diff --git a/achievement/apps.py b/achievement/apps.py
index 3c6023f..98a3f4f 100644
--- a/achievement/apps.py
+++ b/achievement/apps.py
@@ -4,3 +4,7 @@ from django.apps import AppConfig
 class AchievementConfig(AppConfig):
     default_auto_field = "django.db.models.BigAutoField"
     name = "achievement"
+
+    def ready(self):
+        # 导入以触发 @metric 装饰器把指标注册进 METRIC_REGISTRY
+        import achievement.metrics  # noqa: F401
diff --git a/achievement/metrics.py b/achievement/metrics.py
new file mode 100644
index 0000000..a87e02f
--- /dev/null
+++ b/achievement/metrics.py
@@ -0,0 +1,314 @@
+"""成就指标注册表。
+
+这里定义"能测量什么"，后台定义"多少算达成"。管理员在后台看到的指标下拉框
+就是 METRIC_REGISTRY 的 key 列表。加一个新维度必须改本文件并部署，
+之后在该维度上加任意多条成就都是纯配置。
+
+约定：指标从未产生过有效值时，recompute 返回 None，调用方删除该 key。
+metrics 字典里 key 不存在 == 未达标，判定时直接跳过。
+"""
+
+import logging
+
+from django.db.models import Count, Q
+from django.utils import timezone
+
+from submission.models import JudgeStatus, Submission
+
+logger = logging.getLogger(__name__)
+
+METRIC_REGISTRY = {}
+META_METRICS = set()
+
+
+class BaseMetric:
+    key = ""
+    name = ""
+    help_text = ""
+
+    def on_submission(self, metrics, sub, ctx):
+        """判题完成后增量更新 metrics（原地修改）。"""
+        raise NotImplementedError
+
+    def recompute(self, user):
+        """全量重算。返回 None 表示该用户此指标无有效值。"""
+        raise NotImplementedError
+
+
+def metric(key, name, help_text="", meta=False):
+    def deco(cls):
+        cls.key = key
+        cls.name = name
+        cls.help_text = help_text
+        METRIC_REGISTRY[key] = cls()
+        if meta:
+            META_METRICS.add(key)
+        return cls
+
+    return deco
+
+
+def _practice_submissions(user_id):
+    """成就只统计平时练习的提交，比赛提交不计入。"""
+    return Submission.objects.filter(user_id=user_id, contest_id__isnull=True)
+
+
+def build_ctx(user_id, sub):
+    """判题后预查一次，所有指标复用，避免每个指标各自查库。
+
+    比赛提交不参与成就统计，此时返回 skip=True，调用方直接跳过整轮判定。
+    """
+    if sub.contest_id is not None:
+        return {"skip": True}
+
+    prior = _practice_submissions(user_id).filter(problem_id=sub.problem_id).exclude(id=sub.id)
+    prior_stats = prior.aggregate(
+        total=Count("id"),
+        accepted=Count("id", filter=Q(result=JudgeStatus.ACCEPTED)),
+    )
+    local_now = timezone.localtime(sub.create_time)
+    return {
+        "skip": False,
+        "is_accepted": sub.result == JudgeStatus.ACCEPTED,
+        # 该题此前的提交次数与 AC 次数
+        "prior_count": prior_stats["total"],
+        "prior_accepted": prior_stats["accepted"],
+        # 首次 AC 这道题（此前从未 AC 过）
+        "is_first_ac_of_problem": sub.result == JudgeStatus.ACCEPTED and prior_stats["accepted"] == 0,
+        # 一发入魂：此前无任何提交且本次 AC
+        "is_first_try_ac": sub.result == JudgeStatus.ACCEPTED and prior_stats["total"] == 0,
+        "local_date": local_now.date().isoformat(),
+        "local_hour": local_now.hour,
+    }
+
+
+@metric("accepted_count", "AC 题目数", "去重后通过的题目数量（不含比赛）")
+class AcceptedCount(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if ctx["is_first_ac_of_problem"]:
+            metrics["accepted_count"] = metrics.get("accepted_count", 0) + 1
+
+    def recompute(self, user):
+        return _practice_submissions(user.id).filter(result=JudgeStatus.ACCEPTED).values("problem_id").distinct().count()
+
+
+@metric("submission_count", "提交总数", "提交次数（不含比赛）")
+class SubmissionCount(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        metrics["submission_count"] = metrics.get("submission_count", 0) + 1
+
+    def recompute(self, user):
+        return _practice_submissions(user.id).count()
+
+
+@metric("active_days", "活跃天数", "有过提交的累计天数")
+class ActiveDays(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        seen = metrics.get("_active_dates", [])
+        if ctx["local_date"] not in seen:
+            seen.append(ctx["local_date"])
+            metrics["_active_dates"] = seen
