From 100654adb8a3fa2cf4d09ff6c85fc006806d9844 Mon Sep 17 00:00:00 2001
From: Chenyme <118253778+chenyme@users.noreply.github.com>
Date: Mon, 20 Apr 2026 17:45:22 +0800
Subject: [PATCH 1/2] feat: Introduce random selection strategy and cooldown
 configuration for account management

---
 app/dataplane/account/__init__.py  |  44 ++++-
 app/dataplane/account/feedback.py  |  94 ++++++---
 app/dataplane/account/selector.py  | 301 ++++++++++++++++++++---------
 app/dataplane/account/table.py     |   6 +
 app/dataplane/shared/enums.py      |   3 +
 app/main.py                        |  25 ++-
 app/products/_account_selection.py |  25 ++-
 app/products/anthropic/messages.py |   4 +-
 app/products/openai/chat.py        |   4 +-
 app/products/openai/responses.py   |   4 +-
 app/products/web/admin/__init__.py |   2 +
 app/statics/admin/account.html     |  34 +++-
 app/statics/admin/config.html      | 167 ++++++++++++++++
 app/statics/i18n/de.json           |  17 +-
 app/statics/i18n/en.json           |  23 +++
 app/statics/i18n/es.json           |  17 +-
 app/statics/i18n/fr.json           |  17 +-
 app/statics/i18n/ja.json           |  17 +-
 app/statics/i18n/zh.json           |  23 +++
 config.defaults.toml               |  14 ++
 20 files changed, 687 insertions(+), 154 deletions(-)

diff --git a/app/dataplane/account/__init__.py b/app/dataplane/account/__init__.py
index b753edcb3..20937f366 100644
--- a/app/dataplane/account/__init__.py
+++ b/app/dataplane/account/__init__.py
@@ -8,16 +8,17 @@
 import asyncio
 from typing import TYPE_CHECKING
 
+from app.platform.config.snapshot import get_config
 from app.platform.logging.logger import logger
 from app.platform.runtime.clock import now_s
 from app.control.account.repository import AccountRepository
 from app.control.account.enums import FeedbackKind
 from .table import AccountRuntimeTable
 from .lease import AccountLease, new_lease
-from .selector import select, select_any
+from .selector import current_strategy, select, select_any
 from .sync import bootstrap as _bootstrap, apply_changes
 from . import feedback as fb
-from ..shared.enums import StatusId
+from ..shared.enums import POOL_ID_TO_STR, StatusId
 
 if TYPE_CHECKING:
     pass
@@ -250,12 +251,22 @@ async def feedback(
 
         ts = now_s_val if now_s_val is not None else now_s()
 
+        strategy = current_strategy()
+
         async with self._lock:
             if kind == FeedbackKind.SUCCESS:
-                fb.apply_success(table, idx, mode_id)
+                if strategy == "random":
+                    fb.apply_success_random(table, idx)
+                else:
+                    fb.apply_success_quota(table, idx, mode_id)
 
             elif kind == FeedbackKind.RATE_LIMITED:
-                fb.apply_rate_limited(table, idx, mode_id)
+                if strategy == "random":
+                    pool_id = int(table.pool_by_idx[idx])
+                    cooling_sec = _pool_cooling_sec(pool_id)
+                    fb.apply_rate_limited_random(table, idx, cooling_sec=cooling_sec)
+                else:
+                    fb.apply_rate_limited_quota(table, idx, mode_id)
                 fb.update_last_fail(table, idx, ts)
 
             elif kind == FeedbackKind.UNAUTHORIZED:
@@ -271,8 +282,9 @@ async def feedback(
                 fb.apply_server_error(table, idx)
                 fb.update_last_fail(table, idx, ts)
 
-            # Apply authoritative quota data from upstream response headers.
-            if remaining is not None and reset_at_ms is not None:
+            # Quota strategy may receive authoritative quota data from upstream
+            # response headers; the random strategy ignores this entirely.
+            if strategy == "quota" and remaining is not None and reset_at_ms is not None:
                 reset_s = int(reset_at_ms // 1000)
                 fb.apply_quota_update(table, idx, mode_id, remaining, reset_s)
 
@@ -289,6 +301,26 @@ def revision(self) -> int:
         return self._table.revision if self._table else 0
 
 
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+_DEFAULT_COOLING_HOURS: dict[str, int] = {
+    "basic": 20,
+    "super": 2,
+    "heavy": 2,
+}
+
+
+def _pool_cooling_sec(pool_id: int) -> int:
+    """Cooling seconds for a 429 on a given pool (random strategy only)."""
+    pool_str = POOL_ID_TO_STR.get(pool_id, "basic")
+    default_hours = _DEFAULT_COOLING_HOURS.get(pool_str, 2)
+    hours = int(get_config(f"account.random.cooling_{pool_str}_hours", default_hours))
+    return max(0, hours) * 3600
+
+
 # ---------------------------------------------------------------------------
 # Module-level singleton
 # ---------------------------------------------------------------------------
diff --git a/app/dataplane/account/feedback.py b/app/dataplane/account/feedback.py
index b77942c1e..3372b56e3 100644
--- a/app/dataplane/account/feedback.py
+++ b/app/dataplane/account/feedback.py
@@ -1,39 +1,78 @@
 """Apply feedback to the runtime table columns (in-place, lock-free inner ops).
 
-The caller (AccountDirectory) is responsible for holding the state lock
-before calling these functions.
+The caller (AccountDirectory) is responsible for holding the state lock before
+calling these functions.
+
+Strategy-split functions
+------------------------
+``apply_success`` / ``apply_rate_limited`` are split into ``*_quota`` and
+``*_random`` variants. The caller picks exactly one based on the active
+selector strategy. No runtime ``if`` inside the apply helpers themselves.
 """
 
+from app.platform.runtime.clock import now_s
 from ..shared.enums import ALL_MODE_IDS, StatusId
 from .table import AccountRuntimeTable
 
 # Health adjustment constants.
-_SUCCESS_STEP = 0.12
-_AUTH_FACTOR = 0.55
-_FORBIDDEN_FACTOR = 0.25
-_RATE_LIMIT_FACTOR = 0.45
+_SUCCESS_STEP        = 0.12
+_AUTH_FACTOR         = 0.55
+_FORBIDDEN_FACTOR    = 0.25
+_RATE_LIMIT_FACTOR   = 0.45
 _SERVER_ERROR_FACTOR = 0.75
 _MIN_HEALTH = 0.05
 _MAX_HEALTH = 1.0
 
 
-def apply_success(table: AccountRuntimeTable, idx: int, mode_id: int) -> None:
-    """Record a successful call: decrement quota, improve health."""
+# ---------------------------------------------------------------------------
+# apply_success — two independent implementations
+# ---------------------------------------------------------------------------
+
+
+def apply_success_quota(table: AccountRuntimeTable, idx: int, mode_id: int) -> None:
+    """Quota strategy: decrement per-mode quota and improve health."""
     quota_col = table._quota_col(mode_id)
-    new_q = max(0, int(quota_col[idx]) - 1)
-    quota_col[idx] = new_q
+    quota_col[idx] = max(0, int(quota_col[idx]) - 1)
+    _bump_health(table, idx)
+
 
-    # Recover health.
-    h = min(_MAX_HEALTH, float(table.health_by_idx[idx]) + _SUCCESS_STEP)
-    table.health_by_idx[idx] = h
+def apply_success_random(table: AccountRuntimeTable, idx: int) -> None:
+    """Random strategy: only improve health; quota columns are ignored."""
+    _bump_health(table, idx)
 
 
-def apply_rate_limited(table: AccountRuntimeTable, idx: int, mode_id: int) -> None:
-    """Zero the mode quota and reduce health."""
+# ---------------------------------------------------------------------------
+# apply_rate_limited — two independent implementations
+# ---------------------------------------------------------------------------
+
+
+def apply_rate_limited_quota(
+    table: AccountRuntimeTable, idx: int, mode_id: int
+) -> None:
+    """Quota strategy: zero the mode quota and reduce health."""
     table._quota_col(mode_id)[idx] = 0
     _adjust_health(table, idx, _RATE_LIMIT_FACTOR)
 
 
+def apply_rate_limited_random(
+    table: AccountRuntimeTable, idx: int, *, cooling_sec: int
+) -> None:
+    """Random strategy: set per-account cooldown timestamp; do not touch quota.
+
+    The account stays in ``mode_available``; ``_random_select`` filters it out
+    until ``cooling_until_s_by_idx`` has elapsed.
+    """
+    ts = now_s() + max(0, cooling_sec)
+    cooling_col = table.cooling_until_s_by_idx
+    cooling_col[idx] = max(int(cooling_col[idx]), ts)
+    _adjust_health(table, idx, _RATE_LIMIT_FACTOR)
+
+
+# ---------------------------------------------------------------------------
+# Strategy-agnostic feedback helpers (shared)
+# ---------------------------------------------------------------------------
+
+
 def apply_auth_failure(table: AccountRuntimeTable, idx: int) -> None:
     """Reduce health on 401; caller may mark account expired."""
     _adjust_health(table, idx, _AUTH_FACTOR)
@@ -62,13 +101,11 @@ def apply_status_change(
     table.status_by_idx[idx] = new_status_id
 
     if new_status_id != int(StatusId.ACTIVE):
-        # Remove from all mode availability buckets.
         for mode_id in ALL_MODE_IDS:
             bucket = table.mode_available.get((pool_id, mode_id))
             if bucket:
                 bucket.discard(idx)
     else:
-        # Re-add to mode buckets where quota > 0.
         for mode_id in ALL_MODE_IDS:
             if int(table._quota_col(mode_id)[idx]) > 0:
                 table.mode_available.setdefault((pool_id, mode_id), set()).add(idx)
@@ -81,7 +118,11 @@ def apply_quota_update(
     remaining: int,
     reset_s: int,
 ) -> None:
-    """Update quota and reset timestamp from upstream API data."""
+    """Update quota and reset timestamp from upstream API data.
+
+    Used exclusively by the quota strategy (refresh service). The random
+    strategy never calls this.
+    """
     quota_col = table._quota_col(mode_id)
     reset_col = table._reset_col(mode_id)
     quota_col[idx] = max(0, min(remaining, 32767))
@@ -116,14 +157,23 @@ def update_last_fail(table: AccountRuntimeTable, idx: int, now_s: int) -> None:
 # ---------------------------------------------------------------------------
 
 
+def _bump_health(table: AccountRuntimeTable, idx: int) -> None:
+    table.health_by_idx[idx] = min(
+        _MAX_HEALTH, float(table.health_by_idx[idx]) + _SUCCESS_STEP
+    )
+
+
 def _adjust_health(table: AccountRuntimeTable, idx: int, factor: float) -> None:
-    h = max(_MIN_HEALTH, float(table.health_by_idx[idx]) * factor)
-    table.health_by_idx[idx] = h
+    table.health_by_idx[idx] = max(
+        _MIN_HEALTH, float(table.health_by_idx[idx]) * factor
+    )
 
 
 __all__ = [
-    "apply_success",
-    "apply_rate_limited",
+    "apply_success_quota",
+    "apply_success_random",
+    "apply_rate_limited_quota",
+    "apply_rate_limited_random",
     "apply_auth_failure",
     "apply_forbidden",
     "apply_server_error",
diff --git a/app/dataplane/account/selector.py b/app/dataplane/account/selector.py
index 814259049..c7d048cbe 100644
--- a/app/dataplane/account/selector.py
+++ b/app/dataplane/account/selector.py
@@ -1,22 +1,63 @@
-"""Hot-path account selector — pure function, no object allocation.
+"""Hot-path account selector — pluggable strategies.
 
-All scoring operates on typed ``array.array`` columns; no attribute lookup
-overhead.  The caller provides the frozen set of excluded indices and
-pre-resolved tag index set.
+Two fully independent strategies:
+
+* ``_quota_select`` — scores candidates by health / quota / inflight / fails.
+  Used when ``account.refresh.enabled=true``. Behaviour is the historical one,
+  kept unchanged.
+* ``_random_select`` — uniform random choice among non-cooling candidates.
+  Used when ``account.refresh.enabled=false``. Ignores quota and health entirely.
+
+Strategy selection is process-global, registered once by the lifespan via
+:func:`set_strategy`. Callers (``AccountDirectory``) invoke :func:`select` /
+:func:`select_any` and are unaware of which strategy is active.
 """
 
+import array  # noqa: F401  — used in forward-referenced type annotations
+import random
+from typing import Literal
+
+from app.platform.config.snapshot import get_config
 from ..shared.enums import PoolId
 from .table import AccountRuntimeTable
 
-# Scoring weights — tuned for throughput/fairness balance.
-_W_HEALTH  = 100.0
-_W_QUOTA   = 25.0
-_W_RECENT  = 15.0     # penalty for recently used accounts
+# Scoring weights used by the quota strategy.
+_W_HEALTH   = 100.0
+_W_QUOTA    = 25.0
+_W_RECENT   = 15.0     # penalty for recently used accounts
 _W_INFLIGHT = 20.0
-_W_FAIL    = 4.0
+_W_FAIL     = 4.0
 _RECENT_WINDOW_S = 15  # seconds
 
 
+# ---------------------------------------------------------------------------
+# Strategy registry
+# ---------------------------------------------------------------------------
+
+_StrategyName = Literal["quota", "random"]
+_STRATEGY_NAME: _StrategyName = "random"
+
+
+def set_strategy(name: _StrategyName) -> None:
+    """Register the process-global selection strategy.
+
+    Called once by the lifespan after reading ``account.refresh.enabled``.
+    """
+    global _STRATEGY_NAME
+    if name not in ("quota", "random"):
+        raise ValueError(f"unknown selection strategy: {name!r}")
+    _STRATEGY_NAME = name
+
+
+def current_strategy() -> _StrategyName:
+    return _STRATEGY_NAME
+
+
+# ---------------------------------------------------------------------------
+# Public entry points — delegate to the active strategy
+# ---------------------------------------------------------------------------
+
+
 def select(
     table: AccountRuntimeTable,
     pool_id: int,
@@ -26,30 +67,79 @@ def select(
     prefer_tag_idxs: set[int] | None    = None,
     now_s: int,
 ) -> int | None:
-    """Select the best available account slot index.
+    """Select an account slot for ``(pool_id, mode_id)``.
+
+    Returns the slot index or ``None`` when no candidate is available.
+    Does not mutate the table — callers increment inflight separately.
+    """
+    if _STRATEGY_NAME == "random":
+        return _random_select(
+            table, pool_id,
+            exclude_idxs=exclude_idxs,
+            prefer_tag_idxs=prefer_tag_idxs,
+            now_s=now_s,
+        )
+    return _quota_select(
+        table, pool_id, mode_id,
+        exclude_idxs=exclude_idxs,
+        prefer_tag_idxs=prefer_tag_idxs,
+        now_s=now_s,
+    )
+
+
+def select_any(
+    table: AccountRuntimeTable,
+    pool_id: int,
+    *,
+    exclude_idxs: frozenset[int] | None = None,
+    prefer_tag_idxs: set[int] | None    = None,
+    now_s: int,
+) -> int | None:
+    """Select any active account in ``pool_id`` irrespective of per-mode quota.
 
