feat: 收敛 Sub2API 账号调优诊断

.agents/skills/unidesk-sub2api/SKILL.md

@@ -24,10 +24,19 @@ bun scripts/cli.ts platform-infra sub2api ops diagnosis --target PK01
 bun scripts/cli.ts platform-infra sub2api ops channels --target PK01 --window 7d
 bun scripts/cli.ts platform-infra sub2api plan --target PK01
 bun scripts/cli.ts platform-infra sub2api image-prepull --target PK01 --confirm
+bun scripts/cli.ts platform-infra sub2api codex-pool runtime get --target PK01 --account <id-or-exact-name>
 ```
 
 先看报表和状态，再做计划或变更。详细规则按职责拆在 `references/` 下；不要新增 `full.md`、`all.md`、`guide.md` 这类变相超级文件。
 
+- `priority` 数值越小优先级越高，`0` 合法且最高：
+  - 这是 Sub2API 调度的稳定合同，后续调优不得为确认方向重复检索源码；
+  - 只有真实调度结果与该合同冲突时，才下钻官方 scheduler 实现。
+- 查询账号周限保护统一使用 `runtime get --account <id-or-exact-name>`：
+  - 单次展示原生 5h/7d 用量、重置时间、有效自动保护阈值、阈值来源和触发状态；
+  - 账号级 `auto_pause_*_threshold` 优先于全局 Ops 默认值；
+  - 原生 quota probe 失败时明确标记不可用，并保留账号缓存字段，禁止把用户计费额度当成上游账号周限。
+
 ## 边界
 
 - PK01 默认只允许只读诊断：

.agents/skills/unidesk-sub2api/references/codex-pool.md

@@ -198,10 +198,10 @@ bun scripts/cli.ts platform-infra sub2api codex-pool cleanup-probes --target D60
 - 单账号成本标注与优先级调优使用以下最短路径：
   - 已知账号 selector 时，首个诊断入口固定为：
     `runtime errors --target <target> --since <window> --group <group> --account <account>`；
-  - 同一次账号级查询直接读取评分、可归因上游失败率、TTFT、用量、Token、标准 API 计费基数和当前名称成本标注；
-  - 只有需要确认 priority、capacity、schedulable、proxy 或临时不可调度规则时，才补一次 `runtime get --account <account>`；
+  - 同一次账号级查询直接读取评分、可归因上游失败率、TTFT、用量、Token、标准 API 计费基数、当前名称成本标注和 priority；
+  - 只有需要确认 capacity、schedulable、proxy、临时不可调度规则或上游 5h/7d 周窗口保护时，才补一次 `runtime get --account <account>`；
   - 只有需要跨账号比较时才使用全组 `runtime errors --group <group>`；单账号任务不得先跑全组大报表，再重复账号级查询；
-  - priority 数值越小越优先。成本只参与人工经济性判断，必须同时考虑可靠性、TTFT、容量、可调度状态和样本置信度，不得按成本自动排序；
+  - priority 数值越小越优先，`0` 合法且最高；这是稳定合同，除非真实调度结果冲突，不得重复检索源码确认方向。成本只参与人工经济性判断，必须同时考虑可靠性、TTFT、容量、可调度状态和样本置信度，不得按成本自动排序；
   - 名称成本标注和 priority 是两个独立 runtime 字段：分别先 dry-run，再顺序 confirm；禁止并发更新同一账号，避免整对象更新互相覆盖；
   - 两项 confirm 已各自回读一致后，最后只执行一次 `runtime get` 验收完整状态；不重复 confirm，不用全组查询代替账号回读。
 - 模型映射先用 `runtime models --account ...` 获取原生配置与实时上游能力：

docs/MDTODO/details/sub2api-upstream-reliability/R2.19_Task_Report.md

@@ -0,0 +1,20 @@
+# R2.19 任务报告
+
+## 结果
+
+- 在 Sub2API skill 高频入口固化稳定合同：`priority` 数值越小越高，`0` 合法且最高；只有真实调度结果冲突时才下钻官方 scheduler。
+- `runtime get --account <id-or-exact-name>` 现在单次展示原生 5h/7d 上游窗口、重置时间、数据来源、账号级或全局自动保护阈值、触发状态及原生 quota probe 状态。
+- 保护阈值按官方调度语义解析：账号 `auto_pause_*_threshold` 优先，全局 Ops 默认次之；旧版本接口不可用时显示 `global-settings-unavailable`，不会误报为禁用。
+- `runtime errors --account` 的账号质量表直接包含当前 priority，并标注 `lower-is-higher / highest=0`；账号下钻只显示该账号客户错误，不再混入同组其他账号。
+- runtime 文本失败现在直接显示错误原因，不再要求补跑 JSON 才能看到 `account-not-found-or-ambiguous`。
+
+## 只读验收
+
+- `runtime get --target PK01 --account 15` 一次返回账号 `lyon9801 0.0`、priority 0、原生窗口和自动保护状态。
+- 实际运行面显示 7 天原生用量 0%、账号级保护阈值 97%、未触发；这与用户提出的 95% 预期不一致，本任务未修改 runtime。
+- `runtime errors --target PK01 --since 30m --group 2 --account 15` 一次返回 priority、评分、用量、Token、成本和置信度，且不再展示其他账号客户错误。
+- 修改文件通过 Bun 语法检查、`git diff --check` 和 Sub2API skill `quick_validate.py`。
+
+## 边界
+
+未修改 Sub2API 源码、版本、账号 runtime、外部哨兵或生产配置；所有生产调用均为官方原生后端支持的受控只读查询。

docs/MDTODO/sub2api-upstream-reliability.md

@@ -234,6 +234,9 @@
 ### R2.18 [completed]
 
 基于 R2.16 优先级和 R2.17 通用 failover 模板下发后约 3 小时的新窗口，再按 PK01 Sub2API 所有分组的用户可见错误与人民币边际成本调优全局 priority：使用原生 runtime errors 聚合完整账号名称、评分、请求、Token、API USD、成本、可归因失败、TTFT、规则命中、切号和置信度，区分盈利池与自用池；只对新窗口证据充分的账号做精准 dry-run、确认写入和回读，不重复计入旧错误，不修改版本、源码、容量、代理、凭据、分组、可调度、临时规则或外部哨兵，完成任务后将详细报告写入[任务报告](./details/sub2api-upstream-reliability/R2.18_Task_Report.md)。
+### R2.19 [completed]
+
