diff --git a/benchmarks/PowerSystems/README.md b/benchmarks/PowerSystems/README.md
index 6a3f9d2e..e8780e55 100644
--- a/benchmarks/PowerSystems/README.md
+++ b/benchmarks/PowerSystems/README.md
@@ -9,3 +9,7 @@ Current tasks emphasize realistic operational constraints, economic objectives,
   - Unified benchmark: `task=unified task.benchmark=PowerSystems/EV2GymSmartCharging`
   - Quick run: `python -m frontier_eval task=unified task.benchmark=PowerSystems/EV2GymSmartCharging task.runtime.env_name=frontier-eval-driver algorithm.iterations=0`
   - Description: upstream-aligned EV smart charging with transformer constraints in the real `EV2Gym` simulator
+- `TelecomBackup`
+  - Unified benchmark: `task=unified task.benchmark=PowerSystems/TelecomBackup`
+  - Quick run: `python -m frontier_eval task=unified task.benchmark=PowerSystems/TelecomBackup algorithm.iterations=0`
+  - Description: time-sequenced on/off scheduling of telecom backup power supplies to maximize outage backup time while keeping LTE coverage >= 80%
diff --git a/benchmarks/PowerSystems/README_zh-CN.md b/benchmarks/PowerSystems/README_zh-CN.md
index 30fcfd26..10f5e215 100644
--- a/benchmarks/PowerSystems/README_zh-CN.md
+++ b/benchmarks/PowerSystems/README_zh-CN.md
@@ -9,3 +9,7 @@
   - `frontier_eval` 任务：`task=unified task.benchmark=PowerSystems/EV2GymSmartCharging`
   - 快速运行：`python -m frontier_eval task=unified task.benchmark=PowerSystems/EV2GymSmartCharging task.runtime.env_name=frontier-eval-driver algorithm.iterations=0`
   - 简介：在真实上游 `EV2Gym` 模拟器中进行、与上游数据对齐的 EV 智能充电与变压器约束优化
+- `TelecomBackup`
+  - `frontier_eval` 任务：`task=unified task.benchmark=PowerSystems/TelecomBackup`
+  - 快速运行：`python -m frontier_eval task=unified task.benchmark=PowerSystems/TelecomBackup algorithm.iterations=0`
+  - 简介：电信备电电源的时序开关调度，在停电时最大化备电时长同时保持 LTE 覆盖 ≥ 80%
diff --git a/benchmarks/PowerSystems/TelecomBackup/.gitignore b/benchmarks/PowerSystems/TelecomBackup/.gitignore
new file mode 100644
index 00000000..7a60b85e
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/.gitignore
@@ -0,0 +1,2 @@
+__pycache__/
+*.pyc
diff --git a/benchmarks/PowerSystems/TelecomBackup/README.md b/benchmarks/PowerSystems/TelecomBackup/README.md
new file mode 100644
index 00000000..b3801afe
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/README.md
@@ -0,0 +1,164 @@
+# TelecomBackup: Power-Backup Scheduling for Telecom Sites (Frontier-Eng Benchmark)
+
+An **original** Frontier-Engineering benchmark: given a region of telecom sites powered by
+batteries, a solver must produce a **time-sequenced on/off schedule for every power supply** that
+maximizes the region's total backup time while keeping good LTE coverage (RSRP > -105 dBm) above
+80% at every moment. The power-consumption parameters live in each instance and are calibrated so
+that stagger/rotation scheduling has clear, reproducible headroom over the naive always-on strategy
+(see "Scoring").
+
+The full game rules and evaluation semantics are in [Task.md](./Task.md) (Chinese).
+
+## Layout
+
+```
+benchmarks/PowerSystems/TelecomBackup/
+├── baseline/solver.py          # Candidate solver (EVOLVE-BLOCK region is the only editable part)
+├── verification/
+│   ├── generator.py            # Fixed-seed instance generator
+│   ├── simulator.py            # Scoring simulator (coverage/power/battery simulation)
+│   ├── evaluate.py             # Evaluation entry (subprocess + time budget + scoring)
+│   ├── validator.py            # Integrity checks (static + env stripping + determinism)
+│   ├── ref_solver.py           # Reference heuristic (rest-rotation) — documented "good" score
+│   ├── test_simulator.py       # Unit tests: simulator correctness
+│   ├── test_validator.py       # Unit tests: integrity checks / env stripping / determinism
+│   ├── test_evaluator.py       # Unit tests: end-to-end evaluation behavior
+│   ├── data/instances/         # 8 fixed instances (seed-fixed, reproducible)
+│   └── requirements.txt
+├── frontier_eval/              # UnifiedTask metadata (ConnectFour/AntGame pattern)
+├── Task.md                     # Task rules, interface, scoring, reference scores
+└── README.md
+```
+
+## Requirements
+
+- Python >= 3.10, standard library only (no third-party dependencies).
+- Runtime is pure-Python simulation; each instance evaluation takes well under a second for
+  the baseline solver.
+
+## Run
+
+```powershell
+# Score a solver on the fixed 8-instance set (default 60s time budget per instance)
+python verification/evaluate.py baseline/solver.py
+
+# Add runtime-generated instances (anti-hardcoding)
+python verification/evaluate.py baseline/solver.py --generate-seed <SEED>
+
+# Tighter budget (challenge tier: 10s)
+python verification/evaluate.py baseline/solver.py --time-budget 10
+```
+
+### Docker
+
+The evaluator is pure stdlib, so a minimal `python` image suffices. Build it and
+use the unified runtime's `isolation_mode=docker`:
+
+```bash
+# Build (inside the TelecomBackup directory)
+docker build -t telecombackup-benchmark -f verification/docker/Dockerfile .
+
+# From the repo root
+python -m frontier_eval task=unified task.benchmark=PowerSystems/TelecomBackup algorithm.iterations=0 \
+  task.runtime.isolation_mode=docker task.runtime.docker_image=telecombackup-benchmark
+```
+
+> Docker isolation is validated on Linux / WSL. `frontier_eval/eval_command.txt`
+> injects the host-benchmark path via the `{benchmark_source}` placeholder, so
+> scoring works without framework changes; if the container user cannot write
+> the evaluation sandbox, set `task.runtime.docker_user=<host uid>:<host gid>`
+> (e.g. `1000:1000`). On Windows hosts the unified docker path is blocked by a
+> framework path bug (`Path.resolve()` rewrites container paths to drive
+> paths) — run docker mode under WSL instead.
+
+## Tests
+
+```powershell
+# From the TelecomBackup task directory (stdlib unittest, no dependencies)
+python -m unittest discover -s verification -p "test_*.py"
+```
+
+34 tests across four modules (simulator / validator / evaluator / sandbox): simulator correctness (manual golden cases, interval
+normalization, battery depletion, coverage constraint, determinism), validator integrity
+(EVOLVE-BLOCK / forbidden references / absolute paths / per-instance hardcoding / env
+stripping / determinism probe), and evaluator behavior (scoring, malformed/timeout/preflight
+handling, runtime generation, reproducibility).