-    Returns the slot index on success, ``None`` if no account is available.
-    Does **not** mutate any table state — callers must increment inflight.
+    Used by WebSocket-based products that manage their own rate limiting.
     """
+    if _STRATEGY_NAME == "random":
+        return _random_select(
+            table, pool_id,
+            exclude_idxs=exclude_idxs,
+            prefer_tag_idxs=prefer_tag_idxs,
+            now_s=now_s,
+        )
+    return _quota_select_any(
+        table, pool_id,
+        exclude_idxs=exclude_idxs,
+        prefer_tag_idxs=prefer_tag_idxs,
+        now_s=now_s,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Strategy: quota — score-based selection (unchanged behaviour)
+# ---------------------------------------------------------------------------
+
+
+def _quota_select(
+    table: AccountRuntimeTable,
+    pool_id: int,
+    mode_id: int,
+    *,
+    exclude_idxs: frozenset[int] | None,
+    prefer_tag_idxs: set[int] | None,
+    now_s: int,
+) -> int | None:
     candidates: set[int] | None = table.mode_available.get((pool_id, mode_id))
     if not candidates:
         return None
 
-    # Apply window-expiry resets for basic accounts inline.
-    reset_col = table._reset_col(mode_id)
-    quota_col = table._quota_col(mode_id)
-    total_col = table._total_col(mode_id)
+    reset_col  = table._reset_col(mode_id)
+    quota_col  = table._quota_col(mode_id)
+    total_col  = table._total_col(mode_id)
     window_col = table._window_col(mode_id)
     _maybe_reset_windows(
-        table,
-        candidates,
-        mode_id,
-        reset_col,
-        quota_col,
-        total_col,
-        window_col,
-        pool_id,
-        now_s,
+        table, candidates, mode_id,
+        reset_col, quota_col, total_col, window_col,
+        pool_id, now_s,
     )
 
     working: set[int] = candidates.copy()
@@ -59,30 +149,50 @@ def select(
     if not working:
         return None
 
-    # Prefer tagged subset; fall back to full set if empty.
     if prefer_tag_idxs:
         preferred = working & prefer_tag_idxs
         working = preferred if preferred else working
 
-    return _best(table, working, mode_id, quota_col, now_s)
+    return _best(table, working, quota_col, now_s)
+
+
+def _quota_select_any(
+    table: AccountRuntimeTable,
+    pool_id: int,
+    *,
+    exclude_idxs: frozenset[int] | None,
+    prefer_tag_idxs: set[int] | None,
+    now_s: int,
+) -> int | None:
+    candidates: set[int] = _pool_union(table, pool_id)
+    if not candidates:
+        return None
+
+    working = candidates.copy()
+    if exclude_idxs:
+        working -= exclude_idxs
+    if not working:
+        return None
+
+    if prefer_tag_idxs:
+        preferred = working & prefer_tag_idxs
+        working = preferred if preferred else working
+
+    return _best_no_quota(table, working, now_s)
 
 
 def _maybe_reset_windows(
-    table:     AccountRuntimeTable,
+    table: AccountRuntimeTable,
     candidates: set[int],
-    mode_id:   int,
+    mode_id: int,
     reset_col: "array.array",
     quota_col: "array.array",
     total_col: "array.array",
     window_col: "array.array",
-    pool_id:   int,
-    now_s:     int,
+    pool_id: int,
+    now_s: int,
 ) -> None:
-    """Reset expired windows for basic-pool accounts (no API call required).
-
-    Only applies to the basic pool — super/heavy quotas are managed exclusively
-    by the periodic refresh service and must not be reset inline.
-    """
+    """Reset expired windows for basic-pool accounts inline (no API call needed)."""
     if pool_id != int(PoolId.BASIC):
         return
 
@@ -93,22 +203,19 @@ def _maybe_reset_windows(
         if int(table.pool_by_idx[idx]) != pool_id:
             continue
         new_total = int(total_col[idx])
-        window_s = int(window_col[idx])
+        window_s  = int(window_col[idx])
         if new_total <= 0 or window_s <= 0:
             continue
-        # Window expired — restore the last known total for this account/mode.
         quota_col[idx] = new_total
         reset_col[idx] = now_s + window_s
 
 
 def _best(
-    table:     AccountRuntimeTable,
-    working:   set[int],
-    mode_id:   int,
+    table: AccountRuntimeTable,
+    working: set[int],
     quota_col: "array.array",
-    now_s:     int,
-) -> int:
-    """Return the index of the highest-scoring candidate."""
+    now_s: int,
+) -> int | None:
     best_idx   = -1
     best_score = -1e18
 
@@ -118,7 +225,7 @@ def _best(
     last_use_col = table.last_use_at_by_idx
 
     for idx in working:
-        quota    = int(quota_col[idx])
+        quota = int(quota_col[idx])
         if quota <= 0:
             continue
         health   = float(health_col[idx])
@@ -144,60 +251,23 @@ def _best(
     return best_idx if best_idx >= 0 else None
 
 
-def select_any(
-    table: AccountRuntimeTable,
-    pool_id: int,
-    *,
-    exclude_idxs: frozenset[int] | None = None,
-    prefer_tag_idxs: set[int] | None = None,
-    now_s: int,
-) -> int | None:
-    """Select any active account from *pool_id* without mode-specific quota checking.
-
-    Used for WebSocket-based operations that manage their own upstream rate limiting
-    and do not consume tracked quota buckets.  Candidates are the union of all
-    mode_available sets for the pool so that accounts are reachable even when one
-    particular mode bucket (e.g. AUTO) is exhausted.
-    """
-    candidates: set[int] = set()
-    for (pid, _mid), accounts in table.mode_available.items():
-        if pid == pool_id:
-            candidates |= accounts
-
-    if not candidates:
-        return None
-
-    working = candidates.copy()
-    if exclude_idxs:
-        working -= exclude_idxs
-    if not working:
-        return None
-
-    if prefer_tag_idxs:
-        preferred = working & prefer_tag_idxs
-        working = preferred if preferred else working
-
-    return _best_no_quota(table, working, now_s)
-
-
 def _best_no_quota(
     table: AccountRuntimeTable,
     working: set[int],
     now_s: int,
 ) -> int | None:
-    """Score candidates by health and inflight only — no quota weighting."""
-    best_idx = -1
+    best_idx   = -1
     best_score = -1e18
 
-    health_col = table.health_by_idx
+    health_col   = table.health_by_idx
     inflight_col = table.inflight_by_idx
-    fail_col = table.fail_count_by_idx
+    fail_col     = table.fail_count_by_idx
     last_use_col = table.last_use_at_by_idx
 
     for idx in working:
-        health = float(health_col[idx])
+        health   = float(health_col[idx])
         inflight = int(inflight_col[idx])
-        fails = min(int(fail_col[idx]), 10)
+        fails    = min(int(fail_col[idx]), 10)
         last_use = int(last_use_col[idx])
 
         score = health * _W_HEALTH - inflight * _W_INFLIGHT - fails * _W_FAIL
@@ -208,9 +278,62 @@ def _best_no_quota(
 
         if score > best_score:
             best_score = score
-            best_idx = idx
+            best_idx   = idx
 
     return best_idx if best_idx >= 0 else None
 
 
-__all__ = ["select", "select_any"]
+# ---------------------------------------------------------------------------
+# Strategy: random — uniform choice with cooling + inflight filter
+# ---------------------------------------------------------------------------
+
+
+def _random_select(
+    table: AccountRuntimeTable,
+    pool_id: int,
+    *,
+    exclude_idxs: frozenset[int] | None,
+    prefer_tag_idxs: set[int] | None,
+    now_s: int,
+) -> int | None:
+    candidates: set[int] = _pool_union(table, pool_id)
+    if not candidates:
+        return None
+
+    max_inflight = int(get_config("account.selection.max_inflight", 8))
+    cooling_col  = table.cooling_until_s_by_idx
+    inflight_col = table.inflight_by_idx
+
+    working = candidates.copy()
+    if exclude_idxs:
+        working -= exclude_idxs
+    working = {
+        idx for idx in working
+        if int(cooling_col[idx]) <= now_s
+        and int(inflight_col[idx]) < max_inflight
+    }
+    if not working:
+        return None
+
+    if prefer_tag_idxs:
+        preferred = working & prefer_tag_idxs
+        working = preferred if preferred else working
+
+    return random.choice(tuple(working))
+
+
+# ---------------------------------------------------------------------------
+# Shared helper
+# ---------------------------------------------------------------------------
+
+
+def _pool_union(table: AccountRuntimeTable, pool_id: int) -> set[int]:
+    """Union of all ``mode_available`` buckets for ``pool_id``."""
+    out: set[int] = set()
+    for (pid, _mid), accounts in table.mode_available.items():
+        if pid == pool_id:
+            out |= accounts
+    return out
+
+
+__all__ = ["select", "select_any", "set_strategy", "current_strategy"]
diff --git a/app/dataplane/account/table.py b/app/dataplane/account/table.py
index 69ab143a8..16a5719e9 100644
--- a/app/dataplane/account/table.py
+++ b/app/dataplane/account/table.py
@@ -151,6 +151,11 @@ class AccountRuntimeTable:
         default_factory=lambda: array.array("L")
     )
 
+    # --- Per-account cooldown (random strategy only; uint32 epoch-seconds; 0 = not cooling) ---
+    cooling_until_s_by_idx: "array.array[int]" = field(
+        default_factory=lambda: array.array("L")
+    )
+
     # --- Pre-computed selection indexes ---
     # (pool_id, mode_id) → set of idx with a supported quota window and status == ACTIVE
     mode_available: dict[tuple[int, int], set[int]] = field(default_factory=dict)
@@ -300,6 +305,7 @@ def _append_slot(
         self.health_by_idx.append(health)
         self.last_use_at_by_idx.append(last_use_s)
         self.last_fail_at_by_idx.append(last_fail_s)
+        self.cooling_until_s_by_idx.append(0)
         self.size += 1
         self._add_to_indexes(idx)
         self._add_to_tag_idx(idx, tags)
diff --git a/app/dataplane/shared/enums.py b/app/dataplane/shared/enums.py
index bdbe4a9ea..88b9ed66b 100644
--- a/app/dataplane/shared/enums.py
+++ b/app/dataplane/shared/enums.py
@@ -36,6 +36,8 @@ class StatusId(IntEnum):
     "heavy": int(PoolId.HEAVY),
 }
 
+POOL_ID_TO_STR: dict[int, str] = {v: k for k, v in POOL_STR_TO_ID.items()}
+
 STATUS_STR_TO_ID: dict[str, int] = {
     "active": int(StatusId.ACTIVE),
     "cooling": int(StatusId.COOLING),
@@ -56,6 +58,7 @@ class StatusId(IntEnum):
     "PoolId",
     "StatusId",
     "POOL_STR_TO_ID",
+    "POOL_ID_TO_STR",
     "STATUS_STR_TO_ID",
     "ALL_MODE_IDS",
 ]
diff --git a/app/main.py b/app/main.py
index 9a5174928..60bb560a2 100644
--- a/app/main.py
+++ b/app/main.py
@@ -182,7 +182,16 @@ async def _sync_loop() -> None:
     # 4. Account refresh scheduler — only the leader worker.
     #    Uses an advisory file lock so exactly one process runs the heavy
     #    upstream quota-fetch loop regardless of worker count.
+    #
+    #    ``account.refresh.enabled`` selects between two independent runtime
+    #    strategies:
+    #      - true  → "quota" selector + scheduler running (default).
+    #      - false → "random" selector + scheduler idle (no upstream probing).
     from app.control.account.refresh import AccountRefreshService
+    from app.dataplane.account.selector import set_strategy as _set_selection_strategy
+
+    refresh_enabled = _config.get_bool("account.refresh.enabled", False)
+    _set_selection_strategy("quota" if refresh_enabled else "random")
 
     refresh_svc = AccountRefreshService(repo)
     set_refresh_service(refresh_svc)
@@ -190,18 +199,28 @@ async def _sync_loop() -> None:
 
     is_leader = _try_acquire_scheduler_lock()
     scheduler = get_account_refresh_scheduler(refresh_svc)
-    if is_leader:
+    strategy_name = "quota" if refresh_enabled else "random"
+    if is_leader and refresh_enabled:
         scheduler.start()
         logger.info(
-            "scheduler leader: pid={} active_sync_s={} idle_sync_s={}",
+            "scheduler leader: pid={} strategy=quota active_sync_s={} idle_sync_s={}",
+            os.getpid(),
+            _SYNC_ACTIVE_INTERVAL,
+            _SYNC_IDLE_INTERVAL,
+        )
+    elif is_leader:
+        logger.info(
+            "scheduler leader: pid={} strategy=random (scheduler idle) "
+            "active_sync_s={} idle_sync_s={}",
             os.getpid(),
             _SYNC_ACTIVE_INTERVAL,
             _SYNC_IDLE_INTERVAL,
         )
     else:
         logger.info(
-            "scheduler follower: pid={} active_sync_s={} idle_sync_s={}",
+            "scheduler follower: pid={} strategy={} active_sync_s={} idle_sync_s={}",
             os.getpid(),
+            strategy_name,
             _SYNC_ACTIVE_INTERVAL,
             _SYNC_IDLE_INTERVAL,
         )
diff --git a/app/products/_account_selection.py b/app/products/_account_selection.py
index 056583e67..a2f9423d9 100644
--- a/app/products/_account_selection.py
+++ b/app/products/_account_selection.py
@@ -3,8 +3,26 @@
 from app.control.model.enums import ModeId
 from app.control.model.spec import ModelSpec
 from app.control.account.runtime import get_refresh_service
+from app.dataplane.account.selector import current_strategy
 from app.platform.config.snapshot import get_config
 
+# Random strategy has no config key for retry count; it is pinned here so that
+# every retry-driven call site (chat / images / video / anthropic) sees the same
+# value without introducing scattered magic numbers.
+_RANDOM_MAX_RETRIES = 5
+
+
+def selection_max_retries() -> int:
+    """Retry count for account-swap loops, aware of the active selection strategy.
+
+    - ``random`` strategy: fixed at :data:`_RANDOM_MAX_RETRIES` (=5).
+    - ``quota`` strategy:  reads ``retry.max_retries`` (default 1), preserving
+      the historical behaviour.
+    """
+    if current_strategy() == "random":
+        return _RANDOM_MAX_RETRIES
+    return int(get_config("retry.max_retries", 1))
+
 
 def mode_candidates(spec: ModelSpec) -> tuple[int, ...]:
     """Return mode IDs to try for *spec* in priority order.
@@ -32,7 +50,9 @@ async def reserve_account(
 ):
     """Reserve an account and return ``(lease, selected_mode_id)``.
 
-    Returns ``(None, original_mode_id)`` when no account is available.
+    Returns ``(None, original_mode_id)`` when no account is available. Under the
+    random strategy no on-demand refresh fallback is attempted — upstream quota
+    data is never probed.
     """
     original_mode_id = int(spec.mode_id)
 
@@ -52,6 +72,9 @@ async def _try_reserve():
     if lease is not None:
         return lease, selected_mode_id
 
+    if current_strategy() == "random":
+        return None, original_mode_id
+
     refresh_svc = get_refresh_service()
     if refresh_svc is not None:
         await refresh_svc.refresh_on_demand()
diff --git a/app/products/anthropic/messages.py b/app/products/anthropic/messages.py
index 31ef2f90e..42ba98a17 100644
--- a/app/products/anthropic/messages.py
+++ b/app/products/anthropic/messages.py
@@ -33,7 +33,7 @@
     _quota_sync, _fail_sync, _parse_retry_codes, _feedback_kind, _log_task_exception,
     _configured_retry_codes, _should_retry_upstream,
 )
-from app.products._account_selection import reserve_account
+from app.products._account_selection import reserve_account, selection_max_retries
 from app.products.openai._tool_sieve import ToolSieve
 
 
@@ -310,7 +310,7 @@ async def create(
         raise RateLimitError("Account directory not initialised")
     directory = _acct_dir
 
-    max_retries = cfg.get_int("retry.max_retries", 1)
+    max_retries = selection_max_retries()
     retry_codes = _configured_retry_codes(cfg)
     timeout_s   = cfg.get_float("chat.timeout", 120.0)
     msg_id      = _make_msg_id()
diff --git a/app/products/openai/chat.py b/app/products/openai/chat.py
index 4eeab694a..facdd4f55 100644
--- a/app/products/openai/chat.py
+++ b/app/products/openai/chat.py
@@ -54,7 +54,7 @@
     build_usage,
 )
 from ._tool_sieve import ToolSieve
-from app.products._account_selection import reserve_account
+from app.products._account_selection import reserve_account, selection_max_retries
 
 
 def _to_chat_annotations(anns: list[dict]) -> list[dict]:
@@ -470,7 +470,7 @@ async def completions(
         raise RateLimitError("Account directory not initialised")
     directory = _acct_dir
 
-    max_retries = cfg.get_int("retry.max_retries", 1)
+    max_retries = selection_max_retries()
     retry_codes = _configured_retry_codes(cfg)
     response_id = make_response_id()
     timeout_s = cfg.get_float("chat.timeout", 120.0)
diff --git a/app/products/openai/responses.py b/app/products/openai/responses.py
index 95fdc3e7c..d816c7a92 100644
--- a/app/products/openai/responses.py
+++ b/app/products/openai/responses.py
@@ -18,7 +18,7 @@
 from app.control.model.registry import resolve as resolve_model
 from app.control.account.enums import FeedbackKind
 from app.dataplane.reverse.protocol.xai_chat import classify_line, StreamAdapter
-from app.products._account_selection import reserve_account
+from app.products._account_selection import reserve_account, selection_max_retries
 
 from .chat import _stream_chat, _extract_message, _resolve_image, _quota_sync, _fail_sync, _parse_retry_codes, _feedback_kind, _log_task_exception, _upstream_body_excerpt
 from .chat import _configured_retry_codes, _should_retry_upstream
@@ -247,7 +247,7 @@ async def create(
         raise RateLimitError("Account directory not initialised")
     directory = _acct_dir
 