+关联 [UniDesk #2270](https://github.com/pikasTech/unidesk/issues/2270)，固化 priority 数值越小越高且 0 最高的稳定合同，并增强 Sub2API 受控 CLI 单次展示账号原生周窗口状态、YAML 周限保护阈值/来源及紧凑评分摘要，减少重复源码检索、逐账号查询和超量输出；只读验证，不修改 Sub2API 源码、版本、runtime、外部哨兵或生产运行面，完成任务后将详细报告写入[任务报告](./details/sub2api-upstream-reliability/R2.19_Task_Report.md)。
 ## R3 [completed]
 
 增加 Sub2API 客户余额精准批量充值 CLI：支持从原生最近24小时用量提取活跃用户并生成每人增加20美元的补偿计划，也支持显式用户ID或账号 selector；默认 dry-run 输出完整账号、用户ID、当前余额、增额和目标余额，写入必须显式 --confirm 并自动回读对账。本轮只实现并验证 dry-run，禁止实际充值，必须等待用户再次明确授权，完成任务后将详细报告写入[任务报告](./details/sub2api-upstream-reliability/R3_Task_Report.md)。

scripts/src/platform-infra-sub2api-codex/runtime-remote-script-head.ts

@@ -211,7 +211,110 @@ def matching_template(credentials):
             return template.get("id")
     return None
 
-def account_summary(detail, include_rules=False):
+def quota_percent(value):
+    if not isinstance(value, (int, float)):
+        return None
+    return round(value * 100, 4) if 0 <= value <= 1 else round(value, 4)
+
+def cached_codex_window(extra, label):
+    return {
+        "window": label,
+        "usedPercent": extra.get(f"codex_{label}_used_percent"),
+        "resetAfterSeconds": extra.get(f"codex_{label}_reset_after_seconds"),
+        "windowMinutes": extra.get(f"codex_{label}_window_minutes"),
+        "resetAt": extra.get(f"codex_{label}_reset_at"),
+        "source": "sub2api-account-extra",
+    }
+
+def unix_time_text(value):
+    if not isinstance(value, (int, float)) or value <= 0:
+        return value
+    try:
+        return datetime.fromtimestamp(value, tz=timezone.utc).isoformat()
+    except (ValueError, OSError, OverflowError):
+        return value
+
+def openai_quota_auto_pause_settings(token):
+    try:
+        advanced = data_of(curl_api("GET", "/api/v1/admin/ops/advanced-settings", bearer=token), "get advanced settings")
+        settings = advanced.get("openai_account_quota_auto_pause") if isinstance(advanced, dict) else None
+        return {"status": "available", "settings": settings if isinstance(settings, dict) else {}}
+    except Exception as exc:
+        return {"status": "unavailable", "settings": {}, "reason": compact_error(str(exc))}
+
+def openai_quota_summary(token, detail):
+    extra = detail.get("extra") if isinstance(detail.get("extra"), dict) else {}
+    global_result = openai_quota_auto_pause_settings(token)
+    global_settings = global_result.get("settings") or {}
+    cached_windows = [cached_codex_window(extra, label) for label in ("5h", "7d")]
+    probe = {"status": "unsupported", "source": "sub2api-native-openai-quota", "windows": []}
+    if detail.get("platform") == "openai" or detail.get("type") in ("oauth", "setup-token"):
+        try:
+            usage = data_of(curl_api("GET", f"/api/v1/admin/openai/accounts/{detail.get('id')}/quota", bearer=token), "query OpenAI quota")
+            rate_limit = usage.get("rate_limit") if isinstance(usage, dict) else None
+            rows = []
+            if isinstance(rate_limit, dict):
+                for name in ("primary_window", "secondary_window"):
+                    item = rate_limit.get(name)
+                    if not isinstance(item, dict):
+                        continue
+                    seconds = item.get("limit_window_seconds")
+                    label = "5h" if seconds == 18000 else "7d" if seconds == 604800 else name.replace("_window", "")
+                    rows.append({
+                        "window": label,
+                        "usedPercent": item.get("used_percent"),
+                        "resetAfterSeconds": item.get("reset_after_seconds"),
+                        "resetAt": unix_time_text(item.get("reset_at")),