+
+To run inside the Frontier-Eng framework (unified task):
+
+```powershell
+# Windows: point the unified runtime at the venv python (WSL bash can't run Windows exes,
+# so also use an MSYS2/Git bash instead of the default `bash`). Set PYTHONUTF8=1 to avoid
+# GBK decoding crashes in some framework libs, and raise the LLM timeout for thinking models.
+$env:PYTHONUTF8 = "1"
+$env:FRONTIER_EVAL_UNIFIED_PYTHON = "<repo>\.venvs\frontier-eval-driver\Scripts\python.exe"
+$env:TELECOM_EVAL_GENERATE_SEED = "<SEED>"  # runtime-generated instances (anti-hardcoding); 勿用固定值
+python -m frontier_eval task=unified task.benchmark=PowerSystems/TelecomBackup algorithm.iterations=0 "task.runtime.shell=<path to Git Bash>" llm.timeout=300
+```
+
+Note: evaluation spawns solver subprocesses with a time budget; if running very slow solvers,
+increase `FRONTIER_EVAL_EVALUATOR_TIMEOUT_S` accordingly (e.g. 1200).
+
+## Integrity / threat model
+
+- **Runtime-generated instances**: with `TELECOM_EVAL_GENERATE_SEED` set, the evaluator
+  generates fresh instances at evaluation time (temp dir, never in the repo/sandbox), so a
+  candidate cannot pre-position solutions for them.
+- **Candidate env stripping**: candidate subprocesses get `FRONTIER_*` / `TELECOM_EVAL_*`
+  variables stripped (see `verification/validator.py`), closing the host-env side channel.
+- **Static checks**: EVOLVE-BLOCK markers + fixed-region byte diff vs the initial baseline,
+  forbidden imports of evaluation/generation modules, absolute paths, per-instance hardcoding,
+  plus a determinism probe (two runs must match). Any violation scores 0.
+- Honest note: in process mode the candidate has host filesystem access (framework-wide
+  limitation); this benchmark relies on the layered defenses above. `verification/simulator.py`
+  is intentionally exposed as a white-box scorer for candidate-side search.
+- **Sandbox scope** (design trade-off): the 8 fixed instances and the evaluator/validator
+  sources are visible to the candidate during evolution (they are needed for scoring and the
+  simulator is intentionally usable). Anti-hardcoding therefore relies on
+  `TELECOM_EVAL_GENERATE_SEED` (fresh instances at evaluation time — set it, do not use a
+  fixed seed); the name-keyed hardcoding check is best-effort (array-index dispatch can evade
+  it, as in any static check). The reference solver (`ref_solver.py`) and the generator are
+  **not** copied into the sandbox and are additionally forbidden by the validator.
+
+## Scoring
+
+- Instances = 8 fixed (N = 20..40 sites, K = 6..12 power supplies, each instance carries its
+  power-consumption params `p_silent` / `p_work_base` / `p_work_coef`) + runtime-generated when
+  `TELECOM_EVAL_GENERATE_SEED` is set; score = mean backup time (minutes).
+- Malformed output / out-of-range intervals / crash / timeout ⇒ 0 points for that instance.
+- **Power calibration**: instances are generated with `p_silent=0.05`, `p_work_base=3.0`,
+  `p_work_coef=3.0` (silent is cheap, working is expensive), and the generator accepts only
+  instances where a multi-rest stagger (rest 1..3 supplies at a time) beats "always-on" by ≥ 25%
+  per instance — so the scheduling problem has large, reproducible headroom (verified: +28%..+119%,
+  avg +54%).
+- Reference scores (measured on the fixed 8 instances, deepseek-v4-flash agents; agent scores are
+  **verified by directly evaluating the saved programs** from the task directory — candidate
+  solvers resolve `verification/simulator.py` relative to their own location, so re-evaluating a
+  saved program from an arbitrary path silently degrades it to the always-on fallback). **The
+  published agent scores are on the fixed 8 instances only** (no `TELECOM_EVAL_GENERATE_SEED`
+  was set for those runs); set the seed to additionally score fresh instances, which is the
+  recommended anti-hardcoding configuration:
+  - baseline (always-on, no scheduling): **176.2** minutes
+  - agent (openevolve, 25 iterations, best saved program): **414.4** minutes (+135%; run `20260816_130700`); on fixed 8 + 8 generated (seed 42): **410.9**
+  - agent (ShinkaEvolve, 15 generations, best generation program): **312.5** minutes (+77%; run `20260816_214014`, gen 3); on fixed 8 + 8 generated (seed 42): **319.4**
+  - agent (AB-MCTS, 15 iterations, best saved program): **266.9** minutes (+52%; run `20260816_220646`); on fixed 8 + 8 generated (seed 42): **280.3**
+  - reference heuristic (`verification/ref_solver.py`, multi-rest rotation): **271.2** minutes (+54%)
+  - **multi-run statistics** (3 runs per framework, best valid saved program per run):
+    - openevolve (25 iterations): 414.4 / 298.1 / 357.5 → **mean 356.7 ± 47.5**
+    - ShinkaEvolve (15 generations): 312.5 / 357.5 / 325.6 → **mean 331.9 ± 18.9**
+    - AB-MCTS (15 iterations): 266.9 / 208.8 / 227.5 → **mean 234.4 ± 24.2**
+  - horizon: 480 minutes (upper bound if coverage never fails)
+  - note: with 5-15 iterations agents mostly plateau at the baseline; more iterations let all
+    three frameworks discover stagger/coverage-driven schedules that beat it (openevolve even
+    found a near-horizon minimal-covering-subset schedule, surpassing the reference heuristic).
+    ShinkaEvolve's 25-generation run produced a higher-scoring but **non-deterministic** program
+    (time-budgeted simulated annealing) that fails the determinism probe — only deterministic
+    programs count (312.5 from the 15-generation run is the valid best).