-    max_retries  = cfg.get_int("retry.max_retries", 1)
+    max_retries  = selection_max_retries()
     retry_codes  = _configured_retry_codes(cfg)
     response_id  = make_resp_id("resp")
     reasoning_id = make_resp_id("rs")
diff --git a/app/products/web/admin/__init__.py b/app/products/web/admin/__init__.py
index 529452a4c..61494f8d9 100644
--- a/app/products/web/admin/__init__.py
+++ b/app/products/web/admin/__init__.py
@@ -206,6 +206,7 @@ async def get_storage_mode():
 @router.get("/status", tags=[_TAG_ADMIN_SYSTEM])
 async def runtime_status():
     from app.dataplane.account import _directory
+    from app.dataplane.account.selector import current_strategy
 
     if _directory is None:
         raise AppError(
@@ -220,6 +221,7 @@ async def runtime_status():
                 "status": "ok",
                 "size": _directory.size,
                 "revision": _directory.revision,
+                "selection_strategy": current_strategy(),
             }
         ),
         media_type="application/json",
diff --git a/app/statics/admin/account.html b/app/statics/admin/account.html
index f8983e3fa..9107c5de0 100644
--- a/app/statics/admin/account.html
+++ b/app/statics/admin/account.html
@@ -782,7 +782,7 @@
       </div>
       <div class="stat-num" id="s-calls" style="color:#111">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-quota-only>
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statAuto">Auto 余额</div>
         <span class="stat-icon" style="color:#2563eb">
@@ -791,7 +791,7 @@
       </div>
       <div class="stat-num" id="s-qa" style="color:#2563eb">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-quota-only>
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statFast">Fast 余额</div>
         <span class="stat-icon" style="color:#16a34a">
@@ -800,7 +800,7 @@
       </div>
       <div class="stat-num" id="s-qf" style="color:#16a34a">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-quota-only>
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statExpert">Expert 余额</div>
         <span class="stat-icon" style="color:#7c3aed">
@@ -809,7 +809,7 @@
       </div>
       <div class="stat-num" id="s-qe" style="color:#7c3aed">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-quota-only>
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statHeavy">Heavy 余额</div>
         <span class="stat-icon" style="color:#b0632a">
@@ -934,7 +934,7 @@
           <th style="width:200px" data-i18n="account.colToken">Token</th>
           <th class="table-center" style="width:70px" data-i18n="account.colType">类型</th>
           <th class="table-center" style="width:70px" data-i18n="account.colStatus">状态</th>
-          <th>
+          <th data-quota-only>
             <span data-i18n="account.colQuota">额度</span>
             <span style="margin-left:6px;display:inline-flex;align-items:center;gap:3px;vertical-align:middle">
               <span class="q-pill q-a">A</span>
@@ -1134,25 +1134,45 @@
   return r.json();
 }
 
+// Selection strategy reported by /status — "quota" | "random".
+// Controls whether quota columns / stats are rendered (hidden in random mode).
+let selectionStrategy = 'quota';
+
 // ── Load ───────────────────────────────────────────────────────────────────
 async function load() {
   try {
-    const data = await _api('GET', '/tokens');
+    const [data, status] = await Promise.all([
+      _api('GET', '/tokens'),
+      _api('GET', '/status').catch(() => null),
+    ]);
+    if (status && typeof status.selection_strategy === 'string') {
+      selectionStrategy = status.selection_strategy;
+    }
     allTokens = Array.isArray(data.tokens)
       ? data.tokens
       : Object.entries(data.tokens || {}).flatMap(([pool, items]) =>
           Array.isArray(items) ? items.map(t => ({ ...t, pool: t.pool || pool })) : []);
+    applyStrategyUI();
     invalidateTokenView();
     render();
   } catch (e) { showToast(`${tr('account.loadFailed', null, '加载失败')}: ${e.message}`, 'error'); }
 }
 
+function applyStrategyUI() {
+  // Random strategy: hide quota-related stat cells and the table column.
+  const isRandom = selectionStrategy === 'random';
+  document.querySelectorAll('[data-quota-only]').forEach(el => {
+    el.style.display = isRandom ? 'none' : '';
+  });
+}
+
 // ── Render ─────────────────────────────────────────────────────────────────
 function render() {
   const view = getTokenView();
   renderStats(view);
   renderFilters(view);
   renderTable(view);
+  applyStrategyUI();
 }
 
 function getTokenView() {
@@ -1409,7 +1429,7 @@
       </div></td>
     <td class="table-center"><span class="badge badge-${t.pool||'basic'}">${xe(poolLabel(t.pool || 'basic'))}</span></td>
     <td class="table-center">${statusBadge(t.status)}</td>
-    <td><div class="quota-cell">${qpills(q)}</div></td>
+    <td data-quota-only><div class="quota-cell">${qpills(q)}</div></td>
     <td class="table-center" style="font-variant-numeric:tabular-nums;color:#8f8f8f">${success}</td>
     <td class="table-center" style="font-variant-numeric:tabular-nums;color:#9a9a9a">${fail}</td>
     <td class="table-center" style="font-variant-numeric:tabular-nums;color:#9a9a9a">${fmtRate(success, fail)}</td>
diff --git a/app/statics/admin/config.html b/app/statics/admin/config.html
index eb8fd63b3..a6802007d 100644
--- a/app/statics/admin/config.html
+++ b/app/statics/admin/config.html
@@ -111,6 +111,67 @@
       line-height:1.35;
     }
     .cfg-desc  { max-width:520px; font-size:11px; color:#9a9a9a; line-height:1.45 }
+    .cfg-help {
+      position:relative;
+      width:16px;
+      height:16px;
+      display:inline-flex;
+      align-items:center;
+      justify-content:center;
+      padding:0;
+      border-radius:50%;
+      border:1px solid #e6e6e6;
+      background:#fafafa;
+      color:#8d8d8d;
+      font-size:11px;
+      line-height:1;
+      font-weight:700;
+      cursor:help;
+      flex-shrink:0;
+    }
+    .cfg-help:hover,
+    .cfg-help:focus-visible {
+      color:#222;
+      border-color:#d6d6d6;
+      background:#fff;
+      outline:none;
+    }
+    .cfg-help-tip {
+      position:fixed;
+      box-sizing:border-box;
+      max-width:min(320px, calc(100vw - 24px));
+      padding:9px 10px;
+      border-radius:8px;
+      background:#171717;
+      color:#fff;
+      font-size:12px;
+      font-weight:500;
+      line-height:1.5;
+      text-align:left;
+      white-space:pre-line;
+      overflow-wrap:anywhere;
+      box-shadow:0 8px 24px rgba(0,0,0,.16);
+      opacity:0;
+      pointer-events:none;
+      transform:translateY(3px);
+      transition:opacity .15s, transform .15s;
+      z-index:1000;
+    }
+    .cfg-help-tip.visible {
+      opacity:1;
+      transform:translateY(0);
+    }
+    .cfg-help-tip::before {
+      content:'';
+      position:absolute;
+      left:var(--arrow-left, 50%);
+      width:8px;
+      height:8px;
+      background:#171717;
+      transform:translateX(-50%) rotate(45deg);
+    }
+    .cfg-help-tip[data-placement="top"]::before { bottom:-4px }
+    .cfg-help-tip[data-placement="bottom"]::before { top:-4px }
     .cfg-input-col { min-width:0; display:flex; align-items:center; justify-content:flex-start; gap:8px }
     .cfg-input-col.is-bool { justify-content:flex-end }
     .cfg-inline-action {
@@ -434,6 +495,7 @@
         title: '配额刷新调度', titleKey: 'config.schema.groups.refreshSchedule',
         section: 'account.refresh',
         fields: [
+          { key: 'enabled', label: '启用配额刷新', labelKey: 'config.schema.fields.refreshEnabled.label', type: 'bool', desc: '开启配额探测和评分选号；万级以上账号建议关闭。', descKey: 'config.schema.fields.refreshEnabled.desc', help: '开启：配额刷新模式，scheduler 周期同步真实配额，选号按评分。\n关闭：自动重试模式，不主动探测 upstream，选号随机，出错自动换号重试最多 5 次。\n建议：万级以上账号关闭，避免主动探测触发 upstream 429。', helpKey: 'config.schema.fields.refreshEnabled.help' },
           { key: 'basic_interval_sec', label: 'Basic 刷新周期（秒）', labelKey: 'config.schema.fields.basicInterval.label', type: 'number', desc: 'basic 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 36000 s / 10 h）。', descKey: 'config.schema.fields.basicInterval.desc' },
           { key: 'super_interval_sec', label: 'Super 刷新周期（秒）', labelKey: 'config.schema.fields.superInterval.label', type: 'number', desc: 'super 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 7200 s / 2 h）。', descKey: 'config.schema.fields.superInterval.desc' },
           { key: 'heavy_interval_sec', label: 'Heavy 刷新周期（秒）', labelKey: 'config.schema.fields.heavyInterval.label', type: 'number', desc: 'heavy 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 7200 s / 2 h）。', descKey: 'config.schema.fields.heavyInterval.desc' },
@@ -441,6 +503,22 @@
           { key: 'usage_concurrency', label: '刷新并发数', labelKey: 'config.schema.fields.usageConcurrency.label', type: 'number', desc: '并发调用 usage 接口刷新额度时允许的最大并行数。', descKey: 'config.schema.fields.usageConcurrency.desc' },
         ]
       },
+      {
+        title: '自动重试冷却', titleKey: 'config.schema.groups.randomCooling',
+        section: 'account.random',
+        fields: [
+          { key: 'cooling_basic_hours', label: 'Basic 冷却（小时）', labelKey: 'config.schema.fields.coolingBasicHours.label', type: 'number', desc: 'basic 号收到 429 后的冷却时长，默认 20 h（对齐 upstream 20 h 限流窗口）。仅在自动重试模式下生效。', descKey: 'config.schema.fields.coolingBasicHours.desc' },
+          { key: 'cooling_super_hours', label: 'Super 冷却（小时）', labelKey: 'config.schema.fields.coolingSuperHours.label', type: 'number', desc: 'super 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。', descKey: 'config.schema.fields.coolingSuperHours.desc' },
+          { key: 'cooling_heavy_hours', label: 'Heavy 冷却（小时）', labelKey: 'config.schema.fields.coolingHeavyHours.label', type: 'number', desc: 'heavy 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。', descKey: 'config.schema.fields.coolingHeavyHours.desc' },
+        ]
+      },
+      {
+        title: '选号并发限制', titleKey: 'config.schema.groups.selection',
+        section: 'account.selection',
+        fields: [
+          { key: 'max_inflight', label: '单号并发上限', labelKey: 'config.schema.fields.maxInflight.label', type: 'number', desc: '单个账号同时在执行的请求上限，超过则选号路径跳过该号（两种模式共用）。', descKey: 'config.schema.fields.maxInflight.desc' },
+        ]
+      },
       {
         title: '批量操作并发', titleKey: 'config.schema.groups.batchConcurrency',
         section: 'batch',
@@ -722,6 +800,7 @@
     ...field,
     label: field.labelKey ? tr(field.labelKey, null, field.label ?? field.key) : (field.label ?? field.key),
     desc: field.descKey ? tr(field.descKey, null, field.desc ?? '') : (field.desc ?? ''),
+    help: field.helpKey ? tr(field.helpKey, null, field.help ?? '') : (field.help ?? ''),
     options: field.options ? field.options.map(_localizeOption) : field.options,
   };
 }
@@ -824,6 +903,76 @@
   return el;
 }
 
+let _helpTipEl = null;
+let _activeHelpButton = null;
+
+function _hideHelpTip() {
+  if (_activeHelpButton) _activeHelpButton.removeAttribute('aria-describedby');
+  if (_helpTipEl) _helpTipEl.remove();
+  _helpTipEl = null;
+  _activeHelpButton = null;
+}
+
+function _positionHelpTip(button, tip) {
+  const margin = 12;
+  const rect = button.getBoundingClientRect();
+  const viewportWidth = window.innerWidth || document.documentElement.clientWidth;
+  const viewportHeight = window.innerHeight || document.documentElement.clientHeight;
+  const maxWidth = Math.max(120, viewportWidth - margin * 2);
+
+  tip.style.maxWidth = `${Math.min(320, maxWidth)}px`;
+  tip.style.left = '0px';
+  tip.style.top = '0px';
+
+  const tipRect = tip.getBoundingClientRect();
+  let left = rect.left + rect.width / 2 - tipRect.width / 2;
+  left = Math.max(margin, Math.min(left, viewportWidth - tipRect.width - margin));
+
+  let placement = 'top';
+  let top = rect.top - tipRect.height - 8;
+  if (top < margin) {
+    placement = 'bottom';
+    top = rect.bottom + 8;
+  }
+  if (top + tipRect.height > viewportHeight - margin) {
+    top = Math.max(margin, viewportHeight - tipRect.height - margin);
+  }
+
+  const arrowLeft = Math.max(12, Math.min(rect.left + rect.width / 2 - left, tipRect.width - 12));
+  tip.dataset.placement = placement;
+  tip.style.setProperty('--arrow-left', `${arrowLeft}px`);
+  tip.style.left = `${left}px`;
+  tip.style.top = `${top}px`;
+}
+
+function _showHelpTip(button) {
+  const text = button.getAttribute('data-tip');
+  if (!text) return;
+  _hideHelpTip();
+
+  const tip = _el('div', {
+    class: 'cfg-help-tip',
+    id: 'cfg-help-tip',
+    role: 'tooltip',
+  }, text);
+  document.body.appendChild(tip);
+  button.setAttribute('aria-describedby', 'cfg-help-tip');
+
+  _helpTipEl = tip;
+  _activeHelpButton = button;
+  _positionHelpTip(button, tip);
+  requestAnimationFrame(() => tip.classList.add('visible'));
+}
+
+window.addEventListener('resize', _hideHelpTip);
+window.addEventListener('scroll', _hideHelpTip, true);
+document.addEventListener('keydown', (event) => {
+  if (event.key === 'Escape') _hideHelpTip();
+});
+document.addEventListener('click', (event) => {
+  if (!event.target.closest?.('.cfg-help')) _hideHelpTip();
+});
+
 function renderInput(field, section, value) {
   const fp = _fullPath(section, field.key);
 
@@ -941,6 +1090,24 @@
     labelChildren.push(btn);
   }
 
+  if (field.help) {
+    labelChildren.push(_el('button', {
+      class: 'cfg-help',
+      type: 'button',
+      'aria-label': field.help,
+      'data-tip': field.help,
+      onmouseenter: (event) => _showHelpTip(event.currentTarget),
+      onmouseleave: _hideHelpTip,
+      onfocus: (event) => _showHelpTip(event.currentTarget),
+      onblur: _hideHelpTip,
+      onclick: (event) => {
+        event.preventDefault();
+        event.stopPropagation();
+        _showHelpTip(event.currentTarget);
+      },
+    }, '?'));
+  }
+
   labelChildren.push(dot);
 