+                        "source": "sub2api-native-openai-quota",
+                    })
+            probe = {
+                "status": "available",
+                "source": "sub2api-native-openai-quota",
+                "allowed": rate_limit.get("allowed") if isinstance(rate_limit, dict) else None,
+                "limitReached": rate_limit.get("limit_reached") if isinstance(rate_limit, dict) else None,
+                "fetchedAt": unix_time_text(usage.get("fetched_at")) if isinstance(usage, dict) else None,
+                "windows": rows,
+            }
+        except Exception as exc:
+            probe = {"status": "unavailable", "source": "sub2api-native-openai-quota", "reason": compact_error(str(exc)), "windows": []}
+    probed_by_window = {item.get("window"): item for item in probe.get("windows", []) if isinstance(item, dict)}
+    windows = [probed_by_window.get(item.get("window"), item) for item in cached_windows]
+    protections = []
+    for window in windows:
+        label = window.get("window")
+        account_threshold = extra.get(f"auto_pause_{label}_threshold")
+        global_threshold = global_settings.get(f"default_threshold_{label}")
+        if isinstance(account_threshold, (int, float)) and account_threshold > 0:
+            threshold = account_threshold
+            source = "account-extra"
+        elif isinstance(global_threshold, (int, float)) and global_threshold > 0:
+            threshold = global_threshold
+            source = "global-ops-default"
+        elif global_result.get("status") == "available":
+            threshold = None
+            source = "disabled"
+        else:
+            threshold = None
+            source = "global-settings-unavailable"
+        used = window.get("usedPercent")
+        threshold_percent = quota_percent(threshold)
+        protections.append({
+            "window": label,
+            "thresholdPercent": threshold_percent,
+            "source": source,
+            "triggered": isinstance(used, (int, float)) and isinstance(threshold_percent, (int, float)) and used >= threshold_percent,
+        })
+    return {
+        "windows": windows,
+        "cachedWindows": cached_windows,
+        "protections": protections,
+        "nativeProbe": probe,
+        "globalSettings": {"status": global_result.get("status"), "reason": global_result.get("reason")},
+        "updatedAt": extra.get("codex_usage_updated_at"),
+        "attribution": "Window values prefer the native read-only quota probe and fall back to account cache. Protection uses account threshold first, then the global Ops default.",
+    }
+
+def account_summary(detail, include_rules=False, token=None):
     credentials = detail.get("credentials") if isinstance(detail.get("credentials"), dict) else {}
     policy = normalize_policy(credentials)
     name = detail.get("name")
@@ -228,6 +331,7 @@ def account_summary(detail, include_rules=False):
         "currentConcurrency": detail.get("current_concurrency"),
         "loadFactor": detail.get("load_factor"),
         "priority": detail.get("priority"),
+        "prioritySemantics": {"order": "lower-is-higher", "highest": 0},
         "tempUnschedulable": {key: policy[key] for key in ("enabled", "ruleCount", "statusCodes")},
         "matchingTemplate": matching_template(credentials),
         "poolModeEnabled": credentials.get("pool_mode") is True,
@@ -235,6 +339,7 @@ def account_summary(detail, include_rules=False):
     }
     if include_rules:
         result["tempUnschedulable"]["rules"] = policy["rules"]
+        result["upstreamLimits"] = openai_quota_summary(token, detail) if token else None
     return result
 