+            metrics["active_days"] = len(seen)
+
+    def recompute(self, user):
+        dates = {timezone.localtime(t).date().isoformat() for t in _practice_submissions(user.id).values_list("create_time", flat=True)}
+        return len(dates)
+
+
+@metric("max_ac_streak_days", "最长连续 AC 天数", "连续每天至少 AC 一题的最长天数")
+class MaxAcStreakDays(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if not ctx["is_accepted"]:
+            return
+        today = ctx["local_date"]
+        last = metrics.get("_last_ac_date")
+        if last == today:
+            return
+        current = metrics.get("_current_ac_streak", 0)
+        if last and (timezone.datetime.fromisoformat(today) - timezone.datetime.fromisoformat(last)).days == 1:
+            current += 1
+        else:
+            current = 1
+        metrics["_last_ac_date"] = today
+        metrics["_current_ac_streak"] = current
+        metrics["max_ac_streak_days"] = max(metrics.get("max_ac_streak_days", 0), current)
+
+    def recompute(self, user):
+        dates = sorted({timezone.localtime(t).date() for t in _practice_submissions(user.id).filter(result=JudgeStatus.ACCEPTED).values_list("create_time", flat=True)})
+        if not dates:
+            return None
+        best = current = 1
+        for prev, cur in zip(dates, dates[1:]):
+            current = current + 1 if (cur - prev).days == 1 else 1
+            best = max(best, current)
+        return best
+
+
+@metric("languages_used", "使用语言数", "用过多少种编程语言")
+class LanguagesUsed(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        seen = metrics.get("_languages", [])
+        if sub.language not in seen:
+            seen.append(sub.language)
+            metrics["_languages"] = seen
+            metrics["languages_used"] = len(seen)
+
+    def recompute(self, user):
+        return _practice_submissions(user.id).values("language").distinct().count()
+
+
+@metric("contest_joined", "参赛场次", "参加过的比赛数量（本指标是比赛维度，不受比赛提交不计入的限制）")
+class ContestJoined(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        # 比赛提交在 build_ctx 就被跳过，本指标只走 recompute
+        return
+
+    def recompute(self, user):
+        return Submission.objects.filter(user_id=user.id, contest_id__isnull=False).values("contest_id").distinct().count()
+
+
+@metric("badge_count", "题单奖章数", "获得的题单奖章数量")
+class BadgeCount(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        # 奖章由题单流程颁发，本指标只走 recompute，见 Task 7
+        return
+
+    def recompute(self, user):
+        from problemset.models import UserBadge
+
+        return UserBadge.objects.filter(user=user).count()
+
+
+@metric("problemset_completed", "完成题单数", "完成的题单数量")
+class ProblemSetCompleted(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        return
+
+    def recompute(self, user):
+        from problemset.models import ProblemSetProgress
+
+        return ProblemSetProgress.objects.filter(user=user, complete_time__isnull=False).count()
+
+
+@metric("first_try_ac_count", "一发入魂次数", "首次提交即通过的次数")
+class FirstTryAcCount(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if ctx["is_first_try_ac"]:
+            metrics["first_try_ac_count"] = metrics.get("first_try_ac_count", 0) + 1
+
+    def recompute(self, user):
+        count = 0
+        seen = set()
+        for s in _practice_submissions(user.id).order_by("create_time").values("problem_id", "result"):
+            if s["problem_id"] in seen:
+                continue
+            seen.add(s["problem_id"])
+            if s["result"] == JudgeStatus.ACCEPTED:
+                count += 1
+        return count
+
+
+@metric("midnight_submissions", "凌晨提交次数", "0:00–5:00 之间的提交次数")
+class MidnightSubmissions(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if 0 <= ctx["local_hour"] < 5:
+            metrics["midnight_submissions"] = metrics.get("midnight_submissions", 0) + 1
+
+    def recompute(self, user):