   const labelCol = _el('div', { class: 'cfg-label-col' },
diff --git a/app/statics/i18n/de.json b/app/statics/i18n/de.json
index c9d974d92..c75508a86 100644
--- a/app/statics/i18n/de.json
+++ b/app/statics/i18n/de.json
@@ -330,6 +330,8 @@
         "features": "Kernfunktionen",
         "mediaFormat": "Medienausgabe",
         "refreshSchedule": "Quotenaktualisierung",
+        "randomCooling": "Auto-Retry-Abkühlung",
+        "selection": "Auswahl-Nebenläufigkeit",
         "batchConcurrency": "Batch-Konkurrenz",
         "egressProxy": "Ausgehender Proxy",
         "clearance": "Cloudflare Clearance",
@@ -381,11 +383,16 @@
         "customInstruction": { "label": "Globale Zusatzanweisung" },
         "imageFormat": { "label": "Bildausgabeformat" },
         "videoFormat": { "label": "Videoausgabeformat" },
-        "basicInterval": { "label": "Basic-Aktualisierungsintervall (s)" },
-        "superInterval": { "label": "Super-Aktualisierungsintervall (s)" },
-        "heavyInterval": { "label": "Heavy-Aktualisierungsintervall (s)" },
-        "usageConcurrency": { "label": "Aktualisierungs-Konkurrenz" },
-        "onDemandMinInterval": { "label": "Minimales On-Demand-Intervall (s)" },
+        "refreshEnabled": { "label": "Quotenaktualisierung aktivieren", "desc": "Aktiviert Quotenabfrage und bewertete Auswahl; bei mehr als 10.000 Konten empfohlen aus.", "help": "Ein: Quotenaktualisierung. Der Scheduler synchronisiert regelmäßig echte Quoten, die Auswahl erfolgt nach Bewertung.\nAus: Auto-Retry-Modus. Keine aktive upstream-Abfrage, zufällige Auswahl und automatischer Kontowechsel bei Fehlern mit bis zu 5 Wiederholungen.\nEmpfehlung: Bei mehr als 10.000 Konten ausschalten, um upstream 429 durch aktive Abfragen zu vermeiden." },
+        "coolingBasicHours": { "label": "Basic-Abkühlung (Stunden)", "desc": "Abkühldauer nach 429 für Basic-Konten, Standard 20 h (entspricht dem upstream-20-h-Fenster). Nur im Auto-Retry-Modus wirksam." },
+        "coolingSuperHours": { "label": "Super-Abkühlung (Stunden)", "desc": "Abkühldauer nach 429 für Super-Konten, Standard 2 h (entspricht dem upstream-2-h-Fenster). Nur im Auto-Retry-Modus wirksam." },
+        "coolingHeavyHours": { "label": "Heavy-Abkühlung (Stunden)", "desc": "Abkühldauer nach 429 für Heavy-Konten, Standard 2 h (entspricht dem upstream-2-h-Fenster). Nur im Auto-Retry-Modus wirksam." },
+        "maxInflight": { "label": "Max. laufende Anfragen pro Konto", "desc": "Maximale Anzahl gleichzeitig laufender Anfragen pro Konto. Konten am Limit werden von der Auswahl übersprungen (für beide Modi gemeinsam)." },
+        "basicInterval": { "label": "Basic-Aktualisierungsintervall (s)", "desc": "Geplantes Aktualisierungsintervall für den Basic-Pool. Möglichst am Quotenfenster ausrichten (Standard 36000 s / 10 h)." },
+        "superInterval": { "label": "Super-Aktualisierungsintervall (s)", "desc": "Geplantes Aktualisierungsintervall für den Super-Pool. Möglichst am Quotenfenster ausrichten (Standard 7200 s / 2 h)." },
+        "heavyInterval": { "label": "Heavy-Aktualisierungsintervall (s)", "desc": "Geplantes Aktualisierungsintervall für den Heavy-Pool. Möglichst am Quotenfenster ausrichten (Standard 7200 s / 2 h)." },
+        "usageConcurrency": { "label": "Aktualisierungs-Konkurrenz", "desc": "Maximale Anzahl paralleler usage-Aufrufe während der Quotenaktualisierung." },
+        "onDemandMinInterval": { "label": "Minimales On-Demand-Intervall (s)", "desc": "Drosselungsfenster für durch Requests ausgelöste Aktualisierungen, z. B. nach 429. Wiederholte Auslöser innerhalb des Intervalls werden zu einer Aktualisierung zusammengefasst." },
         "nsfwConcurrency": { "label": "Konkurrenz für NSFW-Aktivierung" },
         "refreshConcurrency": { "label": "Usage-Aktualisierungskonkurrenz" },
         "assetUploadConcurrency": { "label": "Asset-Upload-Konkurrenz" },
diff --git a/app/statics/i18n/en.json b/app/statics/i18n/en.json
index 0b8dd3f1f..2d529e25e 100644
--- a/app/statics/i18n/en.json
+++ b/app/statics/i18n/en.json
@@ -331,6 +331,8 @@
         "features": "Core Features",
         "mediaFormat": "Media Output",
         "refreshSchedule": "Quota Refresh",
+        "randomCooling": "Auto-Retry Cooldown",
+        "selection": "Selection Concurrency",
         "batchConcurrency": "Batch Concurrency",
         "egressProxy": "Egress Proxy",
         "clearance": "Cloudflare Clearance",
@@ -433,6 +435,27 @@
           "label": "Video Output Format",
           "desc": "local_* modes download video assets to the server and deliver them through the local proxy. Configure the APP Base URL first and account for outbound bandwidth usage."
         },
+        "refreshEnabled": {
+          "label": "Enable Quota Refresh",
+          "desc": "Enables quota probing and scored selection. Recommended off for >10k accounts.",
+          "help": "On: quota refresh mode. The scheduler periodically syncs real quota, and selection uses scoring.\nOff: auto-retry mode. No upstream probing, uniform random selection, and auto-swap on errors for up to 5 retries.\nRecommendation: turn this off for >10k accounts to avoid self-triggering upstream 429s."
+        },
+        "coolingBasicHours": {
+          "label": "Basic Cooldown (hours)",
+          "desc": "Cooldown duration after a 429 on basic-pool accounts, default 20h (matches upstream 20h window). Only effective in auto-retry mode."
+        },
+        "coolingSuperHours": {
+          "label": "Super Cooldown (hours)",
+          "desc": "Cooldown duration after a 429 on super-pool accounts, default 2h (matches upstream 2h window). Only effective in auto-retry mode."
+        },
+        "coolingHeavyHours": {
+          "label": "Heavy Cooldown (hours)",
+          "desc": "Cooldown duration after a 429 on heavy-pool accounts, default 2h (matches upstream 2h window). Only effective in auto-retry mode."
+        },
+        "maxInflight": {
+          "label": "Max Inflight per Account",
+          "desc": "Maximum concurrent in-flight requests per account. Accounts at this limit are skipped by the selector (shared by both modes)."
+        },
         "basicInterval": {
           "label": "Basic Refresh Interval (s)",
           "desc": "Scheduled quota refresh interval for the basic pool. Align this with the quota window when possible (default 36000 s / 10 h)."
diff --git a/app/statics/i18n/es.json b/app/statics/i18n/es.json
index 3ec90709f..04f3c74be 100644
--- a/app/statics/i18n/es.json
+++ b/app/statics/i18n/es.json
@@ -330,6 +330,8 @@
         "features": "Funciones principales",
         "mediaFormat": "Salida multimedia",
         "refreshSchedule": "Actualización de cuota",
+        "randomCooling": "Enfriamiento de reintento automático",
+        "selection": "Concurrencia de selección",
         "batchConcurrency": "Concurrencia por lotes",
         "egressProxy": "Proxy de salida",
         "clearance": "Cloudflare Clearance",
@@ -381,11 +383,16 @@
         "customInstruction": { "label": "Instrucción suplementaria global" },
         "imageFormat": { "label": "Formato de salida de imagen" },
         "videoFormat": { "label": "Formato de salida de video" },
-        "basicInterval": { "label": "Intervalo de refresco Basic (s)" },
-        "superInterval": { "label": "Intervalo de refresco Super (s)" },
-        "heavyInterval": { "label": "Intervalo de refresco Heavy (s)" },
-        "usageConcurrency": { "label": "Concurrencia de actualización" },
-        "onDemandMinInterval": { "label": "Intervalo mínimo bajo demanda (s)" },
+        "refreshEnabled": { "label": "Activar actualización de cuota", "desc": "Activa la comprobación de cuotas y la selección por puntuación; se recomienda desactivarlo con más de 10.000 cuentas.", "help": "Activado: modo de actualización de cuota. El scheduler sincroniza periódicamente la cuota real y la selección usa puntuación.\nDesactivado: modo de reintento automático. No comprueba upstream activamente, usa selección aleatoria y cambia de cuenta al fallar hasta 5 reintentos.\nRecomendación: desactívalo con más de 10.000 cuentas para evitar provocar 429 de upstream con comprobaciones activas." },
+        "coolingBasicHours": { "label": "Enfriamiento Basic (horas)", "desc": "Duración del enfriamiento tras un 429 en cuentas Basic, por defecto 20 h (coincide con la ventana upstream de 20 h). Solo efectivo en modo de reintento automático." },
+        "coolingSuperHours": { "label": "Enfriamiento Super (horas)", "desc": "Duración del enfriamiento tras un 429 en cuentas Super, por defecto 2 h (coincide con la ventana upstream de 2 h). Solo efectivo en modo de reintento automático." },
+        "coolingHeavyHours": { "label": "Enfriamiento Heavy (horas)", "desc": "Duración del enfriamiento tras un 429 en cuentas Heavy, por defecto 2 h (coincide con la ventana upstream de 2 h). Solo efectivo en modo de reintento automático." },
+        "maxInflight": { "label": "Máximo en curso por cuenta", "desc": "Máximo de solicitudes simultáneas en curso por cuenta. Las cuentas en ese límite se omiten en la selección (compartido por ambos modos)." },
+        "basicInterval": { "label": "Intervalo de refresco Basic (s)", "desc": "Intervalo programado de actualización para el pool Basic. Alinéalo con la ventana de cuota cuando sea posible (por defecto 36000 s / 10 h)." },
+        "superInterval": { "label": "Intervalo de refresco Super (s)", "desc": "Intervalo programado de actualización para el pool Super. Alinéalo con la ventana de cuota cuando sea posible (por defecto 7200 s / 2 h)." },
+        "heavyInterval": { "label": "Intervalo de refresco Heavy (s)", "desc": "Intervalo programado de actualización para el pool Heavy. Alinéalo con la ventana de cuota cuando sea posible (por defecto 7200 s / 2 h)." },
+        "usageConcurrency": { "label": "Concurrencia de actualización", "desc": "Número máximo de llamadas usage paralelas permitidas durante la actualización de cuota." },
+        "onDemandMinInterval": { "label": "Intervalo mínimo bajo demanda (s)", "desc": "Ventana de limitación para actualizaciones activadas por solicitudes, como el manejo de 429. Los disparos repetidos dentro del intervalo se agrupan en una sola actualización." },
         "nsfwConcurrency": { "label": "Concurrencia de activación NSFW" },
         "refreshConcurrency": { "label": "Concurrencia de refresco de Usage" },
         "assetUploadConcurrency": { "label": "Concurrencia de carga de Asset" },
diff --git a/app/statics/i18n/fr.json b/app/statics/i18n/fr.json
index f947d7530..71c87c506 100644
--- a/app/statics/i18n/fr.json
+++ b/app/statics/i18n/fr.json
@@ -330,6 +330,8 @@
         "features": "Fonctionnalités principales",
         "mediaFormat": "Sortie multimédia",
         "refreshSchedule": "Actualisation du quota",
+        "randomCooling": "Refroidissement de relance automatique",
+        "selection": "Concurrence de sélection",
         "batchConcurrency": "Concurrence par lot",
         "egressProxy": "Proxy sortant",
         "clearance": "Cloudflare Clearance",
@@ -381,11 +383,16 @@
         "customInstruction": { "label": "Instruction globale supplémentaire" },
         "imageFormat": { "label": "Format de sortie image" },
         "videoFormat": { "label": "Format de sortie vidéo" },
-        "basicInterval": { "label": "Intervalle d’actualisation Basic (s)" },
-        "superInterval": { "label": "Intervalle d’actualisation Super (s)" },
-        "heavyInterval": { "label": "Intervalle d’actualisation Heavy (s)" },
-        "usageConcurrency": { "label": "Concurrence d’actualisation" },
-        "onDemandMinInterval": { "label": "Intervalle minimal à la demande (s)" },
+        "refreshEnabled": { "label": "Activer l’actualisation du quota", "desc": "Active la sonde de quota et la sélection par score ; recommandé désactivé au-delà de 10 000 comptes.", "help": "Activé : mode d’actualisation du quota. Le scheduler synchronise périodiquement le quota réel et la sélection utilise un score.\nDésactivé : mode de relance automatique. Aucune sonde upstream active, sélection aléatoire et changement automatique de compte en cas d’erreur jusqu’à 5 relances.\nRecommandation : désactivez-le au-delà de 10 000 comptes pour éviter de déclencher des 429 upstream par les sondes actives." },
+        "coolingBasicHours": { "label": "Refroidissement Basic (heures)", "desc": "Durée de refroidissement après un 429 sur les comptes Basic, par défaut 20 h (alignée sur la fenêtre upstream de 20 h). Effectif uniquement en mode de relance automatique." },
+        "coolingSuperHours": { "label": "Refroidissement Super (heures)", "desc": "Durée de refroidissement après un 429 sur les comptes Super, par défaut 2 h (alignée sur la fenêtre upstream de 2 h). Effectif uniquement en mode de relance automatique." },
+        "coolingHeavyHours": { "label": "Refroidissement Heavy (heures)", "desc": "Durée de refroidissement après un 429 sur les comptes Heavy, par défaut 2 h (alignée sur la fenêtre upstream de 2 h). Effectif uniquement en mode de relance automatique." },
+        "maxInflight": { "label": "Maximum en cours par compte", "desc": "Nombre maximal de requêtes simultanées par compte. Les comptes à cette limite sont ignorés par la sélection (partagé par les deux modes)." },
+        "basicInterval": { "label": "Intervalle d’actualisation Basic (s)", "desc": "Intervalle d’actualisation planifiée du pool Basic. À aligner avec la fenêtre de quota si possible (par défaut 36000 s / 10 h)." },
+        "superInterval": { "label": "Intervalle d’actualisation Super (s)", "desc": "Intervalle d’actualisation planifiée du pool Super. À aligner avec la fenêtre de quota si possible (par défaut 7200 s / 2 h)." },
+        "heavyInterval": { "label": "Intervalle d’actualisation Heavy (s)", "desc": "Intervalle d’actualisation planifiée du pool Heavy. À aligner avec la fenêtre de quota si possible (par défaut 7200 s / 2 h)." },
+        "usageConcurrency": { "label": "Concurrence d’actualisation", "desc": "Nombre maximal d’appels usage parallèles autorisés pendant l’actualisation du quota." },
+        "onDemandMinInterval": { "label": "Intervalle minimal à la demande (s)", "desc": "Fenêtre de limitation pour les actualisations déclenchées par les requêtes, par exemple après un 429. Les déclenchements répétés dans l’intervalle sont regroupés en une seule actualisation." },
         "nsfwConcurrency": { "label": "Concurrence d’activation NSFW" },
         "refreshConcurrency": { "label": "Concurrence d’actualisation Usage" },
         "assetUploadConcurrency": { "label": "Concurrence d’envoi Asset" },
diff --git a/app/statics/i18n/ja.json b/app/statics/i18n/ja.json
index b5fb9ae29..227deb341 100644
--- a/app/statics/i18n/ja.json
+++ b/app/statics/i18n/ja.json
@@ -330,6 +330,8 @@
         "features": "基本機能",
         "mediaFormat": "メディア出力",
         "refreshSchedule": "クォータ更新",
+        "randomCooling": "自動再試行クールダウン",
+        "selection": "選択同時実行数",
         "batchConcurrency": "一括処理の並列数",
         "egressProxy": "送信プロキシ",
         "clearance": "Cloudflare Clearance",
@@ -381,11 +383,16 @@
         "customInstruction": { "label": "グローバル補助指示" },
         "imageFormat": { "label": "画像出力形式" },
         "videoFormat": { "label": "動画出力形式" },
-        "basicInterval": { "label": "Basic 更新間隔（秒）" },
-        "superInterval": { "label": "Super 更新間隔（秒）" },
-        "heavyInterval": { "label": "Heavy 更新間隔（秒）" },
-        "usageConcurrency": { "label": "更新並列数" },
-        "onDemandMinInterval": { "label": "最小オンデマンド更新間隔（秒）" },
+        "refreshEnabled": { "label": "クォータ更新を有効化", "desc": "クォータ確認とスコア選択を有効にします。1 万件以上のアカウントでは無効化を推奨します。", "help": "オン：クォータ更新モード。scheduler が実クォータを定期同期し、スコアでアカウントを選択します。\nオフ：自動再試行モード。upstream を能動的に確認せず、ランダム選択し、エラー時は最大 5 回まで別アカウントへ自動切替します。\n推奨：1 万件以上のアカウントではオフにして、能動確認による upstream 429 を避けてください。" },
+        "coolingBasicHours": { "label": "Basic クールダウン（時間）", "desc": "Basic アカウントで 429 を受けた後のクールダウン時間。既定は 20 h（upstream の 20 h ウィンドウに対応）。自動再試行モードでのみ有効です。" },
+        "coolingSuperHours": { "label": "Super クールダウン（時間）", "desc": "Super アカウントで 429 を受けた後のクールダウン時間。既定は 2 h（upstream の 2 h ウィンドウに対応）。自動再試行モードでのみ有効です。" },
+        "coolingHeavyHours": { "label": "Heavy クールダウン（時間）", "desc": "Heavy アカウントで 429 を受けた後のクールダウン時間。既定は 2 h（upstream の 2 h ウィンドウに対応）。自動再試行モードでのみ有効です。" },
+        "maxInflight": { "label": "アカウントごとの同時実行上限", "desc": "1 アカウントで同時に処理中にできるリクエスト数の上限です。上限に達したアカウントは選択時にスキップされます（両モード共通）。" },
+        "basicInterval": { "label": "Basic 更新間隔（秒）", "desc": "Basic プールの定期クォータ更新間隔です。可能ならクォータウィンドウに合わせてください（既定 36000 s / 10 h）。" },
+        "superInterval": { "label": "Super 更新間隔（秒）", "desc": "Super プールの定期クォータ更新間隔です。可能ならクォータウィンドウに合わせてください（既定 7200 s / 2 h）。" },
+        "heavyInterval": { "label": "Heavy 更新間隔（秒）", "desc": "Heavy プールの定期クォータ更新間隔です。可能ならクォータウィンドウに合わせてください（既定 7200 s / 2 h）。" },
+        "usageConcurrency": { "label": "更新並列数", "desc": "クォータ更新中に許可する usage 呼び出しの最大並列数です。" },
+        "onDemandMinInterval": { "label": "最小オンデマンド更新間隔（秒）", "desc": "429 処理など、リクエスト経路から発生する更新のスロットル間隔です。この間隔内の重複トリガーは 1 回の更新にまとめられます。" },
         "nsfwConcurrency": { "label": "NSFW 有効化の並列数" },
         "refreshConcurrency": { "label": "Usage 更新の並列数" },
         "assetUploadConcurrency": { "label": "Asset アップロード並列数" },
diff --git a/app/statics/i18n/zh.json b/app/statics/i18n/zh.json
index 6f88402b7..1f21050bf 100644
--- a/app/statics/i18n/zh.json
+++ b/app/statics/i18n/zh.json
@@ -331,6 +331,8 @@
         "features": "基本功能设置",
         "mediaFormat": "媒体返回格式",
         "refreshSchedule": "配额刷新调度",
+        "randomCooling": "自动重试冷却",
+        "selection": "选号并发限制",
         "batchConcurrency": "批量操作并发",
         "egressProxy": "出站代理设置",
         "clearance": "Cloudflare Clearance",
@@ -433,6 +435,27 @@
           "label": "视频返回格式",
           "desc": "local_* 模式会先将视频下载至服务端，再通过本地 URL 代理分发；请确保已配置 APP 访问地址，并评估出站带宽消耗。"
         },
+        "refreshEnabled": {
+          "label": "启用配额刷新",
+          "desc": "开启配额探测和评分选号；万级以上账号建议关闭。",
+          "help": "开启：配额刷新模式，scheduler 周期同步真实配额，选号按评分。\n关闭：自动重试模式，不主动探测 upstream，选号随机，出错自动换号重试最多 5 次。\n建议：万级以上账号关闭，避免主动探测触发 upstream 429。"
+        },
+        "coolingBasicHours": {
+          "label": "Basic 冷却（小时）",
+          "desc": "basic 号收到 429 后的冷却时长，默认 20 h（对齐 upstream 20 h 限流窗口）。仅在自动重试模式下生效。"
+        },
+        "coolingSuperHours": {
+          "label": "Super 冷却（小时）",
+          "desc": "super 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。"
+        },
+        "coolingHeavyHours": {
+          "label": "Heavy 冷却（小时）",
+          "desc": "heavy 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。"
+        },
+        "maxInflight": {
+          "label": "单号并发上限",
+          "desc": "单个账号同时在执行的请求上限，超过则选号路径跳过该号（两种模式共用）。"
+        },
         "basicInterval": {
           "label": "Basic 刷新周期（秒）",
           "desc": "basic 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 36000 s / 10 h）。"
diff --git a/config.defaults.toml b/config.defaults.toml
index 4a716fefe..2e64cd878 100644
--- a/config.defaults.toml
+++ b/config.defaults.toml
@@ -103,6 +103,8 @@ on_codes = "429,401,503"
 