 def list_proxies(token):

scripts/src/platform-infra-sub2api-codex/runtime-remote-script-tail.ts

@@ -1,5 +1,5 @@
 export function runtimeRemoteScriptTail(): string {
-  return `def native_customer_error_profile(token, group_id, platform):
+  return `def native_customer_error_profile(token, group_id, platform, account_ids=None):
     start_time = quote(since_start_time(PAYLOAD.get("since")), safe="")
     platform_query = "&platform=" + quote(str(platform), safe="") if platform else ""
     page = 1
@@ -11,7 +11,8 @@ export function runtimeRemoteScriptTail(): string {
             response = curl_api("GET", f"/api/v1/admin/ops/request-errors?group_id={group_id}&start_time={start_time}&view=all&page={page}&page_size={page_size}{platform_query}", bearer=token)
             data = data_of(response, "get customer-visible request errors")
             page_rows = items_of(data)
-            rows.extend(page_rows)
+            selected_rows = [row for row in page_rows if not account_ids or str(row.get("account_id")) in account_ids]
+            rows.extend(selected_rows)
             if total is None and isinstance(data, dict) and isinstance(data.get("total"), int):
                 total = data.get("total")
             if not page_rows or len(page_rows) < page_size or (isinstance(total, int) and len(rows) >= total):
@@ -19,7 +20,9 @@ export function runtimeRemoteScriptTail(): string {
             page += 1
     except Exception as exc:
         return {"status": "unavailable", "source": "sub2api-native-admin-ops-request-errors", "reason": compact_error(str(exc))}
-    if total is None:
+    if account_ids:
+        total = len(rows)
+    elif total is None:
         total = len(rows)
     def buckets(key_fn):
         counts = {}
@@ -273,6 +276,7 @@ def account_quality_scores(token, accounts, group_id, native_ops, effectiveness)
         rows.append({
             "accountId": account_id, "accountName": account.get("name"), "status": account.get("status"),
             "schedulable": account.get("schedulable"), "currentlyAvailable": current_available,
+            "priority": account.get("priority"), "priorityOrder": "lower-is-higher",
             "score": score, "grade": grade, "assessment": assessment, "confidence": confidence,
             "scoreComparable": score_comparable,
             "observedAttempts": observed_attempts, "successRequests": success_requests,
@@ -816,7 +820,7 @@ def observed_runtime_errors(token, accounts, group_id, platform=None):
     overview_summary = ((native_ops.get("overview") or {}).get("summary") or {})
     native_ops["responseHeaderTimeout"] = runtime_response_header_timeout()
     native_ops["requestTiming"] = native_request_timing(token, group_id, platform, overview_summary)
-    native_ops["customerErrors"] = native_customer_error_profile(token, group_id, platform)
+    native_ops["customerErrors"] = native_customer_error_profile(token, group_id, platform, selected_ids if selector is not None else None)
     native_ops["accountQuality"] = account_quality_scores(token, selected, group_id, native_ops, effectiveness)
     request_total = overview_summary.get("request_count_total")
     error_total = overview_summary.get("error_count_total")
@@ -1204,7 +1208,7 @@ def runtime_result():
         return {**base, "ok": True, "operation": "errors", "mutation": False, "errors": errors}
     detail = find_account(token, accounts, str(PAYLOAD["account"])) if action in ("get", "infrastructure") else None
     if action == "get":
-        return {**base, "ok": True, "operation": "get", "mutation": False, "account": account_summary(detail, True)}