+
+## Time Budget Tiers (from the original problem)
+
+| Tier | --time-budget | Notes |
+|---|---|---|
+| Base | 300 s | T+1: 10-min solve |
+| Advanced (default) | 60 s | T+2: 1-min solve |
+| Challenge | 10 s | T+3: 10-s solve |
diff --git a/benchmarks/PowerSystems/TelecomBackup/Task.md b/benchmarks/PowerSystems/TelecomBackup/Task.md
new file mode 100644
index 00000000..15795f81
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/Task.md
@@ -0,0 +1,126 @@
+# 区域备电栅格级优化（TelecomBackup）
+
+## 任务概述
+
+现网某片区域停电时，需要基于各基站（站点）的备电电量、功耗与覆盖关系，设置**每个电源的时序开关策略**，
+在"任意时刻良好覆盖栅格比例 ≥ 80%"的约束下，**最大化区域总备电时长**。
+
+本质是一个受限的调度/优化问题：覆盖约束让问题解空间巨大，因此要求在给定的求解时间预算内
+给出尽可能优的开关调度。
+
+## 实例输入
+
+评测会传入一个实例 JSON（路径作为命令行参数），包含：
+
+| 字段                           | 含义                                    |
+| ---------------------------- | ------------------------------------- |
+| `grid`                       | 区域 200m×200m，栅格化为 nx×ny（默认 20×20=400） |
+| `sites`                      | 站点坐标列表（N 个）                           |
+| `groups`                     | 电源分组：每个电源管理一组站点（K 个电源，每组共享一块电池）       |
+| `battery`                    | 每块电池电量（kWh）                           |
+| `demand`                     | 每个栅格的需求（相对负载单位，0\~1）                  |
+| `pt_dbm / n_exp / threshold` | 覆盖参数：Pt=20dBm、路径损耗指数 n=6、阈值 -105dBm   |
+| `coverage_ratio`             | 覆盖约束比例 0.8                            |
+| `delta_min`                  | 时隙长度 5 分钟                             |
+| `horizon`                    | 时隙总数 96（=8 小时规划期）                     |
+| `site_cap`                   | 站点额定容量（归一化负载用）                        |
+
+## 规则（评分模拟器语义）
+
+- **覆盖**：站点 s 对栅格 g 的电平 `P(g,s) = pt_dbm - 10*n_exp*log10(d(g,s)+1)`（dBm，d 为米）。
+  栅格接入**最强存活站点**（存活 = 所属电源开启且电池未耗尽）；栅格"良好" ⟺ 覆盖电平 > -105dBm。
+- **功耗**：站点工作功耗 `p_work_base + p_work_coef*min(load_s/site_cap, 1.0)` kW（load\_s 为其覆盖栅格需求之和，负载相关；当前实例 `p_work_base=3.0, p_work_coef=3.0`）；
+  站点静默功耗 **`p_silent`** kW（当前实例 `p_silent=0.05`；电源关闭时其下站点静默：不提供覆盖，但仍耗静默电）。功耗参数随实例提供（见实例 JSON 与 README）。
+- **负载迁移**：电源关闭 → 站点静默 → 其覆盖栅格**即时**接入最强存活站点 → 接收站点负载与功耗上升。
+- **电量**：电源 k 每时隙扣电 `Δt × Σ(站点功耗)`；电量 ≤ 0 → 该电源停服（不耗电、不覆盖）。
+- **备电时长**：从时隙 0 起模拟推进，首个"良好栅格比例 < 80%"的时隙的前一时刻即为备电终点；
+  撑满 horizon 则备电时长 = horizon×5 分钟。
+
+## 决策输出（你的程序必须输出的格式）
+
+运行方式：`python baseline/solver.py <instance.json>`，在 **stdout** 打印一个 JSON：
+
+```json
+{"on": [
+  [[0, 96]],                  // 电源 0：全程开启
+  [[0, 40], [60, 96]],        // 电源 1：0..39 与 60..95 开启
+  []                          // 电源 2：从不开启
+]}
+```
+
+- `on[k]` 是电源 k 的**开启时隙区间**列表，每个区间为半开区间 `[a, b)`（0-index，0 ≤ a ≤ b ≤ horizon）。
+- 电源 k 在时隙 s 开启 ⟺ 存在区间使 `a ≤ s < b`。相邻/重叠区间会被合并，写错顺序没关系。
+- `[[0, horizon]]` = 全程开启（最简单合法的解）；`[]` = 全程关闭。
+- **格式错误、区间越界、崩溃、超时 → 该实例得 0 分**（模拟器会验证每个调度）。
+
+## 评测与分数
+
+- 实例池 = 固定公开实例（默认 8 个，跨规模 N=20\~40、K=6\~12）+ **运行时生成实例**（防硬编码）。
+- 分数 = 各实例备电时长的**平均值（分钟）**。
+- 每个实例给固定求解时间预算（默认 **60s**），超时判 0 分。
+- 时间预算可配置：`--time-budget 300`（10min）/ `60`（1min）/ `10`（10s），预算越紧分数通常越低。
+
+本地运行：
+
+```powershell
+python verification/evaluate.py baseline/solver.py            # 固定 8 实例
+python verification/evaluate.py baseline/solver.py --time-budget 10
+python verification/evaluate.py baseline/solver.py --generate-seed 42   # 固定 8 + 运行时生成 8（防硬编码）
+```
+
+框架（unified）评测时，**必须在宿主环境设置** **`TELECOM_EVAL_GENERATE_SEED`** 开启运行时生成
+（否则只有固定公开实例）：
+
+```powershell
+$env:TELECOM_EVAL_GENERATE_SEED = "<SEED>"
+python -m frontier_eval task=unified task.benchmark=PowerSystems/TelecomBackup algorithm.iterations=0 "task.runtime.shell=<path to Git Bash>"
+```
+
+### 完整性 / 防作弊（威胁模型）
+
+- **运行时生成实例**：设了 `TELECOM_EVAL_GENERATE_SEED` 后，评测现场按种子生成新实例
+  （只存在于临时目录，不落仓库/沙箱），候选无法预置针对它们的解。
+- **候选环境剥离**：候选子进程的环境变量剥离 `FRONTIER_*` 与 `TELECOM_EVAL_*`，
+  封死通过宿主环境定位评测基线的侧信道。
+- **静态检查**：评分前检查 EVOLVE-BLOCK 标记与标记外代码（与初始 baseline 逐字节比对）、
+  禁 import 评测/生成模块、禁绝对路径、禁按实例名硬编码；同一实例跑两次输出必须一致
+  （确定性探针）。任何违规判 0 分。
+- **诚实说明**：process 模式下候选进程有宿主文件系统访问权（框架的限制）；本基准
+  依赖"运行时生成 + env 剥离 + 静态检查"多层防御，评测器（`verification/simulator.py`）
+  是有意暴露给候选做内部搜索的白盒。
+
+## 参考分数（实测，固定 8 实例）
+
+| 策略                                | 平均备电时长                      |
+| --------------------------------- | --------------------------- |
+| baseline（朴素全程开启，不调度）           | **176.2** 分钟               |
+| agent（AB-MCTS，15 迭代，最优保存程序实测）   | **266.9** 分钟（+52%）      |
+| 参考启发式（`verification/ref_solver.py`，多路休息轮换） | **271.2** 分钟（+54%） |
+| agent（ShinkaEvolve，15 代，最佳代程序实测）  | **312.5** 分钟（+77%）      |
+| agent（openevolve，25 迭代，最优保存程序实测） | **414.4** 分钟（+135%）    |
+| horizon（覆盖永不跌破 80% 的上限）         | 480 分钟                   |
+
+baseline 只做"全程开启"（覆盖最高但电池并行耗尽）；`ref_solver.py` 的"多路轮流休息"错峰调度
+能显著延长备电时长；三个 agent 框架在足够迭代下都能发现超过全程开启的调度，其中 openevolve
+在 25 迭代下找到接近 horizon 上限的**最小覆盖子集轮换**调度（甚至超过简单参考启发式 53%）。
+实例已校准（静默功耗便宜、工作功耗贵），生成器保证错峰相对全程开启每实例 ≥25% 的可复现提升
+（实测 +28%..+119%，平均 +54%）。
+
+> 注意：agent 分数以**保存程序直跑**为准，且必须**从任务目录内**重新评估——候选求解器按自身
+> 位置解析 `verification/simulator.py`，从任意路径直跑会静默退化为全程开启（分数虚低）。
+> ShinkaEvolve 的"保存 best"会低估其最佳代（框架追踪问题），取其最高分代为准。
+> ShinkaEvolve 25 代时曾产生更高分（~409）但为**时间型模拟退火、非确定性**，不过确定性探针，
+> 故不计（有效最高为 15 代的 312.5）。
+> 5~15 迭代时 agent 大多停在基线附近；迭代越多越能充分搜索——本任务奖励"彻底的模拟器内搜索"。
+
+## 优化方向提示
+
+1. 备电终点几乎总是"某电源耗尽导致覆盖跌破"——策略本质是**让电池错峰放电**。
+2. 覆盖是**冗余**的（80% 约束远小于全开覆盖），可以让部分电源轮换休息（静默省电），
+   只要保证任意时刻覆盖 ≥ 80%。
+3. 关闭电源的负载会迁移到剩余站点，其功耗上升——轮换分组要**地理邻近**（同簇一组），
+   避免单组覆盖不足。
+4. 你可以在求解器里 `import verification/simulator.py`（只读）来评估候选调度的备电时长，
+   做"生成-模拟-改进"的搜索（时间预算内迭代）。
+5. 先保证输出**永远合法**（宁可全开也别格式错），再追求调度质量。
+
diff --git a/benchmarks/PowerSystems/TelecomBackup/baseline/result_log.txt b/benchmarks/PowerSystems/TelecomBackup/baseline/result_log.txt
new file mode 100644
index 00000000..54f74d19
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/baseline/result_log.txt
@@ -0,0 +1,52 @@
+TelecomBackup evaluation log
+=============================
+All scores are mean backup time (minutes) over the fixed 8-instance set,
+verified by directly evaluating the saved programs from the task directory
+(candidates resolve verification/simulator.py relative to their own location,
+so re-evaluation must run from the task root). Only deterministic programs
+(passing the cross-size determinism probe) count.