 
 [account.refresh]
+# 总开关：true=配额刷新模式（主动探测，选号评分）；false=自动重试模式（随机选号，零探测）
+enabled = true
 basic_interval_sec = 36000   # basic 号池刷新周期（秒），默认 10 h
 super_interval_sec = 7200    # super 号池刷新周期（秒），默认 2 h
 heavy_interval_sec = 7200    # heavy 号池刷新周期（秒），默认 2 h
@@ -110,6 +112,18 @@ usage_concurrency = 50
 on_demand_min_interval_sec = 300
 
 
+[account.random]
+# 仅 account.refresh.enabled=false 时生效
+# 429 后按 pool 冷却的小时数（对齐 upstream 限流窗口）
+cooling_basic_hours = 20
+cooling_heavy_hours = 2
+
+
+[account.selection]
+# 单号并发上限（两模式共用）
+max_inflight = 8
+
+
 # ==================== 对话配置 ====================
 [chat]
 timeout = 60

From 54748c378a00e63bd3b5d2596f79c18a074aaede Mon Sep 17 00:00:00 2001
From: Chenyme <118253778+chenyme@users.noreply.github.com>
Date: Mon, 20 Apr 2026 20:36:30 +0800
Subject: [PATCH 2/2] refactor: Update account refresh configuration and
 enhance runtime management

---
 app/control/account/quota_defaults.py |  8 +--
 app/control/account/refresh.py        | 18 ++++--
 app/control/account/runtime.py        | 86 ++++++++++++++++++++++++++-
 app/control/account/scheduler.py      | 17 +++++-
 app/dataplane/account/__init__.py     | 15 ++---
 app/main.py                           | 24 ++++++--
 app/products/openai/chat.py           | 15 ++++-
 app/products/web/admin/__init__.py    | 14 ++++-
 app/statics/admin/account.html        | 71 ++++++++++++++++++----
 app/statics/admin/config.html         | 19 ++----
 app/statics/i18n/de.json              | 15 ++---
 app/statics/i18n/en.json              | 30 +++-------
 app/statics/i18n/es.json              | 15 ++---
 app/statics/i18n/fr.json              | 15 ++---
 app/statics/i18n/ja.json              | 15 ++---
 app/statics/i18n/zh.json              | 30 +++-------
 config.defaults.toml                  | 13 +---
 17 files changed, 275 insertions(+), 145 deletions(-)

diff --git a/app/control/account/quota_defaults.py b/app/control/account/quota_defaults.py
index f5464806e..6ab8d861a 100644
--- a/app/control/account/quota_defaults.py
+++ b/app/control/account/quota_defaults.py
@@ -2,10 +2,10 @@
 
 Canonical quota totals per pool type (from upstream rate-limits API):
 
-              auto    fast    expert    heavy
-  basic         20      60       8        —      window: 72000 / 36000 s
-  super         50     140      50        —      window: 7200 s
-  heavy        150     400     150       20      window: 7200 s
+              auto    fast    expert    heavy    grok_4_3
+  basic         20      60       8        —         —        window: 72000 / 36000 s
+  super         50     140      50        —        50        window: 7200 s
+  heavy        150     400     150       20       150        window: 7200 s
 
 Pool inference uses ``auto.total`` as the primary signal — the three values
 (20 / 50 / 150) are mutually exclusive across pool types.
diff --git a/app/control/account/refresh.py b/app/control/account/refresh.py
index c21238d45..203c27784 100644
--- a/app/control/account/refresh.py
+++ b/app/control/account/refresh.py
@@ -57,9 +57,9 @@ class AccountRefreshService:
     """Fetches real quota data from the upstream usage API and persists it.
 
     Triggers:
-      1. Import   — super accounts: fetch all 3 modes.
-      2. Call     — super accounts: fetch the called mode (async, non-blocking).
-      3. Schedule — super: fetch all 3 modes; basic: static window reset check.
+      1. Import   — fetch all modes supported by the account's pool.
+      2. Call     — fetch the called mode only (async, non-blocking).
+      3. Schedule — refresh one pool per loop using that pool's supported modes.
     """
 
     def __init__(self, repository: "AccountRepository") -> None:
@@ -75,7 +75,13 @@ def __init__(self, repository: "AccountRepository") -> None:
     async def _fetch_all_quotas(
         self, token: str, pool: str
     ) -> dict[int, QuotaWindow] | None:
-        """Fetch quota windows for modes supported by *pool*. Returns {mode_id: window}."""
+        """Fetch quota windows for every mode supported by *pool*.
+
+        Examples:
+          - basic -> auto / fast / expert
+          - super -> auto / fast / expert / grok_4_3
+          - heavy -> auto / fast / expert / heavy / grok_4_3
+        """
         try:
             from app.dataplane.reverse.protocol.xai_usage import fetch_all_quotas
 
@@ -142,7 +148,7 @@ async def refresh_on_import(self, tokens: list[str]) -> RefreshResult:
         return agg
 
     async def refresh_call_async(self, token: str, mode_id: int) -> None:
-        """Fire-and-forget quota sync after a successful call (all accounts)."""
+        """Fire-and-forget single-mode quota sync after a successful call."""
         record = (await self._repo.get_accounts([token]) or [None])[0]
         if record is None or record.is_deleted():
             return
@@ -219,7 +225,7 @@ async def _refresh_one(
         *,
         apply_fallback: bool = False,
     ) -> RefreshResult:
-        """Fetch all 3 modes from the usage API and persist real quota data.
+        """Fetch all pool-supported modes from the usage API and persist them.
 
         apply_fallback=True  — used by scheduled/import paths: when API fails,
                                decrement REAL quotas or reset expired DEFAULT windows.
diff --git a/app/control/account/runtime.py b/app/control/account/runtime.py
index 00a5778c0..fbf5f7f3a 100644
--- a/app/control/account/runtime.py
+++ b/app/control/account/runtime.py
@@ -1,11 +1,19 @@
-"""Account runtime singletons exposed without importing app.main."""
+"""Account runtime singletons and hot-apply helpers.
 
-from typing import TYPE_CHECKING
+These helpers expose the process-local account refresh runtime without making
+callers import ``app.main``. Admin handlers use them to reconcile strategy and
+scheduler state after hot config updates.
+"""
+
+from typing import TYPE_CHECKING, Literal
 
 if TYPE_CHECKING:
     from .refresh import AccountRefreshService
+    from .scheduler import AccountRefreshScheduler
 
 _refresh_service: "AccountRefreshService | None" = None
+_refresh_scheduler: "AccountRefreshScheduler | None" = None
+_refresh_scheduler_leader = False
 
 
 def set_refresh_service(service: "AccountRefreshService | None") -> None:
@@ -19,4 +27,76 @@ def get_refresh_service() -> "AccountRefreshService | None":
     return _refresh_service
 
 
-__all__ = ["get_refresh_service", "set_refresh_service"]
+def set_refresh_scheduler(scheduler: "AccountRefreshScheduler | None") -> None:
+    """Register the process-global account refresh scheduler."""
+    global _refresh_scheduler
+    _refresh_scheduler = scheduler
+
+
+def get_refresh_scheduler() -> "AccountRefreshScheduler | None":
+    """Return the registered account refresh scheduler, if any."""
+    return _refresh_scheduler
+
+
+def set_refresh_scheduler_leader(is_leader: bool) -> None:
+    """Record whether this worker currently owns the refresh scheduler lock."""
+    global _refresh_scheduler_leader
+    _refresh_scheduler_leader = bool(is_leader)
+
+
+def is_refresh_scheduler_leader() -> bool:
+    """Return True when this worker is the active refresh-scheduler leader."""
+    return _refresh_scheduler_leader
+
+
+def reconcile_refresh_runtime(
+    enabled: bool | None = None,
+) -> Literal["quota", "random"]:
+    """Hot-apply refresh strategy and scheduler state for the current worker."""
+    from app.dataplane.account.selector import current_strategy, set_strategy
+    from app.platform.config.snapshot import config
+    from app.platform.logging.logger import logger
+
+    refresh_enabled = (
+        config.get_bool("account.refresh.enabled", False)
+        if enabled is None
+        else bool(enabled)
+    )
+    target_strategy: Literal["quota", "random"] = (
+        "quota" if refresh_enabled else "random"
+    )
+    previous_strategy = current_strategy()
+    if previous_strategy != target_strategy:
+        set_strategy(target_strategy)
+
+    scheduler_action = "unchanged"
+    scheduler = _refresh_scheduler
+    if scheduler is not None and _refresh_scheduler_leader:
+        if refresh_enabled:
+            if not scheduler.is_running():
+                scheduler.start()
+                scheduler_action = "started"
+        elif scheduler.is_running():
+            scheduler.stop()
+            scheduler_action = "stopped"
+
+    if previous_strategy != target_strategy or scheduler_action != "unchanged":
+        logger.info(
+            "account refresh runtime reconciled: previous_strategy={} strategy={} leader={} scheduler_action={}",
+            previous_strategy,
+            target_strategy,
+            _refresh_scheduler_leader,
+            scheduler_action,
+        )
+    return target_strategy
+
+
+__all__ = [
+    "get_refresh_service",
+    "set_refresh_service",
+    "get_refresh_scheduler",
+    "set_refresh_scheduler",
+    "is_refresh_scheduler_leader",
+    "set_refresh_scheduler_leader",
+    "reconcile_refresh_runtime",
+]
diff --git a/app/control/account/scheduler.py b/app/control/account/scheduler.py
index 0fa265fae..8d384d219 100644
--- a/app/control/account/scheduler.py
+++ b/app/control/account/scheduler.py
@@ -39,8 +39,16 @@ def __init__(self, refresh_service: AccountRefreshService) -> None:
         self._tasks:  list[asyncio.Task] = []
         self._stop    = asyncio.Event()
 
+    def bind_service(self, refresh_service: AccountRefreshService) -> None:
+        """Update the refresh service used by the singleton scheduler."""
+        self._service = refresh_service
+
+    def is_running(self) -> bool:
+        """Return True while any pool refresh loop is still active."""
+        return any(not task.done() for task in self._tasks)
+
     def start(self) -> None:
-        if self._tasks and not all(t.done() for t in self._tasks):
+        if self.is_running():
             return
         self._stop.clear()
         self._tasks = [
@@ -54,11 +62,14 @@ def start(self) -> None:
         )
 
     def stop(self) -> None:
+        was_running = self.is_running()
         self._stop.set()
         for t in self._tasks:
             if not t.done():
                 t.cancel()
-        logger.info("account refresh scheduler stopped")
+        self._tasks = []
+        if was_running:
+            logger.info("account refresh scheduler stopped")
 
     async def _loop(self, pool: str) -> None:
         while not self._stop.is_set():
@@ -102,6 +113,8 @@ def get_account_refresh_scheduler(
     global _scheduler
     if _scheduler is None:
         _scheduler = AccountRefreshScheduler(refresh_service)
+    else:
+        _scheduler.bind_service(refresh_service)
     return _scheduler
 
 
diff --git a/app/dataplane/account/__init__.py b/app/dataplane/account/__init__.py
index 20937f366..11c074f00 100644
--- a/app/dataplane/account/__init__.py
+++ b/app/dataplane/account/__init__.py
@@ -306,19 +306,20 @@ def revision(self) -> int:
 # ---------------------------------------------------------------------------
 
 
-_DEFAULT_COOLING_HOURS: dict[str, int] = {
-    "basic": 20,
-    "super": 2,
-    "heavy": 2,
+_POOL_INTERVAL_CONFIG: dict[str, tuple[str, int]] = {
+    "basic": ("account.refresh.basic_interval_sec", 36_000),
+    "super": ("account.refresh.super_interval_sec", 7_200),
+    "heavy": ("account.refresh.heavy_interval_sec", 7_200),
 }
 
 
 def _pool_cooling_sec(pool_id: int) -> int:
     """Cooling seconds for a 429 on a given pool (random strategy only)."""
     pool_str = POOL_ID_TO_STR.get(pool_id, "basic")
-    default_hours = _DEFAULT_COOLING_HOURS.get(pool_str, 2)
-    hours = int(get_config(f"account.random.cooling_{pool_str}_hours", default_hours))
-    return max(0, hours) * 3600
+    interval_key, default_interval = _POOL_INTERVAL_CONFIG.get(
+        pool_str, _POOL_INTERVAL_CONFIG["basic"]
+    )
+    return max(0, int(get_config(interval_key, default_interval)))
 
 
 # ---------------------------------------------------------------------------
diff --git a/app/main.py b/app/main.py
index 60bb560a2..e69498c8d 100644
--- a/app/main.py
+++ b/app/main.py
@@ -116,7 +116,12 @@ async def lifespan(app: FastAPI):
         create_repository,
         describe_repository_target,
     )
-    from app.control.account.runtime import set_refresh_service
+    from app.control.account.runtime import (
+        reconcile_refresh_runtime,
+        set_refresh_scheduler,
+        set_refresh_scheduler_leader,
+        set_refresh_service,
+    )
     from app.control.account.scheduler import get_account_refresh_scheduler
     from app.dataplane.account import get_account_directory
 
@@ -188,10 +193,8 @@ async def _sync_loop() -> None:
     #      - true  → "quota" selector + scheduler running (default).
     #      - false → "random" selector + scheduler idle (no upstream probing).
     from app.control.account.refresh import AccountRefreshService
-    from app.dataplane.account.selector import set_strategy as _set_selection_strategy
 
     refresh_enabled = _config.get_bool("account.refresh.enabled", False)
-    _set_selection_strategy("quota" if refresh_enabled else "random")
 
     refresh_svc = AccountRefreshService(repo)
     set_refresh_service(refresh_svc)
@@ -199,9 +202,13 @@ async def _sync_loop() -> None:
 
     is_leader = _try_acquire_scheduler_lock()
     scheduler = get_account_refresh_scheduler(refresh_svc)
-    strategy_name = "quota" if refresh_enabled else "random"
-    if is_leader and refresh_enabled:
-        scheduler.start()
+    set_refresh_scheduler(scheduler)
+    set_refresh_scheduler_leader(is_leader)
+    app.state.account_refresh_scheduler = scheduler
+    app.state.account_refresh_is_leader = is_leader
+
+    strategy_name = reconcile_refresh_runtime(refresh_enabled)
+    if is_leader and strategy_name == "quota":
         logger.info(
             "scheduler leader: pid={} strategy=quota active_sync_s={} idle_sync_s={}",
             os.getpid(),
@@ -252,6 +259,8 @@ async def _sync_loop() -> None:
         proxy_scheduler.stop()
         _release_scheduler_lock()
 