+        return sum(1 for t in _practice_submissions(user.id).values_list("create_time", flat=True) if 0 <= timezone.localtime(t).hour < 5)
+
+
+@metric("compile_error_count", "编译错误次数", "累计编译错误的次数")
+class CompileErrorCount(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if sub.result == JudgeStatus.COMPILE_ERROR:
+            metrics["compile_error_count"] = metrics.get("compile_error_count", 0) + 1
+
+    def recompute(self, user):
+        return _practice_submissions(user.id).filter(result=JudgeStatus.COMPILE_ERROR).count()
+
+
+@metric("max_wa_before_ac", "屡败屡战", "单题失败最多多少次后终于通过")
+class MaxWaBeforeAc(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if ctx["is_first_ac_of_problem"]:
+            metrics["max_wa_before_ac"] = max(metrics.get("max_wa_before_ac", 0), ctx["prior_count"])
+
+    def recompute(self, user):
+        best = None
+        attempts = {}
+        for s in _practice_submissions(user.id).order_by("create_time").values("problem_id", "result"):
+            pid = s["problem_id"]
+            if pid in attempts and attempts[pid] is None:
+                continue
+            if s["result"] == JudgeStatus.ACCEPTED:
+                best = max(best or 0, attempts.get(pid, 0))
+                attempts[pid] = None
+            else:
+                attempts[pid] = attempts.get(pid, 0) + 1
+        return best
+
+
+@metric("max_ac_in_one_day", "单日最多 AC", "一天之内最多通过多少题")
+class MaxAcInOneDay(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if not ctx["is_first_ac_of_problem"]:
+            return
+        counts = metrics.get("_ac_per_day", {})
+        counts[ctx["local_date"]] = counts.get(ctx["local_date"], 0) + 1
+        metrics["_ac_per_day"] = counts
+        metrics["max_ac_in_one_day"] = max(counts.values())
+
+    def recompute(self, user):
+        counts = {}
+        seen = set()
+        for s in _practice_submissions(user.id).filter(result=JudgeStatus.ACCEPTED).order_by("create_time").values("problem_id", "create_time"):
+            if s["problem_id"] in seen:
+                continue
+            seen.add(s["problem_id"])
+            day = timezone.localtime(s["create_time"]).date().isoformat()
+            counts[day] = counts.get(day, 0) + 1
+        return max(counts.values()) if counts else None
+
+
+@metric("min_ac_code_chars", "最短 AC 代码", "通过的代码里最短的字符数（配小于等于使用）")
+class MinAcCodeChars(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        if not ctx["is_accepted"]:
+            return
+        length = len(sub.code)
+        cur = metrics.get("min_ac_code_chars")
+        metrics["min_ac_code_chars"] = length if cur is None else min(cur, length)
+
+    def recompute(self, user):
+        lengths = [len(c) for c in _practice_submissions(user.id).filter(result=JudgeStatus.ACCEPTED).values_list("code", flat=True)]
+        return min(lengths) if lengths else None
+
+
+@metric("max_code_lines", "最长代码行数", "提交过的最长代码有多少行")
+class MaxCodeLines(BaseMetric):
+    def on_submission(self, metrics, sub, ctx):
+        lines = len(sub.code.splitlines())
+        metrics["max_code_lines"] = max(metrics.get("max_code_lines", 0), lines)
+
+    def recompute(self, user):
+        counts = [len(c.splitlines()) for c in _practice_submissions(user.id).values_list("code", flat=True)]
+        return max(counts) if counts else None
+
+
+@metric("achievement_unlocked_count", "已解锁成就数", "已解锁的成就数量（不含白金档）", meta=True)
+class AchievementUnlockedCount(BaseMetric):
+    """自引用指标：解锁成就会改变它。
+
+    因此判定流程限定为最多两轮（见 checker.py），且口径排除白金档自身，
+    避免「集齐 N 个成就」这类白金奖杯把自己算进分子。
+    """
+
+    def on_submission(self, metrics, sub, ctx):
+        # 由 checker 在第一轮解锁后显式重算，不参与增量更新
+        return
+
+    def recompute(self, user):
+        from achievement.models import Rarity, UserAchievement
+
+        return UserAchievement.objects.filter(user=user).exclude(achievement__rarity=Rarity.PLATINUM).count()