+        return {**base, "ok": True, "operation": "get", "mutation": False, "account": account_summary(detail, True, token)}
     if action == "infrastructure":
         return {
             **base,

scripts/src/platform-infra-sub2api-codex/runtime-render.ts

@@ -227,6 +227,8 @@ function compactAccountQuality(value: unknown, detailed: boolean): unknown {
       accountId: account.accountId,
       accountName: account.accountName,
       status: account.status,
+      priority: account.priority,
+      priorityOrder: account.priorityOrder,
       currentlyAvailable: account.currentlyAvailable,
       score: account.score,
       grade: account.grade,
@@ -254,6 +256,8 @@ function compactAccountQuality(value: unknown, detailed: boolean): unknown {
     } : {
       accountId: account.accountId,
       accountName: account.accountName,
+      priority: account.priority,
+      priorityOrder: account.priorityOrder,
       currentlyAvailable: account.currentlyAvailable,
       score: account.score,
       grade: account.grade,
@@ -311,6 +315,11 @@ export function renderRuntimeResult(response: Record<string, unknown>, options:
   const runtime = runtimeRecord(response.runtime) ?? {};
   const ok = response.ok === true;
   const lines: string[] = [];
+  if (runtime.ok !== true && typeof runtime.error === "string") {
+    lines.push(`SUB2API RUNTIME ${options.action.toUpperCase()} ok=false`, `ERROR: ${runtime.error}`);
+    lines.push("", `JSON: bun scripts/cli.ts platform-infra sub2api codex-pool runtime ${options.action} <same-options> --json`);
+    return renderedCliResult(ok, `platform-infra sub2api codex-pool runtime ${options.action}`, lines.join("\n"));
+  }
   if (options.action === "errors") renderRuntimeErrors(lines, runtime, options);
   else if (options.action === "infrastructure") renderRuntimeInfrastructure(lines, runtime);
   else if (options.action === "list") renderRuntimeAccountList(lines, runtime);
@@ -406,9 +415,9 @@ function renderRuntimeErrors(lines: string[], runtime: Record<string, unknown>,
   const qualityAccounts = runtimeRecords(accountQuality.accounts);
   if (qualityAccounts.length > 0) {
     lines.push("", "ACCOUNT QUALITY", renderTable([
-      ["ACCOUNT", "ID", "SCORE", "GRADE", "QUALITY", "CONF", "ATTEMPTS", "SUCCESS", "FAIL", "UP_CUST", "EXCLUDED", "FAIL_RATE", "TTFT_P95_MS", "TTFT_N", "AVAILABLE", "REASONS"],
+      ["ACCOUNT", "ID", "PRIORITY", "SCORE", "GRADE", "QUALITY", "CONF", "ATTEMPTS", "SUCCESS", "FAIL", "UP_CUST", "EXCLUDED", "FAIL_RATE", "TTFT_P95_MS", "TTFT_N", "AVAILABLE", "REASONS"],
       ...qualityAccounts.map((item) => [
-        runtimeText(item.accountName), runtimeText(item.accountId), runtimeText(item.score), runtimeText(item.grade), runtimeText(item.assessment), runtimeText(item.confidence),
+        runtimeText(item.accountName), runtimeText(item.accountId), runtimeText(item.priority), runtimeText(item.score), runtimeText(item.grade), runtimeText(item.assessment), runtimeText(item.confidence),
         runtimeText(item.observedAttempts), runtimeText(item.successRequests), runtimeText(item.failureRequests), runtimeText(item.scoreableUpstreamErrorRequests),
         runtimeText(item.excludedNonUpstreamErrorRequests), runtimeRate(item.failureRate),
         runtimeText(item.ttftP95Ms), runtimeText(item.firstTokenSamples), runtimeText(item.currentlyAvailable),
@@ -439,7 +448,7 @@ function renderRuntimeErrors(lines: string[], runtime: Record<string, unknown>,
       ]),
     ]));
     const scoreRule = runtimeRecord(accountQuality.scoreRule) ?? {};
-    lines.push(`SCORE RULE reliability=${runtimeText(scoreRule.reliabilityWeight)} latency=${runtimeText(scoreRule.latencyWeight)} availability=${runtimeText(scoreRule.availabilityWeight)} confidence=separate`);