+    set_refresh_scheduler(None)
+    set_refresh_scheduler_leader(False)
     set_refresh_service(None)
     await repo.close()
     logger.info("application shutdown completed")
@@ -332,7 +341,10 @@ def create_app() -> FastAPI:
     # Ensure config is loaded on every request.
     @app.middleware("http")
     async def _ensure_config(request: Request, call_next):
+        from app.control.account.runtime import reconcile_refresh_runtime
+
         await _config.load()
+        reconcile_refresh_runtime()
         return await call_next(request)
 
     # Global exception handler — converts AppError to JSON.
diff --git a/app/products/openai/chat.py b/app/products/openai/chat.py
index facdd4f55..0d45bb30c 100644
--- a/app/products/openai/chat.py
+++ b/app/products/openai/chat.py
@@ -22,6 +22,7 @@
 from app.control.model.registry import resolve as resolve_model
 from app.control.model.enums import ModeId
 from app.control.account.enums import FeedbackKind
+from app.dataplane.account.selector import current_strategy
 from app.dataplane.proxy.adapters.headers import build_http_headers
 from app.dataplane.proxy import get_proxy_runtime
 from app.dataplane.proxy.adapters.session import (
@@ -106,6 +107,8 @@ def _transport_upstream_error(exc: BaseException, *, context: str) -> UpstreamEr
 async def _quota_sync(token: str, mode_id: int) -> None:
     """Fire-and-forget: fetch real quota after a successful call."""
     try:
+        if current_strategy() != "quota":
+            return
         svc = get_refresh_service()
         if svc:
             await svc.refresh_call_async(token, mode_id)
@@ -121,12 +124,20 @@ async def _quota_sync(token: str, mode_id: int) -> None:
 async def _fail_sync(
     token: str, mode_id: int, exc: BaseException | None = None
 ) -> None:
-    """Fire-and-forget: persist failure counter after a failed call."""
+    """Fire-and-forget: persist failure metadata after a failed call.
+
+    In random mode this helper must not trigger upstream quota probes. It still
+    records failures so 401 invalidation and local failure accounting continue
+    to work unchanged.
+    """
     try:
         svc = get_refresh_service()
         if svc:
             await svc.record_failure_async(token, mode_id, exc)
-            if getattr(exc, "status", None) == 429:
+            if (
+                current_strategy() == "quota"
+                and getattr(exc, "status", None) == 429
+            ):
                 result = await svc.refresh_on_demand()
                 logger.info(
                     "account on-demand refresh triggered: token={}... mode_id={} refreshed={} failed={} rate_limited={}",
diff --git a/app/products/web/admin/__init__.py b/app/products/web/admin/__init__.py
index 61494f8d9..8b87c6d7d 100644
--- a/app/products/web/admin/__init__.py
+++ b/app/products/web/admin/__init__.py
@@ -184,6 +184,8 @@ async def get_config_endpoint():
 
 @router.post("/config", tags=[_TAG_ADMIN_SYSTEM])
 async def update_config(req: ConfigPatchRequest):
+    from app.control.account.runtime import reconcile_refresh_runtime
+
     patch = _sanitize_proxy_config(req.root)
     _ensure_runtime_patch_allowed(patch)
     await config.update(patch)
@@ -195,7 +197,12 @@ async def update_config(req: ConfigPatchRequest):
         file_level=config.get_str("logging.file_level", "") or None,
         max_files=config.get_int("logging.max_files", 7),
     )
-    return {"status": "success", "message": "配置已更新"}
+    strategy_name = reconcile_refresh_runtime()
+    return {
+        "status": "success",
+        "message": "配置已更新",
+        "selection_strategy": strategy_name,
+    }
 
 
 @router.get("/storage", tags=[_TAG_ADMIN_SYSTEM])
@@ -205,8 +212,8 @@ async def get_storage_mode():
 
 @router.get("/status", tags=[_TAG_ADMIN_SYSTEM])
 async def runtime_status():
+    from app.control.account.runtime import reconcile_refresh_runtime
     from app.dataplane.account import _directory
-    from app.dataplane.account.selector import current_strategy
 
     if _directory is None:
         raise AppError(
@@ -215,13 +222,14 @@ async def runtime_status():
             code="directory_not_initialised",
             status=503,
         )
+    strategy_name = reconcile_refresh_runtime()
     return Response(
         content=orjson.dumps(
             {
                 "status": "ok",
                 "size": _directory.size,
                 "revision": _directory.revision,
-                "selection_strategy": current_strategy(),
+                "selection_strategy": strategy_name,
             }
         ),
         media_type="application/json",
diff --git a/app/statics/admin/account.html b/app/statics/admin/account.html
index 9107c5de0..d72e814c5 100644
--- a/app/statics/admin/account.html
+++ b/app/statics/admin/account.html
@@ -83,6 +83,9 @@
       gap:12px;
       margin-bottom:20px;
     }
+    .stat-grid.mode-random {
+      grid-template-columns:repeat(4, minmax(0, 1fr));
+    }
     .stat-cell {
       min-height:88px;
       padding:14px 16px;
@@ -126,6 +129,16 @@
       margin-top:auto;
     }
     .stat-muted { color:#a1a1aa !important; }
+    [data-random-only] { display:none }
+    .stat-grid.mode-random [data-random-only] { display:flex }
+    .stat-grid.mode-random [data-stat-card="active"] { order:1 }
+    .stat-grid.mode-random [data-stat-card="cooling"] { order:2 }
+    .stat-grid.mode-random [data-stat-card="invalid"] { order:3 }
+    .stat-grid.mode-random [data-stat-card="disabled"] { order:4 }
+    .stat-grid.mode-random [data-stat-card="total"] { order:5 }
+    .stat-grid.mode-random [data-stat-card="calls"] { order:6 }
+    .stat-grid.mode-random [data-stat-card="success"] { order:7 }
+    .stat-grid.mode-random [data-stat-card="rate"] { order:8 }
 
     /* Filters */
     .filter-bar {
@@ -726,9 +739,9 @@
     <div class="section-title" data-i18n="account.sectionOverview">账户概览</div>
   </div>
 
-  <!-- Stats 2×5 -->
-  <div class="stat-grid">
-    <div class="stat-cell">
+  <!-- Stats -->
+  <div class="stat-grid" id="overview-stats">
+    <div class="stat-cell" data-stat-card="total">
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statTotal">账户总数</div>
         <span class="stat-icon">
@@ -737,7 +750,7 @@
       </div>
       <div class="stat-num" id="s-total">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-stat-card="active">
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statActive">正常账户</div>
         <span class="stat-icon" style="color:#16a34a">
@@ -746,7 +759,7 @@
       </div>
       <div class="stat-num" id="s-active" style="color:#16a34a">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-stat-card="cooling">
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statCooling">限流账户</div>
         <span class="stat-icon" style="color:#ea580c">
@@ -755,7 +768,7 @@
       </div>
       <div class="stat-num" id="s-cooling" style="color:#ea580c">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-stat-card="invalid">
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statInvalid">异常账户</div>
         <span class="stat-icon" style="color:#dc2626">
@@ -764,7 +777,7 @@
       </div>
       <div class="stat-num" id="s-invalid" style="color:#dc2626">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-stat-card="disabled">
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statDisabled">禁用账户</div>
         <span class="stat-icon" style="color:#6f675d">
@@ -773,7 +786,7 @@
       </div>
       <div class="stat-num" id="s-disabled" style="color:#6f675d">0</div>
     </div>
-    <div class="stat-cell">
+    <div class="stat-cell" data-stat-card="calls">
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statCalls">调用总数</div>
         <span class="stat-icon" style="color:#111">
@@ -782,6 +795,24 @@
       </div>
       <div class="stat-num" id="s-calls" style="color:#111">0</div>
     </div>
+    <div class="stat-cell" data-stat-card="success" data-random-only>
+      <div class="stat-top">
+        <div class="stat-label" data-i18n="account.statSuccessTotal">成功总数</div>
+        <span class="stat-icon" style="color:#2563eb">
+          <svg viewBox="0 0 24 24" fill="none" stroke-width="1.8"><path d="M7 12.5 10 15.5 17 8.5"/><circle cx="12" cy="12" r="8"/></svg>
+        </span>
+      </div>
+      <div class="stat-num" id="s-success" style="color:#2563eb">0</div>
+    </div>
+    <div class="stat-cell" data-stat-card="rate" data-random-only>
+      <div class="stat-top">
+        <div class="stat-label" data-i18n="account.colRate">成功率</div>
+        <span class="stat-icon" style="color:#7c3aed">
+          <svg viewBox="0 0 24 24" fill="none" stroke-width="1.8"><path d="M6 17 10 13 13 16 18 9"/><path d="M18 9h-4"/><path d="M18 9v4"/></svg>
+        </span>
+      </div>
+      <div class="stat-num" id="s-rate" style="color:#7c3aed">0%</div>
+    </div>
     <div class="stat-cell" data-quota-only>
       <div class="stat-top">
         <div class="stat-label" data-i18n="account.statAuto">Auto 余额</div>
@@ -1161,6 +1192,8 @@
 function applyStrategyUI() {
   // Random strategy: hide quota-related stat cells and the table column.
   const isRandom = selectionStrategy === 'random';
+  const statsGrid = document.getElementById('overview-stats');
+  if (statsGrid) statsGrid.classList.toggle('mode-random', isRandom);
   document.querySelectorAll('[data-quota-only]').forEach(el => {
     el.style.display = isRandom ? 'none' : '';
   });
@@ -1179,7 +1212,19 @@
   const cacheKey = `${_tokenViewVersion}|${curStatus}|${curNsfw}|${curPool}`;
   if (_tokenViewCache && _tokenViewCacheKey === cacheKey) return _tokenViewCache;
 
-  const stats = { active: 0, cooling: 0, invalid: 0, disabled: 0, calls: 0, qa: 0, qf: 0, qe: 0, qh: 0 };
+  const stats = {
+    active: 0,
+    cooling: 0,
+    invalid: 0,
+    disabled: 0,
+    calls: 0,
+    success: 0,
+    fail: 0,
+    qa: 0,
+    qf: 0,
+    qe: 0,
+    qh: 0,
+  };
   const statusCounts = { all: 0, active: 0, cooling: 0, invalid: 0, disabled: 0 };
   const nsfwCounts = { all: 0, enabled: 0, disabled: 0 };
   const poolCounts = new Map([['all', 0]]);
@@ -1198,7 +1243,11 @@
     const matchesNsfw = curNsfw === 'all' || (curNsfw === 'enabled' ? nsfwEnabled : !nsfwEnabled);
 
     poolsSet.add(pool);
-    stats.calls += (token.use_count || 0) + (token.fail_count || 0);
+    const successCount = token.use_count || 0;
+    const failCount = token.fail_count || 0;
+    stats.success += successCount;
+    stats.fail += failCount;
+    stats.calls += successCount + failCount;
     stats.qa += quota.auto?.remaining || 0;
     stats.qf += quota.fast?.remaining || 0;
     stats.qe += quota.expert?.remaining || 0;
@@ -1268,6 +1317,8 @@
   $('s-invalid', stats.invalid);
   $('s-disabled', stats.disabled);
   $('s-calls', fmt(stats.calls));
+  $('s-success', fmt(stats.success));
+  $('s-rate', fmtRate(stats.success, stats.fail));
   $('s-qa', fmt(stats.qa)); $('s-qf', fmt(stats.qf)); $('s-qe', fmt(stats.qe)); $('s-qh', fmt(stats.qh));
 }
 