+
+Date       : 2026-08-17
+Evaluator  : python verification/evaluate.py <solver>  (stdlib only, time budget 60s)
+
+Solvers
+-------
+baseline (always-on, no scheduling)              : 176.25   (valid 1.0)
+reference heuristic (verification/ref_solver.py): 271.2    (multi-rest rotation,
+                                                           direct simulate)
+
+Agents (deepseek-v4-flash, run IDs under runs/unified__TelecomBackup/...)
+-----------------------------------------------------------------------
+openevolve  25 iterations  414.38  run 20260816_130700 (openevolve/best/best_program.py)
+ShinkaEvolve 15 generations 312.50  run 20260816_214014 (shinkaevolve/gen_3/main.py)
+AB-MCTS     15 iterations  266.88  run 20260816_220646 (abmcts/best/solver.py)
+
+Notes
+-----
+- A ShinkaEvolve 25-generation run (20260817_035543) reached ~409 in a single
+  evaluation, but that program is a time-budgeted simulated-annealing solver
+  (non-deterministic) and fails the cross-size determinism probe — excluded.
+- openevolve's 25-iteration schedule (414.38) is near the 480-min horizon.
+- AB-MCTS stays ~250-270 regardless of reasoning mode.
+
+Note: the agent scores above are on the fixed 8 instances only (no
+TELECOM_EVAL_GENERATE_SEED was set for those runs). Set the seed to
+additionally score fresh (anti-hardcoding) instances.
+
+Generalization (fixed 8 + 8 generated, TELECOM_EVAL_GENERATE_SEED=42)
+-------------------------------------------------------------------
+openevolve  best program  410.94  (valid 1.0, 16 instances)
+ShinkaEvolve gen_3         319.38  (valid 1.0, 16 instances)
+AB-MCTS     best solver    280.31  (valid 1.0, 16 instances)
+The saved programs generalize to freshly generated instances (not overfit
+to the fixed 8); generated instances pass the same _stagger_ok acceptance.
+
+Multi-run statistics (3 runs per framework; best valid saved program per run)
+------------------------------------------------------------------------------
+openevolve  25 iters: 414.38 / 298.12 / 357.50 -> mean 356.67, std 47.47
+ShinkaEvolve 15 gens: 312.50 / 357.50 / 325.62 -> mean 331.87, std 18.90
+AB-MCTS    15 iters: 266.88 / 208.75 / 227.50 -> mean 234.38, std 24.22
+Run IDs (runs/unified__PowerSystems__TelecomBackup/...):
+  openevolve: 20260816_130700, 20260819_123913, 20260819_135819
+  ShinkaEvolve: 20260816_214014 (gen 3), 20260819_203149 (gen 2), 20260819_221722 (gen 10)
+  AB-MCTS: 20260816_220646, 20260819_151209, 20260819_171751
diff --git a/benchmarks/PowerSystems/TelecomBackup/baseline/solver.py b/benchmarks/PowerSystems/TelecomBackup/baseline/solver.py
new file mode 100644
index 00000000..6c47cf8d
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/baseline/solver.py
@@ -0,0 +1,46 @@
+#!/usr/bin/env python3
+"""区域备电 baseline 求解器（朴素基线：全程开启，不做时序调度）。
+
+用法：python baseline/solver.py <instance.json>
+输出：stdout 打印 {"on": [[[a,b),...], ...]}，on[k] 为电源 k 的开启时隙区间列表
+（0-index，半开区间；[[0, horizon]] = 全程开启，[] = 全程关闭）。
+
+只允许修改 EVOLVE-BLOCK 区域内的代码；接口契约（main/stdin-json/stdout-json）必须保留。
+
+基线取"全程开启"（朴素、无调度）：覆盖最高但电池并行耗尽。校准后的实例保证
+"错峰调度"（多路轮流休息）能显著超越它（参考启发式 +54%），因此基线是一个
+"正常且可被明显超越"的起点。
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+
+def solve(inst: dict) -> list[list[list[int]]]:
+    """返回 on_intervals 列表（每电源的开启时隙区间）。"""
+    # EVOLVE-BLOCK-START
+    # 朴素策略：所有电源全程开启（不调度）。这是合法的基线解——
+    # 覆盖最高但电池并行耗尽；通过"时序开关调度"（错峰/轮换供电，
+    # 让电池错峰放电、利用覆盖冗余）可以显著延长备电时长。
+    k = len(inst["groups"])
+    horizon = int(inst["horizon"])
+    return [[[0, horizon]] for _ in range(k)]
+    # EVOLVE-BLOCK-END
+
+
+def main() -> int:
+    if len(sys.argv) < 2:
+        print("usage: python solver.py <instance.json>", file=sys.stderr)
+        return 2
+    inst_path = Path(sys.argv[1])
+    inst = json.loads(inst_path.read_text(encoding="utf-8"))
+    on = solve(inst)
+    print(json.dumps({"on": on}))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/agent_files.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/agent_files.txt
new file mode 100644
index 00000000..640607a1
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/agent_files.txt
@@ -0,0 +1,4 @@
+README.md
+Task.md
+baseline/solver.py
+frontier_eval/constraints.txt
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/artifact_files.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/artifact_files.txt
new file mode 100644
index 00000000..a52b8b16
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/artifact_files.txt
@@ -0,0 +1,2 @@
+# No extra artifact files are auto-collected by default for this benchmark.