+    lines.push(`SCORE RULE reliability=${runtimeText(scoreRule.reliabilityWeight)} latency=${runtimeText(scoreRule.latencyWeight)} availability=${runtimeText(scoreRule.availabilityWeight)} confidence=separate priority=lower-is-higher highest=0`);
     if (typeof accountQuality.attribution === "string") lines.push(`NOTE: ${accountQuality.attribution}`);
   }
   const errorModels = runtimeRecords(customerErrors.modelBuckets);
@@ -535,6 +544,12 @@ function renderRuntimeAccountList(lines: string[], runtime: Record<string, unkno
 function renderRuntimeAccountDetail(lines: string[], runtime: Record<string, unknown>): void {
   const account = runtimeRecord(runtime.account) ?? {};
   const temp = runtimeRecord(account.tempUnschedulable) ?? {};
+  const priority = runtimeRecord(account.prioritySemantics) ?? {};
+  const limits = runtimeRecord(account.upstreamLimits) ?? {};
+  const probe = runtimeRecord(limits.nativeProbe) ?? {};
+  const globalSettings = runtimeRecord(limits.globalSettings) ?? {};
+  const windows = runtimeRecords(limits.windows);
+  const protections = runtimeRecords(limits.protections);
   lines.push(`SUB2API RUNTIME ACCOUNT ok=${runtime.ok === true ? "true" : "false"}`);
   lines.push(renderTable([
     ["NAME", "ID", "MANAGEMENT", "STATUS", "SCHED", "PROXY", "CONCURRENCY", "LOAD", "PRIORITY", "TEMPLATE"],
@@ -544,10 +559,39 @@ function renderRuntimeAccountDetail(lines: string[], runtime: Record<string, unk
       runtimeText(account.loadFactor), runtimeText(account.priority), runtimeText(account.matchingTemplate),
     ],
   ]));
+  lines.push(renderTable([
+    ["PRIORITY_ORDER", "HIGHEST_PRIORITY"],
+    [runtimeText(priority.order), runtimeText(priority.highest)],
+  ]));
   lines.push(renderTable([
     ["TEMP_ENABLED", "RULES", "STATUS_CODES", "POOL_MODE"],
     [runtimeText(temp.enabled), runtimeText(temp.ruleCount), runtimeText(temp.statusCodes), runtimeText(account.poolModeEnabled)],
   ]));
+  lines.push("", "UPSTREAM LIMITS");
+  lines.push(renderTable([
+    ["WINDOW", "USED", "RESET_AFTER", "RESET_AT", "DATA_SOURCE", "PROTECT_AT", "PROTECT_SOURCE", "TRIGGERED"],
+    ...windows.map((window) => {
+      const protection = protections.find((item) => item.window === window.window) ?? {};
+      return [
+        runtimeText(window.window), runtimePercent(window.usedPercent), runtimeSeconds(window.resetAfterSeconds),
+        runtimeText(window.resetAt), runtimeText(window.source), runtimePercent(protection.thresholdPercent),
+        runtimeText(protection.source), runtimeText(protection.triggered),
+      ];
+    }),
+  ]));
+  lines.push(`NATIVE_QUOTA status=${runtimeText(probe.status)} allowed=${runtimeText(probe.allowed)} limitReached=${runtimeText(probe.limitReached)} fetchedAt=${runtimeText(probe.fetchedAt)}`);
+  if (typeof probe.reason === "string") lines.push(`NATIVE_QUOTA_REASON: ${probe.reason}`);
+  lines.push(`GLOBAL_AUTO_PAUSE status=${runtimeText(globalSettings.status)}`);
+  if (typeof globalSettings.reason === "string") lines.push(`GLOBAL_AUTO_PAUSE_REASON: ${globalSettings.reason}`);
+  if (typeof limits.attribution === "string") lines.push(`NOTE: ${limits.attribution}`);
+}
+
+function runtimePercent(value: unknown): string {
+  return typeof value === "number" && Number.isFinite(value) ? `${value.toFixed(2)}%` : "-";
+}
+
+function runtimeSeconds(value: unknown): string {
+  return typeof value === "number" && Number.isFinite(value) ? `${Math.round(value)}s` : "-";
 }
 
 function renderRuntimeInfrastructure(lines: string[], runtime: Record<string, unknown>): void {