diff --git a/app/statics/admin/config.html b/app/statics/admin/config.html
index a6802007d..d6b6715d9 100644
--- a/app/statics/admin/config.html
+++ b/app/statics/admin/config.html
@@ -495,23 +495,14 @@
         title: '配额刷新调度', titleKey: 'config.schema.groups.refreshSchedule',
         section: 'account.refresh',
         fields: [
-          { key: 'enabled', label: '启用配额刷新', labelKey: 'config.schema.fields.refreshEnabled.label', type: 'bool', desc: '开启配额探测和评分选号；万级以上账号建议关闭。', descKey: 'config.schema.fields.refreshEnabled.desc', help: '开启：配额刷新模式，scheduler 周期同步真实配额，选号按评分。\n关闭：自动重试模式，不主动探测 upstream，选号随机，出错自动换号重试最多 5 次。\n建议：万级以上账号关闭，避免主动探测触发 upstream 429。', helpKey: 'config.schema.fields.refreshEnabled.help' },
-          { key: 'basic_interval_sec', label: 'Basic 刷新周期（秒）', labelKey: 'config.schema.fields.basicInterval.label', type: 'number', desc: 'basic 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 36000 s / 10 h）。', descKey: 'config.schema.fields.basicInterval.desc' },
-          { key: 'super_interval_sec', label: 'Super 刷新周期（秒）', labelKey: 'config.schema.fields.superInterval.label', type: 'number', desc: 'super 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 7200 s / 2 h）。', descKey: 'config.schema.fields.superInterval.desc' },
-          { key: 'heavy_interval_sec', label: 'Heavy 刷新周期（秒）', labelKey: 'config.schema.fields.heavyInterval.label', type: 'number', desc: 'heavy 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 7200 s / 2 h）。', descKey: 'config.schema.fields.heavyInterval.desc' },
-          { key: 'on_demand_min_interval_sec', label: '按需刷新最小间隔（秒）', labelKey: 'config.schema.fields.onDemandMinInterval.label', type: 'number', desc: '请求链路触发刷新（例如收到 429）时的节流间隔。N 秒内重复触发只执行一次，避免批量打爆配额接口。', descKey: 'config.schema.fields.onDemandMinInterval.desc' },
+          { key: 'enabled', label: '启用配额刷新', labelKey: 'config.schema.fields.refreshEnabled.label', type: 'bool', desc: '开启后自动进入配额刷新模式，关闭后自动进入自动重试模式。', descKey: 'config.schema.fields.refreshEnabled.desc', help: '开启：配额刷新模式，scheduler 周期同步真实配额，选号按评分。\n关闭：自动重试模式，不主动探测 upstream，选号随机，出错自动换号重试最多 5 次。\n建议：万级以上账号关闭，避免主动探测触发 upstream 429。', helpKey: 'config.schema.fields.refreshEnabled.help' },
+          { key: 'basic_interval_sec', label: 'Basic 周期（秒）', labelKey: 'config.schema.fields.basicInterval.label', type: 'number', desc: 'basic 号池共用周期：quota 模式下用于后台刷新，random 模式下用于 429 冷却。默认 36000s。', descKey: 'config.schema.fields.basicInterval.desc' },
+          { key: 'super_interval_sec', label: 'Super 周期（秒）', labelKey: 'config.schema.fields.superInterval.label', type: 'number', desc: 'super 号池共用周期：quota 模式下用于后台刷新，random 模式下用于 429 冷却。默认 7200s。', descKey: 'config.schema.fields.superInterval.desc' },
+          { key: 'heavy_interval_sec', label: 'Heavy 周期（秒）', labelKey: 'config.schema.fields.heavyInterval.label', type: 'number', desc: 'heavy 号池共用周期：quota 模式下用于后台刷新，random 模式下用于 429 冷却。默认 7200s。', descKey: 'config.schema.fields.heavyInterval.desc' },
+          { key: 'on_demand_min_interval_sec', label: '按需刷新间隔（秒）', labelKey: 'config.schema.fields.onDemandMinInterval.label', type: 'number', desc: '请求链路触发刷新（例如收到 429）时的节流间隔。N 秒内重复触发只执行一次，避免批量打爆配额接口。', descKey: 'config.schema.fields.onDemandMinInterval.desc' },
           { key: 'usage_concurrency', label: '刷新并发数', labelKey: 'config.schema.fields.usageConcurrency.label', type: 'number', desc: '并发调用 usage 接口刷新额度时允许的最大并行数。', descKey: 'config.schema.fields.usageConcurrency.desc' },
         ]
       },
-      {
-        title: '自动重试冷却', titleKey: 'config.schema.groups.randomCooling',
-        section: 'account.random',
-        fields: [
-          { key: 'cooling_basic_hours', label: 'Basic 冷却（小时）', labelKey: 'config.schema.fields.coolingBasicHours.label', type: 'number', desc: 'basic 号收到 429 后的冷却时长，默认 20 h（对齐 upstream 20 h 限流窗口）。仅在自动重试模式下生效。', descKey: 'config.schema.fields.coolingBasicHours.desc' },
-          { key: 'cooling_super_hours', label: 'Super 冷却（小时）', labelKey: 'config.schema.fields.coolingSuperHours.label', type: 'number', desc: 'super 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。', descKey: 'config.schema.fields.coolingSuperHours.desc' },
-          { key: 'cooling_heavy_hours', label: 'Heavy 冷却（小时）', labelKey: 'config.schema.fields.coolingHeavyHours.label', type: 'number', desc: 'heavy 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。', descKey: 'config.schema.fields.coolingHeavyHours.desc' },
-        ]
-      },
       {
         title: '选号并发限制', titleKey: 'config.schema.groups.selection',
         section: 'account.selection',
diff --git a/app/statics/i18n/de.json b/app/statics/i18n/de.json
index c75508a86..b26d1b5aa 100644
--- a/app/statics/i18n/de.json
+++ b/app/statics/i18n/de.json
@@ -160,6 +160,7 @@
     "statInvalid": "Fehlerhaft",
     "statDisabled": "Deaktiviert",
     "statCalls": "Aufrufe gesamt",
+    "statSuccessTotal": "Erfolgreiche Aufrufe",
     "statAuto": "Auto-Guthaben",
     "statFast": "Fast-Guthaben",
     "statExpert": "Expert-Guthaben",
@@ -330,7 +331,6 @@
         "features": "Kernfunktionen",
         "mediaFormat": "Medienausgabe",
         "refreshSchedule": "Quotenaktualisierung",
-        "randomCooling": "Auto-Retry-Abkühlung",
         "selection": "Auswahl-Nebenläufigkeit",
         "batchConcurrency": "Batch-Konkurrenz",
         "egressProxy": "Ausgehender Proxy",
@@ -383,16 +383,13 @@
         "customInstruction": { "label": "Globale Zusatzanweisung" },
         "imageFormat": { "label": "Bildausgabeformat" },
         "videoFormat": { "label": "Videoausgabeformat" },
-        "refreshEnabled": { "label": "Quotenaktualisierung aktivieren", "desc": "Aktiviert Quotenabfrage und bewertete Auswahl; bei mehr als 10.000 Konten empfohlen aus.", "help": "Ein: Quotenaktualisierung. Der Scheduler synchronisiert regelmäßig echte Quoten, die Auswahl erfolgt nach Bewertung.\nAus: Auto-Retry-Modus. Keine aktive upstream-Abfrage, zufällige Auswahl und automatischer Kontowechsel bei Fehlern mit bis zu 5 Wiederholungen.\nEmpfehlung: Bei mehr als 10.000 Konten ausschalten, um upstream 429 durch aktive Abfragen zu vermeiden." },
-        "coolingBasicHours": { "label": "Basic-Abkühlung (Stunden)", "desc": "Abkühldauer nach 429 für Basic-Konten, Standard 20 h (entspricht dem upstream-20-h-Fenster). Nur im Auto-Retry-Modus wirksam." },
-        "coolingSuperHours": { "label": "Super-Abkühlung (Stunden)", "desc": "Abkühldauer nach 429 für Super-Konten, Standard 2 h (entspricht dem upstream-2-h-Fenster). Nur im Auto-Retry-Modus wirksam." },
-        "coolingHeavyHours": { "label": "Heavy-Abkühlung (Stunden)", "desc": "Abkühldauer nach 429 für Heavy-Konten, Standard 2 h (entspricht dem upstream-2-h-Fenster). Nur im Auto-Retry-Modus wirksam." },
+        "refreshEnabled": { "label": "Quotenaktualisierung aktivieren", "desc": "Wenn aktiviert, wechselt das System automatisch in den Quotenaktualisierungsmodus. Wenn deaktiviert, wechselt es automatisch in den Auto-Retry-Modus.", "help": "Ein: Quotenaktualisierung. Der Scheduler synchronisiert regelmäßig echte Quoten, die Auswahl erfolgt nach Bewertung.\nAus: Auto-Retry-Modus. Keine aktive upstream-Abfrage, zufällige Auswahl und automatischer Kontowechsel bei Fehlern mit bis zu 5 Wiederholungen.\nEmpfehlung: Bei mehr als 10.000 Konten ausschalten, um upstream 429 durch aktive Abfragen zu vermeiden." },
         "maxInflight": { "label": "Max. laufende Anfragen pro Konto", "desc": "Maximale Anzahl gleichzeitig laufender Anfragen pro Konto. Konten am Limit werden von der Auswahl übersprungen (für beide Modi gemeinsam)." },
-        "basicInterval": { "label": "Basic-Aktualisierungsintervall (s)", "desc": "Geplantes Aktualisierungsintervall für den Basic-Pool. Möglichst am Quotenfenster ausrichten (Standard 36000 s / 10 h)." },
-        "superInterval": { "label": "Super-Aktualisierungsintervall (s)", "desc": "Geplantes Aktualisierungsintervall für den Super-Pool. Möglichst am Quotenfenster ausrichten (Standard 7200 s / 2 h)." },
-        "heavyInterval": { "label": "Heavy-Aktualisierungsintervall (s)", "desc": "Geplantes Aktualisierungsintervall für den Heavy-Pool. Möglichst am Quotenfenster ausrichten (Standard 7200 s / 2 h)." },
+        "basicInterval": { "label": "Basic-Zyklus (s)", "desc": "Gemeinsamer Zyklus für den Basic-Pool: wird im Quota-Modus für den Hintergrund-Refresh und im Random-Modus für die 429-Abkühlung verwendet. Standardwert: 36000s." },
+        "superInterval": { "label": "Super-Zyklus (s)", "desc": "Gemeinsamer Zyklus für den Super-Pool: wird im Quota-Modus für den Hintergrund-Refresh und im Random-Modus für die 429-Abkühlung verwendet. Standardwert: 7200s." },
+        "heavyInterval": { "label": "Heavy-Zyklus (s)", "desc": "Gemeinsamer Zyklus für den Heavy-Pool: wird im Quota-Modus für den Hintergrund-Refresh und im Random-Modus für die 429-Abkühlung verwendet. Standardwert: 7200s." },
         "usageConcurrency": { "label": "Aktualisierungs-Konkurrenz", "desc": "Maximale Anzahl paralleler usage-Aufrufe während der Quotenaktualisierung." },
-        "onDemandMinInterval": { "label": "Minimales On-Demand-Intervall (s)", "desc": "Drosselungsfenster für durch Requests ausgelöste Aktualisierungen, z. B. nach 429. Wiederholte Auslöser innerhalb des Intervalls werden zu einer Aktualisierung zusammengefasst." },
+        "onDemandMinInterval": { "label": "On-Demand-Aktualisierungsintervall (s)", "desc": "Drosselungsfenster für durch Requests ausgelöste Aktualisierungen, z. B. nach 429. Wiederholte Auslöser innerhalb des Intervalls werden zu einer Aktualisierung zusammengefasst." },
         "nsfwConcurrency": { "label": "Konkurrenz für NSFW-Aktivierung" },
         "refreshConcurrency": { "label": "Usage-Aktualisierungskonkurrenz" },
         "assetUploadConcurrency": { "label": "Asset-Upload-Konkurrenz" },
diff --git a/app/statics/i18n/en.json b/app/statics/i18n/en.json
index 2d529e25e..c2b0c066c 100644
--- a/app/statics/i18n/en.json
+++ b/app/statics/i18n/en.json
@@ -161,6 +161,7 @@
     "statInvalid": "Abnormal",
     "statDisabled": "Disabled",
     "statCalls": "Total Calls",
+    "statSuccessTotal": "Successful Calls",
     "statAuto": "Auto Balance",
     "statFast": "Fast Balance",
     "statExpert": "Expert Balance",
@@ -331,7 +332,6 @@
         "features": "Core Features",
         "mediaFormat": "Media Output",
         "refreshSchedule": "Quota Refresh",
-        "randomCooling": "Auto-Retry Cooldown",
         "selection": "Selection Concurrency",
         "batchConcurrency": "Batch Concurrency",
         "egressProxy": "Egress Proxy",
@@ -437,43 +437,31 @@
         },
         "refreshEnabled": {
           "label": "Enable Quota Refresh",
-          "desc": "Enables quota probing and scored selection. Recommended off for >10k accounts.",
+          "desc": "When enabled, the system automatically switches to quota refresh mode. When disabled, it automatically switches to auto-retry mode.",
           "help": "On: quota refresh mode. The scheduler periodically syncs real quota, and selection uses scoring.\nOff: auto-retry mode. No upstream probing, uniform random selection, and auto-swap on errors for up to 5 retries.\nRecommendation: turn this off for >10k accounts to avoid self-triggering upstream 429s."
         },
-        "coolingBasicHours": {
-          "label": "Basic Cooldown (hours)",
-          "desc": "Cooldown duration after a 429 on basic-pool accounts, default 20h (matches upstream 20h window). Only effective in auto-retry mode."
-        },
-        "coolingSuperHours": {
-          "label": "Super Cooldown (hours)",
-          "desc": "Cooldown duration after a 429 on super-pool accounts, default 2h (matches upstream 2h window). Only effective in auto-retry mode."
-        },
-        "coolingHeavyHours": {
-          "label": "Heavy Cooldown (hours)",
-          "desc": "Cooldown duration after a 429 on heavy-pool accounts, default 2h (matches upstream 2h window). Only effective in auto-retry mode."
-        },
         "maxInflight": {
           "label": "Max Inflight per Account",
           "desc": "Maximum concurrent in-flight requests per account. Accounts at this limit are skipped by the selector (shared by both modes)."
         },
         "basicInterval": {
-          "label": "Basic Refresh Interval (s)",
-          "desc": "Scheduled quota refresh interval for the basic pool. Align this with the quota window when possible (default 36000 s / 10 h)."
+          "label": "Basic Cycle (s)",
+          "desc": "Shared cycle for the basic pool: used for background refresh in quota mode and for 429 cooldown in random mode. Default 36000s."
         },
         "superInterval": {
-          "label": "Super Refresh Interval (s)",
-          "desc": "Scheduled quota refresh interval for the super pool. Align this with the quota window when possible (default 7200 s / 2 h)."
+          "label": "Super Cycle (s)",
+          "desc": "Shared cycle for the super pool: used for background refresh in quota mode and for 429 cooldown in random mode. Default 7200s."
         },
         "heavyInterval": {
-          "label": "Heavy Refresh Interval (s)",
-          "desc": "Scheduled quota refresh interval for the heavy pool. Align this with the quota window when possible (default 7200 s / 2 h)."
+          "label": "Heavy Cycle (s)",
+          "desc": "Shared cycle for the heavy pool: used for background refresh in quota mode and for 429 cooldown in random mode. Default 7200s."
         },
         "usageConcurrency": {
           "label": "Refresh Concurrency",
           "desc": "Maximum number of parallel usage calls allowed during quota refresh."
         },
         "onDemandMinInterval": {
-          "label": "Minimum On-Demand Refresh Interval (s)",
+          "label": "On-Demand Refresh Interval (s)",
           "desc": "Throttle window for request-triggered refreshes such as 429 handling. Repeated triggers within the interval are coalesced into a single refresh."
         },
         "nsfwConcurrency": {
diff --git a/app/statics/i18n/es.json b/app/statics/i18n/es.json
index 04f3c74be..eb092769d 100644
--- a/app/statics/i18n/es.json
+++ b/app/statics/i18n/es.json
@@ -160,6 +160,7 @@
     "statInvalid": "Anómalas",
     "statDisabled": "Desactivadas",
     "statCalls": "Llamadas totales",
+    "statSuccessTotal": "Éxitos totales",
     "statAuto": "Saldo Auto",
     "statFast": "Saldo Fast",
     "statExpert": "Saldo Expert",
@@ -330,7 +331,6 @@
         "features": "Funciones principales",
         "mediaFormat": "Salida multimedia",
         "refreshSchedule": "Actualización de cuota",
-        "randomCooling": "Enfriamiento de reintento automático",
         "selection": "Concurrencia de selección",
         "batchConcurrency": "Concurrencia por lotes",
         "egressProxy": "Proxy de salida",
@@ -383,16 +383,13 @@
         "customInstruction": { "label": "Instrucción suplementaria global" },
         "imageFormat": { "label": "Formato de salida de imagen" },
         "videoFormat": { "label": "Formato de salida de video" },
-        "refreshEnabled": { "label": "Activar actualización de cuota", "desc": "Activa la comprobación de cuotas y la selección por puntuación; se recomienda desactivarlo con más de 10.000 cuentas.", "help": "Activado: modo de actualización de cuota. El scheduler sincroniza periódicamente la cuota real y la selección usa puntuación.\nDesactivado: modo de reintento automático. No comprueba upstream activamente, usa selección aleatoria y cambia de cuenta al fallar hasta 5 reintentos.\nRecomendación: desactívalo con más de 10.000 cuentas para evitar provocar 429 de upstream con comprobaciones activas." },
-        "coolingBasicHours": { "label": "Enfriamiento Basic (horas)", "desc": "Duración del enfriamiento tras un 429 en cuentas Basic, por defecto 20 h (coincide con la ventana upstream de 20 h). Solo efectivo en modo de reintento automático." },
-        "coolingSuperHours": { "label": "Enfriamiento Super (horas)", "desc": "Duración del enfriamiento tras un 429 en cuentas Super, por defecto 2 h (coincide con la ventana upstream de 2 h). Solo efectivo en modo de reintento automático." },