+# metrics.json and artifacts.json are handled separately by UnifiedTask.
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/constraints.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/constraints.txt
new file mode 100644
index 00000000..dd6a7cc9
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/constraints.txt
@@ -0,0 +1,12 @@
+UnifiedTask constraints:
+1) Only modify `baseline/solver.py`, and only inside the EVOLVE-BLOCK-START / EVOLVE-BLOCK-END region.
+2) Preserve the public contract: run as `python baseline/solver.py <instance.json>`, print a single JSON object
+   `{"on": [[[a,b),...], ...]}` to stdout -- one list of half-open slot intervals [a, b) per power supply,
+   each interval within [0, horizon] and a <= b ([[0, horizon]] means "always on", [] means "never on").
+3) Do not modify benchmark assets, documentation, verification code, instance data, or `frontier_eval/` metadata.
+4) Your output must always be a valid schedule. Malformed output, out-of-range intervals, crashes, or timeouts
+   score 0 for that instance (the simulator validates every schedule).
+5) You may import `verification/simulator.py` (read-only) to evaluate candidate schedules while searching,
+   but the schedule you finally print must come from your own algorithm.
+6) Prioritize validity before optimization. The solver is given a fixed time budget per instance (default 60s);
+   keep it fast enough, or your schedule is discarded.
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/copy_files.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/copy_files.txt
new file mode 100644
index 00000000..79f8b1a5
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/copy_files.txt
@@ -0,0 +1,6 @@
+baseline
+verification/evaluate.py
+verification/simulator.py
+verification/validator.py
+verification/data/instances
+frontier_eval
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/eval_command.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/eval_command.txt
new file mode 100644
index 00000000..3a31d525
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/eval_command.txt
@@ -0,0 +1 @@
+FRONTIER_EVAL_UNIFIED_SOURCE_BENCHMARK_DIR={benchmark_source} {python} frontier_eval/run_eval.py --candidate {candidate} --metrics-out metrics.json --artifacts-out artifacts.json
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/eval_cwd.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/eval_cwd.txt
new file mode 100644
index 00000000..9c558e35
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/eval_cwd.txt
@@ -0,0 +1 @@
+.
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/evaluator.py b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/evaluator.py
new file mode 100644
index 00000000..78274a19
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/evaluator.py
@@ -0,0 +1,36 @@
+"""Unified evaluator entry point for the TelecomBackup benchmark.
+
+This module is loaded by `frontier_eval/run_eval.py` and must expose a
+top-level `evaluate(program_path, **kwargs)` callable. The real
+implementation lives in `verification/evaluate.py`.
+"""
+
+from __future__ import annotations
+
+import importlib.util
+from pathlib import Path
+from typing import Any
+
+TIME_BUDGET_S = 60.0
+
+
+def _load_verification_evaluator() -> Any:
+    evaluator_path = (
+        Path(__file__).resolve().parent.parent / "verification" / "evaluate.py"
+    )
+    spec = importlib.util.spec_from_file_location(
+        "_telecombackup_verification_evaluator", evaluator_path
+    )
+    if spec is None or spec.loader is None:
+        raise RuntimeError(f"Failed to load verification evaluator from {evaluator_path}")
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def evaluate(program_path: str, **kwargs: Any) -> Any:
+    module = _load_verification_evaluator()
+    result = module.evaluate(program_path, time_budget=TIME_BUDGET_S, **kwargs)
+    if isinstance(result, dict) and "metrics" in result:
+        return result
+    return {"metrics": result, "artifacts": {}}
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/initial_program.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/initial_program.txt
new file mode 100644
index 00000000..6645b02f
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/initial_program.txt
@@ -0,0 +1 @@
+baseline/solver.py
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/readonly_files.txt b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/readonly_files.txt
new file mode 100644
index 00000000..1205de75
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/readonly_files.txt
@@ -0,0 +1,4 @@
+README.md
+Task.md
+verification
+frontier_eval
diff --git a/benchmarks/PowerSystems/TelecomBackup/frontier_eval/run_eval.py b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/run_eval.py
new file mode 100644
index 00000000..720d0e09
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/frontier_eval/run_eval.py
@@ -0,0 +1,126 @@
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+import traceback
+from importlib.util import module_from_spec, spec_from_file_location
+from pathlib import Path
+from typing import Any
+
+INVALID_COMBINED_SCORE = -1e18
+
+
+def _write_json(path: Path, obj: Any) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(
+        json.dumps(obj, ensure_ascii=False, indent=2, default=str) + "\n",
+        encoding="utf-8",
+    )
+
+
+def _normalize_result(result: Any) -> tuple[dict[str, Any], dict[str, Any]]:
+    if hasattr(result, "metrics") and hasattr(result, "artifacts"):
+        return dict(getattr(result, "metrics")), dict(getattr(result, "artifacts"))
+
+    if isinstance(result, dict):
+        raw_metrics = result.get("metrics")
+        raw_artifacts = result.get("artifacts")
+        if isinstance(raw_metrics, dict):
+            return dict(raw_metrics), dict(raw_artifacts or {})
+        return dict(result), {}
+
+    raise TypeError(
+        "Evaluator must return an EvaluationResult-like object or a dict of metrics."
+    )
+
+
+def _load_local_evaluator() -> Any:
+    evaluator_path = Path(__file__).with_name("evaluator.py").resolve()
+    spec = spec_from_file_location("_frontier_eval_local_evaluator", evaluator_path)
+    if spec is None or spec.loader is None:
+        raise RuntimeError(f"Failed to load local evaluator from {evaluator_path}")
+    module = module_from_spec(spec)
+    spec.loader.exec_module(module)
+    try:
+        return getattr(module, "evaluate")
+    except AttributeError as exc:
+        raise RuntimeError(
+            f"Local evaluator does not define evaluate(): {evaluator_path}"
+        ) from exc
+
+
+def _find_repo_root() -> Path:
+    import os
+
+    env_root = os.environ.get("FRONTIER_ENGINEERING_ROOT")
+    if env_root:
+        return Path(env_root).expanduser().resolve()
+
+    here = Path(__file__).resolve()
+    for parent in [here.parent, *here.parents]:
+        if (parent / "frontier_eval").is_dir() and (parent / "benchmarks").is_dir():
+            return parent
+    return Path.cwd().resolve()
+
+
+def _build_kwargs(evaluate_fn: Any) -> dict[str, Any]:
+    kwargs: dict[str, Any] = {}
+    try:
+        parameters = inspect_signature(evaluate_fn)
+    except Exception:
+        return kwargs
+
+    if "repo_root" in parameters:
+        kwargs["repo_root"] = _find_repo_root()
+    return kwargs
+
+
+def inspect_signature(fn: Any) -> set[str]:
+    import inspect
+
+    return set(inspect.signature(fn).parameters)
+
+
+def _parse_args(argv: list[str]) -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description="Run a benchmark-local unified evaluator and export metrics/artifacts JSON."
