __init__(self, config, *, profile: str, cwd: str,
                 raise ValueError("replan_after_todos requires a heartbeat profile")
         if replan_after_turns is not None:
             if (type(replan_after_turns) is not int or
-                    not 1 <= replan_after_turns <= 5):
-                raise ValueError("replan_after_turns must be an integer between 1 and 5")
+                    not 1 <= replan_after_turns <= 6):
+                raise ValueError("replan_after_turns must be an integer between 1 and 6")
             if not profile.startswith("heartbeat-"):
                 raise ValueError("replan_after_turns requires a heartbeat profile")
         if (replan_after_turns is None and replan_after_todos is None

benchmark/tests/test_sforge_runtime.py

@@ -174,7 +174,7 @@ async def installed(self, environment):
     env = worker.runtime._worker_env(cwd="/task")
     assert float(env["LOOPX_CODEX_TURN_TIMEOUT_SEC"]) == expected
     assert worker.runtime.scheduler_timeout == total
-    expected_cadence = ({"replan_after_effective_turns": turns or 3}
+    expected_cadence = ({"replan_after_effective_turns": turns or 6}
                         if profile.startswith("heartbeat-") else
                         {"replan_after_completed_todos": 3})
     assert worker.runtime._replan_receipt() == expected_cadence
@@ -396,7 +396,7 @@ async def installed(self, environment):
     assert json.loads((tmp_path / "worker-profile.json").read_text())["task_entry"] == entry
 @pytest.mark.parametrize('profile', ['heartbeat-resume', 'heartbeat-explore'])
 @pytest.mark.parametrize('enabled', [False, True])
-@pytest.mark.parametrize('cadence', [None, 2])
+@pytest.mark.parametrize('cadence', [None, 2, 6])
 def test_envelope_treatment_reaches_shared_worker_and_receipts(tmp_path, monkeypatch, profile, enabled, cadence):
     pytest.importorskip('sforge')
     pytest.importorskip('harbor')
@@ -413,7 +413,7 @@ async def installed(self, environment):
     env = worker.runtime._worker_env(cwd='/task')
     assert env.get('LOOPX_TURN_ENVELOPE') == ('1' if enabled else None)
     assert worker.runtime.execution.turn_envelope is enabled
-    assert worker.runtime.replan_after_turns == (cadence or 3)
+    assert worker.runtime.replan_after_turns == (cadence or 6)
     receipt = json.loads((tmp_path / 'worker-profile.json').read_text())
     assert receipt['outer_resume'] is False
     assert receipt.get('turn_envelope') is (True if enabled else None)
@@ -462,7 +462,7 @@ def test_edgebench_rejects_invalid_profile_settings_before_creating_trial(tmp_pa
 
 
 @pytest.mark.parametrize("cadence_args,field,count", [
-    ([], "replan_after_effective_turns", 3),
+    ([], "replan_after_effective_turns", 6),
     (["--replan-after-turns", "2"], "replan_after_effective_turns", 2),
     (["--replan-after-todos", "3"], "replan_after_completed_todos", 3),
 ])
@@ -519,7 +519,7 @@ def stop_before_solver(**kwargs):
     assert receipt["status"] == "runner_failed"
 
 
-@pytest.mark.parametrize("value", [0, 6, True, 2.5, "3"])
+@pytest.mark.parametrize("value", [0, 7, True, 2.5, "3"])
 def test_effective_turn_cadence_rejects_invalid_values_before_install(tmp_path, monkeypatch, value):
     pytest.importorskip("sforge")
     pytest.importorskip("harbor")

benchmark/tests/test_shared_codex_runtime.py

@@ -359,12 +359,12 @@ def test_baseline_and_treatment_use_same_harbor_entry(tmp_path):
         assert env["LOOPX_EXECUTION_MODE"] == mode
         assert env["LOOPX_PROJECT"] == "/workspace"
         assert env["MODEL_NAME"] == "fixture"
-        assert agent.replan_after_turns == (3 if agent.execution.uses_loopx else None)
+        assert agent.replan_after_turns == (6 if agent.execution.uses_loopx else None)
 
 
 @pytest.mark.parametrize("existing", [False, True])
 @pytest.mark.parametrize("settings,field,value", [
-    ({}, "replan_after_effective_turns", 3),
+    ({}, "replan_after_effective_turns", 6),
     ({"replan_after_turns": 2}, "replan_after_effective_turns", 2),
     ({"replan_after_todos": 3}, "replan_after_completed_todos", 3),
 ])