-        "coolingHeavyHours": { "label": "Enfriamiento Heavy (horas)", "desc": "Duración del enfriamiento tras un 429 en cuentas Heavy, por defecto 2 h (coincide con la ventana upstream de 2 h). Solo efectivo en modo de reintento automático." },
+        "refreshEnabled": { "label": "Activar actualización de cuota", "desc": "Al activarlo, el sistema cambia automáticamente al modo de actualización de cuota. Al desactivarlo, cambia automáticamente al modo de reintento automático.", "help": "Activado: modo de actualización de cuota. El scheduler sincroniza periódicamente la cuota real y la selección usa puntuación.\nDesactivado: modo de reintento automático. No comprueba upstream activamente, usa selección aleatoria y cambia de cuenta al fallar hasta 5 reintentos.\nRecomendación: desactívalo con más de 10.000 cuentas para evitar provocar 429 de upstream con comprobaciones activas." },
         "maxInflight": { "label": "Máximo en curso por cuenta", "desc": "Máximo de solicitudes simultáneas en curso por cuenta. Las cuentas en ese límite se omiten en la selección (compartido por ambos modos)." },
-        "basicInterval": { "label": "Intervalo de refresco Basic (s)", "desc": "Intervalo programado de actualización para el pool Basic. Alinéalo con la ventana de cuota cuando sea posible (por defecto 36000 s / 10 h)." },
-        "superInterval": { "label": "Intervalo de refresco Super (s)", "desc": "Intervalo programado de actualización para el pool Super. Alinéalo con la ventana de cuota cuando sea posible (por defecto 7200 s / 2 h)." },
-        "heavyInterval": { "label": "Intervalo de refresco Heavy (s)", "desc": "Intervalo programado de actualización para el pool Heavy. Alinéalo con la ventana de cuota cuando sea posible (por defecto 7200 s / 2 h)." },
+        "basicInterval": { "label": "Ciclo Basic (s)", "desc": "Ciclo compartido del pool Basic: se usa para la actualización en segundo plano en modo quota y para el enfriamiento tras 429 en modo random. Valor por defecto: 36000s." },
+        "superInterval": { "label": "Ciclo Super (s)", "desc": "Ciclo compartido del pool Super: se usa para la actualización en segundo plano en modo quota y para el enfriamiento tras 429 en modo random. Valor por defecto: 7200s." },
+        "heavyInterval": { "label": "Ciclo Heavy (s)", "desc": "Ciclo compartido del pool Heavy: se usa para la actualización en segundo plano en modo quota y para el enfriamiento tras 429 en modo random. Valor por defecto: 7200s." },
         "usageConcurrency": { "label": "Concurrencia de actualización", "desc": "Número máximo de llamadas usage paralelas permitidas durante la actualización de cuota." },
-        "onDemandMinInterval": { "label": "Intervalo mínimo bajo demanda (s)", "desc": "Ventana de limitación para actualizaciones activadas por solicitudes, como el manejo de 429. Los disparos repetidos dentro del intervalo se agrupan en una sola actualización." },
+        "onDemandMinInterval": { "label": "Intervalo de actualización bajo demanda (s)", "desc": "Ventana de limitación para actualizaciones activadas por solicitudes, como el manejo de 429. Los disparos repetidos dentro del intervalo se agrupan en una sola actualización." },
         "nsfwConcurrency": { "label": "Concurrencia de activación NSFW" },
         "refreshConcurrency": { "label": "Concurrencia de refresco de Usage" },
         "assetUploadConcurrency": { "label": "Concurrencia de carga de Asset" },
diff --git a/app/statics/i18n/fr.json b/app/statics/i18n/fr.json
index 71c87c506..70246469d 100644
--- a/app/statics/i18n/fr.json
+++ b/app/statics/i18n/fr.json
@@ -160,6 +160,7 @@
     "statInvalid": "Anormaux",
     "statDisabled": "Désactivés",
     "statCalls": "Appels totaux",
+    "statSuccessTotal": "Succès totaux",
     "statAuto": "Solde Auto",
     "statFast": "Solde Fast",
     "statExpert": "Solde Expert",
@@ -330,7 +331,6 @@
         "features": "Fonctionnalités principales",
         "mediaFormat": "Sortie multimédia",
         "refreshSchedule": "Actualisation du quota",
-        "randomCooling": "Refroidissement de relance automatique",
         "selection": "Concurrence de sélection",
         "batchConcurrency": "Concurrence par lot",
         "egressProxy": "Proxy sortant",
@@ -383,16 +383,13 @@
         "customInstruction": { "label": "Instruction globale supplémentaire" },
         "imageFormat": { "label": "Format de sortie image" },
         "videoFormat": { "label": "Format de sortie vidéo" },
-        "refreshEnabled": { "label": "Activer l’actualisation du quota", "desc": "Active la sonde de quota et la sélection par score ; recommandé désactivé au-delà de 10 000 comptes.", "help": "Activé : mode d’actualisation du quota. Le scheduler synchronise périodiquement le quota réel et la sélection utilise un score.\nDésactivé : mode de relance automatique. Aucune sonde upstream active, sélection aléatoire et changement automatique de compte en cas d’erreur jusqu’à 5 relances.\nRecommandation : désactivez-le au-delà de 10 000 comptes pour éviter de déclencher des 429 upstream par les sondes actives." },
-        "coolingBasicHours": { "label": "Refroidissement Basic (heures)", "desc": "Durée de refroidissement après un 429 sur les comptes Basic, par défaut 20 h (alignée sur la fenêtre upstream de 20 h). Effectif uniquement en mode de relance automatique." },
-        "coolingSuperHours": { "label": "Refroidissement Super (heures)", "desc": "Durée de refroidissement après un 429 sur les comptes Super, par défaut 2 h (alignée sur la fenêtre upstream de 2 h). Effectif uniquement en mode de relance automatique." },
-        "coolingHeavyHours": { "label": "Refroidissement Heavy (heures)", "desc": "Durée de refroidissement après un 429 sur les comptes Heavy, par défaut 2 h (alignée sur la fenêtre upstream de 2 h). Effectif uniquement en mode de relance automatique." },
+        "refreshEnabled": { "label": "Activer l’actualisation du quota", "desc": "Lorsqu’il est activé, le système passe automatiquement en mode d’actualisation du quota. Lorsqu’il est désactivé, il passe automatiquement en mode de relance automatique.", "help": "Activé : mode d’actualisation du quota. Le scheduler synchronise périodiquement le quota réel et la sélection utilise un score.\nDésactivé : mode de relance automatique. Aucune sonde upstream active, sélection aléatoire et changement automatique de compte en cas d’erreur jusqu’à 5 relances.\nRecommandation : désactivez-le au-delà de 10 000 comptes pour éviter de déclencher des 429 upstream par les sondes actives." },
         "maxInflight": { "label": "Maximum en cours par compte", "desc": "Nombre maximal de requêtes simultanées par compte. Les comptes à cette limite sont ignorés par la sélection (partagé par les deux modes)." },
-        "basicInterval": { "label": "Intervalle d’actualisation Basic (s)", "desc": "Intervalle d’actualisation planifiée du pool Basic. À aligner avec la fenêtre de quota si possible (par défaut 36000 s / 10 h)." },
-        "superInterval": { "label": "Intervalle d’actualisation Super (s)", "desc": "Intervalle d’actualisation planifiée du pool Super. À aligner avec la fenêtre de quota si possible (par défaut 7200 s / 2 h)." },
-        "heavyInterval": { "label": "Intervalle d’actualisation Heavy (s)", "desc": "Intervalle d’actualisation planifiée du pool Heavy. À aligner avec la fenêtre de quota si possible (par défaut 7200 s / 2 h)." },
+        "basicInterval": { "label": "Cycle Basic (s)", "desc": "Cycle partagé du pool Basic : utilisé pour l’actualisation en arrière-plan en mode quota et pour le refroidissement après 429 en mode random. Valeur par défaut : 36000s." },
+        "superInterval": { "label": "Cycle Super (s)", "desc": "Cycle partagé du pool Super : utilisé pour l’actualisation en arrière-plan en mode quota et pour le refroidissement après 429 en mode random. Valeur par défaut : 7200s." },
+        "heavyInterval": { "label": "Cycle Heavy (s)", "desc": "Cycle partagé du pool Heavy : utilisé pour l’actualisation en arrière-plan en mode quota et pour le refroidissement après 429 en mode random. Valeur par défaut : 7200s." },
         "usageConcurrency": { "label": "Concurrence d’actualisation", "desc": "Nombre maximal d’appels usage parallèles autorisés pendant l’actualisation du quota." },
-        "onDemandMinInterval": { "label": "Intervalle minimal à la demande (s)", "desc": "Fenêtre de limitation pour les actualisations déclenchées par les requêtes, par exemple après un 429. Les déclenchements répétés dans l’intervalle sont regroupés en une seule actualisation." },
+        "onDemandMinInterval": { "label": "Intervalle d’actualisation à la demande (s)", "desc": "Fenêtre de limitation pour les actualisations déclenchées par les requêtes, par exemple après un 429. Les déclenchements répétés dans l’intervalle sont regroupés en une seule actualisation." },
         "nsfwConcurrency": { "label": "Concurrence d’activation NSFW" },
         "refreshConcurrency": { "label": "Concurrence d’actualisation Usage" },
         "assetUploadConcurrency": { "label": "Concurrence d’envoi Asset" },
diff --git a/app/statics/i18n/ja.json b/app/statics/i18n/ja.json
index 227deb341..512765dfe 100644
--- a/app/statics/i18n/ja.json
+++ b/app/statics/i18n/ja.json
@@ -160,6 +160,7 @@
     "statInvalid": "異常",
     "statDisabled": "無効化",
     "statCalls": "総呼び出し数",
+    "statSuccessTotal": "成功総数",
     "statAuto": "Auto 残高",
     "statFast": "Fast 残高",
     "statExpert": "Expert 残高",
@@ -330,7 +331,6 @@
         "features": "基本機能",
         "mediaFormat": "メディア出力",
         "refreshSchedule": "クォータ更新",
-        "randomCooling": "自動再試行クールダウン",
         "selection": "選択同時実行数",
         "batchConcurrency": "一括処理の並列数",
         "egressProxy": "送信プロキシ",
@@ -383,16 +383,13 @@
         "customInstruction": { "label": "グローバル補助指示" },
         "imageFormat": { "label": "画像出力形式" },
         "videoFormat": { "label": "動画出力形式" },
-        "refreshEnabled": { "label": "クォータ更新を有効化", "desc": "クォータ確認とスコア選択を有効にします。1 万件以上のアカウントでは無効化を推奨します。", "help": "オン：クォータ更新モード。scheduler が実クォータを定期同期し、スコアでアカウントを選択します。\nオフ：自動再試行モード。upstream を能動的に確認せず、ランダム選択し、エラー時は最大 5 回まで別アカウントへ自動切替します。\n推奨：1 万件以上のアカウントではオフにして、能動確認による upstream 429 を避けてください。" },
-        "coolingBasicHours": { "label": "Basic クールダウン（時間）", "desc": "Basic アカウントで 429 を受けた後のクールダウン時間。既定は 20 h（upstream の 20 h ウィンドウに対応）。自動再試行モードでのみ有効です。" },
-        "coolingSuperHours": { "label": "Super クールダウン（時間）", "desc": "Super アカウントで 429 を受けた後のクールダウン時間。既定は 2 h（upstream の 2 h ウィンドウに対応）。自動再試行モードでのみ有効です。" },
-        "coolingHeavyHours": { "label": "Heavy クールダウン（時間）", "desc": "Heavy アカウントで 429 を受けた後のクールダウン時間。既定は 2 h（upstream の 2 h ウィンドウに対応）。自動再試行モードでのみ有効です。" },
+        "refreshEnabled": { "label": "クォータ更新を有効化", "desc": "オンにすると自動的にクォータ更新モードへ切り替わり、オフにすると自動的に自動再試行モードへ切り替わります。", "help": "オン：クォータ更新モード。scheduler が実クォータを定期同期し、スコアでアカウントを選択します。\nオフ：自動再試行モード。upstream を能動的に確認せず、ランダム選択し、エラー時は最大 5 回まで別アカウントへ自動切替します。\n推奨：1 万件以上のアカウントではオフにして、能動確認による upstream 429 を避けてください。" },
         "maxInflight": { "label": "アカウントごとの同時実行上限", "desc": "1 アカウントで同時に処理中にできるリクエスト数の上限です。上限に達したアカウントは選択時にスキップされます（両モード共通）。" },
-        "basicInterval": { "label": "Basic 更新間隔（秒）", "desc": "Basic プールの定期クォータ更新間隔です。可能ならクォータウィンドウに合わせてください（既定 36000 s / 10 h）。" },
-        "superInterval": { "label": "Super 更新間隔（秒）", "desc": "Super プールの定期クォータ更新間隔です。可能ならクォータウィンドウに合わせてください（既定 7200 s / 2 h）。" },
-        "heavyInterval": { "label": "Heavy 更新間隔（秒）", "desc": "Heavy プールの定期クォータ更新間隔です。可能ならクォータウィンドウに合わせてください（既定 7200 s / 2 h）。" },
+        "basicInterval": { "label": "Basic 周期（秒）", "desc": "Basic プールの共通周期です。quota モードではバックグラウンド更新に、random モードでは 429 クールダウンに使われます。既定値は 36000s です。" },
+        "superInterval": { "label": "Super 周期（秒）", "desc": "Super プールの共通周期です。quota モードではバックグラウンド更新に、random モードでは 429 クールダウンに使われます。既定値は 7200s です。" },
+        "heavyInterval": { "label": "Heavy 周期（秒）", "desc": "Heavy プールの共通周期です。quota モードではバックグラウンド更新に、random モードでは 429 クールダウンに使われます。既定値は 7200s です。" },
         "usageConcurrency": { "label": "更新並列数", "desc": "クォータ更新中に許可する usage 呼び出しの最大並列数です。" },
-        "onDemandMinInterval": { "label": "最小オンデマンド更新間隔（秒）", "desc": "429 処理など、リクエスト経路から発生する更新のスロットル間隔です。この間隔内の重複トリガーは 1 回の更新にまとめられます。" },
+        "onDemandMinInterval": { "label": "オンデマンド更新間隔（秒）", "desc": "429 処理など、リクエスト経路から発生する更新のスロットル間隔です。この間隔内の重複トリガーは 1 回の更新にまとめられます。" },
         "nsfwConcurrency": { "label": "NSFW 有効化の並列数" },
         "refreshConcurrency": { "label": "Usage 更新の並列数" },
         "assetUploadConcurrency": { "label": "Asset アップロード並列数" },
diff --git a/app/statics/i18n/zh.json b/app/statics/i18n/zh.json
index 1f21050bf..12c64b25a 100644
--- a/app/statics/i18n/zh.json
+++ b/app/statics/i18n/zh.json
@@ -161,6 +161,7 @@
     "statInvalid": "异常账户",
     "statDisabled": "禁用账户",
     "statCalls": "调用总数",
+    "statSuccessTotal": "成功总数",
     "statAuto": "Auto 余额",
     "statFast": "Fast 余额",
     "statExpert": "Expert 余额",
@@ -331,7 +332,6 @@
         "features": "基本功能设置",
         "mediaFormat": "媒体返回格式",
         "refreshSchedule": "配额刷新调度",
-        "randomCooling": "自动重试冷却",
         "selection": "选号并发限制",
         "batchConcurrency": "批量操作并发",
         "egressProxy": "出站代理设置",
@@ -437,43 +437,31 @@
         },
         "refreshEnabled": {
           "label": "启用配额刷新",
-          "desc": "开启配额探测和评分选号；万级以上账号建议关闭。",
+          "desc": "开启后自动进入配额刷新模式，关闭后自动进入自动重试模式。",
           "help": "开启：配额刷新模式，scheduler 周期同步真实配额，选号按评分。\n关闭：自动重试模式，不主动探测 upstream，选号随机，出错自动换号重试最多 5 次。\n建议：万级以上账号关闭，避免主动探测触发 upstream 429。"
         },
-        "coolingBasicHours": {
-          "label": "Basic 冷却（小时）",
-          "desc": "basic 号收到 429 后的冷却时长，默认 20 h（对齐 upstream 20 h 限流窗口）。仅在自动重试模式下生效。"
-        },
-        "coolingSuperHours": {
-          "label": "Super 冷却（小时）",
-          "desc": "super 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。"
-        },
-        "coolingHeavyHours": {
-          "label": "Heavy 冷却（小时）",
-          "desc": "heavy 号收到 429 后的冷却时长，默认 2 h（对齐 upstream 2 h 限流窗口）。仅在自动重试模式下生效。"
-        },
         "maxInflight": {
           "label": "单号并发上限",
           "desc": "单个账号同时在执行的请求上限，超过则选号路径跳过该号（两种模式共用）。"
         },
         "basicInterval": {
-          "label": "Basic 刷新周期（秒）",
-          "desc": "basic 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 36000 s / 10 h）。"
+          "label": "Basic 周期（秒）",
+          "desc": "basic 号池共用周期：quota 模式下用于后台刷新，random 模式下用于 429 冷却。默认 36000s。"
         },
         "superInterval": {
-          "label": "Super 刷新周期（秒）",
-          "desc": "super 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 7200 s / 2 h）。"
+          "label": "Super 周期（秒）",
+          "desc": "super 号池共用周期：quota 模式下用于后台刷新，random 模式下用于 429 冷却。默认 7200s。"
         },
         "heavyInterval": {
-          "label": "Heavy 刷新周期（秒）",
-          "desc": "heavy 号池额度的定时刷新间隔，建议与配额窗口对齐（默认 7200 s / 2 h）。"
+          "label": "Heavy 周期（秒）",
+          "desc": "heavy 号池共用周期：quota 模式下用于后台刷新，random 模式下用于 429 冷却。默认 7200s。"
         },
         "usageConcurrency": {
           "label": "刷新并发数",
           "desc": "并发调用 usage 接口刷新额度时允许的最大并行数。"
         },
         "onDemandMinInterval": {
-          "label": "按需刷新最小间隔（秒）",
+          "label": "按需刷新间隔（秒）",
           "desc": "请求链路触发刷新（例如收到 429）时的节流间隔。N 秒内重复触发只执行一次，避免批量打爆配额接口。"
         },
         "nsfwConcurrency": {
diff --git a/config.defaults.toml b/config.defaults.toml
index 2e64cd878..363d164a1 100644
--- a/config.defaults.toml
+++ b/config.defaults.toml
@@ -105,20 +105,13 @@ on_codes = "429,401,503"
 [account.refresh]
 # 总开关：true=配额刷新模式（主动探测，选号评分）；false=自动重试模式（随机选号，零探测）
 enabled = true
-basic_interval_sec = 36000   # basic 号池刷新周期（秒），默认 10 h
-super_interval_sec = 7200    # super 号池刷新周期（秒），默认 2 h
-heavy_interval_sec = 7200    # heavy 号池刷新周期（秒），默认 2 h
+basic_interval_sec = 36000   # basic 号池周期（秒）：quota 模式用于后台刷新，random 模式用于 429 冷却；默认 36000s
+super_interval_sec = 7200    # super 号池周期（秒）：quota 模式用于后台刷新，random 模式用于 429 冷却；默认 7200s
+heavy_interval_sec = 7200    # heavy 号池周期（秒）：quota 模式用于后台刷新，random 模式用于 429 冷却；默认 7200s
 usage_concurrency = 50
 on_demand_min_interval_sec = 300
 
 
-[account.random]
-# 仅 account.refresh.enabled=false 时生效
-# 429 后按 pool 冷却的小时数（对齐 upstream 限流窗口）
-cooling_basic_hours = 20
-cooling_heavy_hours = 2
-
-
 [account.selection]
 # 单号并发上限（两模式共用）
 max_inflight = 8