+    )
+    parser.add_argument("--candidate", required=True)
+    parser.add_argument("--metrics-out", default="metrics.json")
+    parser.add_argument("--artifacts-out", default="artifacts.json")
+    return parser.parse_args(argv)
+
+
+def main(argv: list[str]) -> int:
+    args = _parse_args(argv)
+
+    candidate_path = Path(args.candidate).expanduser().resolve()
+    metrics_out = Path(args.metrics_out).expanduser().resolve()
+    artifacts_out = Path(args.artifacts_out).expanduser().resolve()
+
+    metrics: dict[str, Any] = {
+        "combined_score": INVALID_COMBINED_SCORE,
+        "valid": 0.0,
+    }
+    artifacts: dict[str, Any] = {
+        "local_evaluator_path": str(Path(__file__).with_name("evaluator.py").resolve()),
+        "candidate_path": str(candidate_path),
+    }
+
+    try:
+        evaluate_fn = _load_local_evaluator()
+        result = evaluate_fn(str(candidate_path), **_build_kwargs(evaluate_fn))
+        metrics, evaluator_artifacts = _normalize_result(result)
+        artifacts.update(evaluator_artifacts)
+    except Exception as exc:
+        artifacts["error_message"] = str(exc)
+        artifacts["traceback"] = traceback.format_exc()
+
+    _write_json(metrics_out, metrics)
+    _write_json(artifacts_out, artifacts)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main(sys.argv[1:]))
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_1.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_1.json
new file mode 100644
index 00000000..7aa0a680
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_1.json
@@ -0,0 +1,545 @@
+{
+ "seed": 1,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   128.2,
+   50.6
+  ],
+  [
+   152.1,
+   141.4
+  ],
+  [
+   165.6,
+   37.7
+  ],
+  [
+   144.7,
+   35.2
+  ],
+  [
+   42.2,
+   145.0
+  ],
+  [
+   54.5,
+   50.9
+  ],
+  [
+   168.2,
+   42.7
+  ],
+  [
+   81.6,
+   143.8
+  ],
+  [
+   54.1,
+   102.8
+  ],
+  [
+   33.0,
+   174.8
+  ],
+  [
+   193.7,
+   136.5
+  ],
+  [
+   163.3,
+   128.3
+  ],
+  [
+   174.5,
+   107.6
+  ],
+  [
+   63.0,
+   62.4
+  ],
+  [
+   106.1,
+   147.9
+  ],
+  [
+   41.2,
+   175.1
+  ],
+  [
+   163.5,
+   130.5
+  ],
+  [
+   117.3,
+   54.1
+  ],
+  [
+   139.4,
+   156.3
+  ],
+  [
+   75.3,
+   65.3
+  ]
+ ],
+ "groups": [
+  [
+   0,
+   2,
+   3,
+   6,
+   17
+  ],
+  [
+   5,
+   8,
+   13,
+   19
+  ],
+  [
+   1,
+   14,
+   18
+  ],
+  [
+   12
+  ],
+  [
+   10,
+   11,
+   16
+  ],
+  [
+   4,
+   7,
+   9,
+   15
+  ]
+ ],
+ "battery": [
+  26.0,
+  22.0,
+  19.6,
+  27.7,
+  24.2,
+  23.7
+ ],
+ "demand": [
+  0.647,
+  0.169,
+  0.227,
+  0.012,
+  0.2,
+  0.92,
+  0.548,
+  0.404,
+  0.344,
+  0.847,
+  0.353,
+  0.91,
+  0.659,
+  0.609,
+  0.729,
+  0.384,
+  0.857,
+  0.955,
+  0.938,
+  0.512,
+  0.129,
+  0.777,
+  0.205,
+  0.95,
+  0.481,
+  0.365,
+  0.554,
+  0.941,
+  0.413,
+  0.813,
+  0.414,
+  0.002,
+  0.54,
+  0.786,
+  0.331,
+  0.6,
+  0.805,
+  0.635,
+  0.551,
+  0.181,
+  0.092,
+  0.551,
+  0.851,
+  0.931,
+  0.032,
+  0.944,
+  0.07,
+  0.868,
+  0.453,
+  0.754,
+  0.281,
+  0.269,
+  0.797,
+  0.185,
+  0.29,
+  0.167,
+  0.255,
+  0.952,
+  0.657,
+  0.648,
+  0.294,
+  0.703,
+  0.496,
+  0.114,
+  0.312,
+  0.343,
+  0.796,
+  0.258,
+  0.253,
+  0.73,
+  0.977,
+  0.966,
+  0.432,
+  0.976,
+  0.225,
+  0.397,
+  0.035,
+  0.96,
+  0.446,
+  0.506,
+  0.427,
+  0.832,
+  0.977,
+  0.631,
+  0.695,
+  0.451,
+  0.524,
+  0.031,
+  0.675,
+  0.803,
+  0.66,
+  0.426,
+  0.737,
+  0.126,
+  0.212,
+  0.047,
+  0.071,
+  0.076,
+  0.917,
+  0.298,
+  0.158,
+  0.565,
+  0.13,
+  0.561,
+  0.851,
+  0.591,
+  0.218,
+  0.901,
+  0.461,
+  0.828,
+  0.87,
+  0.78,
+  0.623,
+  0.037,
+  0.2,
+  0.099,
+  0.573,
+  0.897,
+  0.591,
+  0.492,
+  0.938,
+  0.39,
+  0.504,
+  0.017,
+  0.612,
+  0.402,
+  0.281,
+  0.157,
+  0.858,
+  0.811,
+  0.563,
+  0.135,
+  0.429,
+  0.267,
+  0.096,
+  0.379,
+  0.548,
+  0.914,
+  0.838,
+  0.534,
+  0.768,
+  0.533,
+  0.065,
+  0.04,
+  0.133,
+  0.167,
+  0.538,
+  0.268,
+  0.332,
+  0.506,
+  0.255,
+  0.339,
+  0.114,
+  0.235,
+  0.944,
+  0.78,
+  0.715,
+  0.489,
+  0.58,
+  0.77,
+  0.321,
+  0.407,
+  0.38,
+  0.991,
+  0.147,
+  0.125,
+  0.115,
+  0.587,
+  0.926,
+  0.077,
+  0.55,
+  0.566,
+  0.952,
+  0.365,
+  0.296,
+  0.534,
+  0.114,
+  0.897,
+  0.108,
+  0.046,
+  0.296,
+  0.614,
+  0.015,
+  0.414,
+  0.826,
+  0.79,
+  0.188,
+  0.786,
+  0.587,
+  0.162,
+  0.451,
+  0.681,
+  0.159,
+  0.845,
+  0.435,
+  0.965,
+  0.806,
+  0.543,
+  0.818,
+  0.55,
+  0.712,
+  0.314,
+  0.208,
+  0.317,
+  0.027,
+  0.787,
+  0.926,
+  0.726,
+  0.32,
+  0.391,
+  0.399,
+  0.064,
+  0.317,
+  0.601,
+  0.456,
+  0.25,
+  0.785,
+  0.778,
+  0.891,
+  0.868,
+  0.469,
+  0.356,
+  0.183,
+  0.208,
+  0.199,
+  0.36,
+  0.82,
+  0.089,
+  0.753,
+  0.09,
+  0.574,
+  0.339,
+  0.227,
+  0.967,
+  0.041,
+  0.187,
+  0.793,
+  0.579,
+  0.921,
+  0.246,
+  0.101,
+  0.611,
+  0.808,
+  0.092,
+  0.22,
+  0.808,
+  0.402,
+  0.268,
+  0.868,
+  0.729,
+  0.022,
+  0.01,
+  0.751,
+  0.359,
+  0.469,
+  0.859,
+  0.101,
+  0.778,
+  0.328,
+  0.509,
+  0.665,
+  0.18,
+  0.15,
+  0.142,
+  0.866,
+  0.306,
+  0.709,
+  0.835,
+  0.602,
+  0.126,
+  0.207,
+  0.545,
+  0.723,
+  0.78,
+  0.821,
+  0.624,
+  0.672,
+  0.553,
+  0.943,
+  0.987,
+  0.205,
+  0.299,
+  0.537,
+  0.049,
+  0.862,
+  0.247,
+  0.778,
+  0.682,
+  0.447,
+  0.43,
+  0.25,
+  0.439,
+  0.538,
+  0.011,
+  0.836,
+  0.172,
+  0.486,
+  0.793,
+  0.933,
+  0.976,
+  0.019,
+  0.692,
+  0.58,
+  0.594,
+  0.138,
+  0.983,
+  0.277,
+  0.564,
+  0.172,
+  0.089,
+  0.486,
+  0.178,
+  0.317,
+  0.893,
+  0.92,
+  0.93,
+  0.639,
+  0.226,
+  0.313,
+  0.687,
+  0.957,
+  0.713,
+  0.337,
+  0.611,
+  0.728,
+  0.653,
+  0.972,
+  0.219,
+  0.922,
+  0.763,
+  0.645,
+  0.369,
+  0.512,
+  0.792,
+  0.204,
+  0.299,
+  0.3,
+  0.552,
+  0.165,
+  0.701,
+  0.465,
+  0.085,
+  0.123,
+  0.606,
+  0.514,
+  0.377,
+  0.156,
+  0.427,
+  0.942,
+  0.72,
+  0.782,
+  0.495,
+  0.394,
+  0.637,
+  0.384,
+  0.845,
+  0.544,
+  0.994,
+  0.524,
+  0.09,
+  0.255,
+  0.101,
+  0.737,
+  0.084,
+  0.975,
+  0.969,
+  0.617,
+  0.967,
+  0.687,
+  0.082,
+  0.851,
+  0.241,
+  0.851,
+  0.94,
+  0.903,
+  0.397,
+  0.91,
+  0.438,
+  0.622,
+  0.488,
+  0.212,
+  0.431,
+  0.534,
+  0.909,
+  0.661,
+  0.278,
+  0.379,
+  0.559,
+  0.96,
+  0.528,
+  0.579,
+  0.031,
+  0.973,
+  0.242,
+  0.26,
+  0.173,
+  0.148,
+  0.2,
+  0.311,
+  0.757
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_2.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_2.json
new file mode 100644
index 00000000..9723b2dc
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_2.json
@@ -0,0 +1,568 @@
+{
+ "seed": 2,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   147.4,
+   120.3
+  ],
+  [
+   49.9,
+   163.8
+  ],
+  [
+   39.8,
+   44.3
+  ],
+  [
+   126.7,
+   22.4
+  ],
+  [
+   29.1,
+   144.1
+  ],
+  [
+   82.9,
+   126.5
+  ],
+  [
+   32.9,
+   153.8
+  ],
+  [
+   113.2,
+   69.2
+  ],
+  [
+   136.4,
+   144.6
+  ],
+  [
+   55.5,
+   61.1
+  ],
+  [
+   80.0,
+   149.8
+  ],
+  [
+   171.7,
+   149.3
+  ],
+  [
+   145.7,
+   6.7
+  ],
+  [
+   54.6,
+   131.9
+  ],
+  [
+   167.7,
+   78.4
+  ],
+  [
+   141.5,
+   37.5
+  ],
+  [
+   71.0,
+   139.3
+  ],
+  [
+   38.4,
+   131.3
+  ],
+  [
+   131.0,
+   75.9
+  ],
+  [
+   135.9,
+   51.2
+  ],
+  [
+   65.0,
+   115.1
+  ],
+  [
+   53.3,
+   127.4
+  ],
+  [
+   80.2,
+   141.7
+  ],
+  [
+   144.8,
+   62.6
+  ]
+ ],
+ "groups": [
+  [
+   1,
+   4,
+   6
+  ],
+  [
+   11
+  ],
+  [
+   13,
+   17,
+   20,
+   21
+  ],
+  [
+   3,
+   7,
+   12,
+   14,
+   15,
+   18,
+   19,
+   23
+  ],
+  [
+   2,
+   9
+  ],
+  [
+   5,
+   10,
+   16,
+   22
+  ],
+  [
+   0,
+   8
+  ]
+ ],
+ "battery": [
+  28.6,
+  17.9,
+  26.2,
+  15.9,
+  24.8,
+  19.1,
+  18.4
+ ],
+ "demand": [
+  0.875,
+  0.106,
+  0.522,
+  0.854,
+  0.245,
+  0.21,
+  0.881,
+  0.423,
+  0.717,
+  0.032,
+  0.362,
+  0.172,
+  0.673,
+  0.083,
+  0.955,
+  0.025,
+  0.729,
+  0.021,
+  0.256,
+  0.813,
+  0.157,
+  0.184,
+  0.691,
+  0.386,
+  0.043,
+  0.99,
+  0.151,
+  0.036,
+  0.344,
+  0.615,
+  0.742,
+  0.113,
+  0.337,
+  0.031,
+  0.449,
+  0.766,
+  0.74,
+  0.902,
+  0.756,
+  0.862,
+  0.705,
+  0.473,
+  0.226,
+  0.661,
+  0.316,
+  0.102,
+  0.448,
+  0.875,
+  0.128,
+  0.585,
+  0.393,
+  0.515,
+  0.144,
+  0.96,
+  0.259,
+  0.606,
+  0.42,
+  0.018,
+  0.558,
+  0.141,
+  0.057,
+  0.034,
+  0.161,
+  0.096,
+  0.635,
+  0.508,
+  0.983,
+  0.934,
+  0.995,
+  0.232,
+  0.445,
+  0.251,
+  0.591,
+  0.624,
+  0.8,
+  0.709,
+  0.257,
+  0.423,
+  0.526,
+  0.005,
+  0.035,
+  0.409,
+  0.111,
+  0.724,
+  0.241,
+  0.1,
+  0.182,
+  0.232,
+  0.217,
+  0.521,
+  0.464,
+  0.31,
+  0.642,
+  0.212,
+  0.907,
+  0.963,
+  0.729,
+  0.434,
+  0.512,
+  0.581,
+  0.051,
+  0.418,
+  0.525,
+  0.181,
+  0.094,
+  0.803,
+  0.366,
+  0.519,
+  0.921,
+  0.611,
+  0.29,
+  0.984,
+  0.372,
+  0.019,
+  0.685,
+  0.101,
+  0.306,
+  0.841,
+  0.673,
+  0.016,
+  0.451,
+  0.411,
+  0.486,
+  0.208,
+  0.589,
+  0.074,
+  0.284,
+  0.373,
+  0.935,
+  0.077,
+  0.755,
+  0.192,
+  0.572,
+  0.392,
+  0.463,
+  0.754,
+  0.395,
+  0.122,
+  0.122,
+  0.081,
+  0.85,
+  0.641,
+  0.96,
+  0.693,
+  0.025,
+  0.659,
+  0.777,
+  0.724,
+  0.498,
+  0.358,
+  0.457,
+  0.799,
+  0.269,
+  0.526,
+  0.478,
+  0.955,
+  0.804,
+  0.932,
+  0.836,
+  0.297,
+  0.232,
+  0.489,
+  0.259,
+  0.428,
+  0.679,
+  0.919,
+  0.586,
+  0.818,
+  0.096,
+  0.356,
+  0.998,
+  0.147,
+  0.417,
+  0.067,
+  0.086,
+  0.896,
+  0.989,
+  0.648,
+  0.129,
+  0.296,
+  0.232,
+  0.671,
+  0.681,
+  0.439,
+  0.524,
+  0.112,
+  0.541,
+  0.95,
+  0.756,
+  0.096,
+  0.517,
+  0.715,
+  0.257,
+  0.895,
+  0.461,
+  0.703,
+  0.404,
+  0.995,
+  0.783,
+  0.573,
+  0.145,
+  0.441,
+  0.029,
+  0.595,
+  0.882,
+  0.18,
+  0.51,
+  0.482,
+  0.405,
+  0.71,
+  0.937,
+  0.705,
+  0.472,
+  0.962,
+  0.331,
+  0.746,
+  0.658,
+  0.762,
+  0.852,
+  0.225,
+  0.621,
+  0.403,
+  0.667,
+  0.977,
+  0.635,
+  0.012,
+  0.465,
+  0.712,
+  0.883,
+  0.65,
+  0.816,
+  0.017,
+  0.943,
+  0.729,
+  0.606,
+  0.905,
+  0.885,
+  0.1,
+  0.816,
+  0.767,
+  0.2,
+  0.744,
+  0.586,
+  0.191,
+  0.804,
+  0.138,
+  0.612,
+  0.434,
+  0.254,
+  0.566,
+  0.467,
+  0.205,
+  0.967,
+  0.073,
+  0.003,
+  0.485,
+  0.837,
+  0.658,
+  0.755,
+  0.485,
+  0.675,
+  0.335,
+  0.267,
+  0.503,
+  0.028,
+  0.08,
+  0.754,
+  0.174,
+  0.75,
+  0.784,
+  0.404,
+  0.675,
+  0.787,
+  0.864,
+  0.135,
+  0.163,
+  0.382,
+  0.465,
+  0.295,
+  0.01,
+  0.557,
+  0.967,
+  0.366,
+  0.538,
+  0.382,
+  0.443,
+  0.87,
+  0.308,
+  0.649,
+  0.484,
+  0.539,
+  0.915,
+  0.077,
+  0.824,
+  0.304,
+  0.646,
+  0.796,
+  0.653,
+  0.393,
+  0.841,
+  0.093,
+  0.633,
+  0.391,
+  0.53,
+  0.851,
+  0.798,
+  0.629,
+  0.308,
+  0.233,
+  0.458,
+  0.232,
+  0.277,
+  0.958,
+  0.112,
+  0.819,
+  0.379,
+  0.365,
+  0.318,
+  0.077,
+  0.457,
+  0.166,
+  0.442,
+  0.292,
+  0.895,
+  0.922,
+  0.442,
+  0.64,
+  0.93,
+  0.326,
+  0.1,
+  0.238,
+  0.19,
+  0.678,
+  0.374,
+  0.356,
+  0.795,
+  0.233,
+  0.809,
+  0.633,
+  0.4,
+  0.824,
+  0.342,
+  0.879,
+  0.926,
+  0.503,
+  0.69,
+  0.949,
+  0.743,
+  0.751,
+  0.869,
+  0.936,
+  0.754,
+  0.979,
+  0.292,
+  0.622,
+  0.671,
+  0.367,
+  0.395,
+  0.175,
+  0.958,
+  0.354,
+  0.477,
+  0.894,
+  0.186,
+  0.961,
+  0.127,
+  0.028,
+  0.351,
+  0.359,
+  0.918,
+  0.883,
+  0.762,
+  0.436,
+  0.543,
+  0.237,
+  0.834,
+  0.39,
+  0.285,
+  0.638,
+  0.151,
+  0.316,
+  0.926,
+  0.095,
+  0.142,
+  0.204,
+  0.251,
+  0.42,
+  0.25,
+  0.343,
+  0.246,
+  0.24,
+  0.611,
+  0.336,
+  0.373,
+  0.768,
+  0.062,
+  0.144,
+  0.851,
+  0.43,
+  0.779
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_3.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_3.json
new file mode 100644
index 00000000..58b056bc
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_3.json
@@ -0,0 +1,591 @@
+{
+ "seed": 104,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   60.3,
+   150.4
+  ],
+  [
+   58.0,
+   45.1
+  ],
+  [
+   171.5,
+   169.7
+  ],
+  [
+   45.2,
+   155.2
+  ],
+  [
+   132.6,
+   140.2
+  ],
+  [
+   61.6,
+   146.0
+  ],
+  [
+   31.1,
+   162.1
+  ],
+  [
+   149.3,
+   162.6
+  ],
+  [
+   71.8,
+   65.9
+  ],
+  [
+   57.2,
+   36.5
+  ],
+  [
+   145.8,
+   56.5
+  ],
+  [
+   153.5,
+   141.8
+  ],
+  [
+   158.6,
+   26.3
+  ],
+  [
+   195.0,
+   115.5
+  ],
+  [
+   158.1,
+   140.2
+  ],
+  [
+   171.8,
+   54.4
+  ],
+  [
+   144.1,
+   144.1
+  ],
+  [
+   159.5,
+   154.6
+  ],
+  [
+   166.1,
+   165.8
+  ],
+  [
+   132.8,
+   31.6
+  ],
+  [
+   146.7,
+   133.1
+  ],
+  [
+   76.4,
+   17.3
+  ],
+  [
+   125.9,
+   47.1
+  ],
+  [
+   45.8,
+   47.7
+  ],
+  [
+   57.2,
+   71.9
+  ],
+  [
+   159.3,
+   59.2
+  ],
+  [
+   119.1,
+   130.3
+  ],
+  [
+   78.7,
+   171.5
+  ]
+ ],
+ "groups": [
+  [
+   1,
+   8,
+   9,
+   23,
+   24
+  ],
+  [
+   4,
+   11,
+   13,
+   14,
+   16,
+   20,
+   26
+  ],
+  [
+   2,
+   7,
+   17,
+   18
+  ],
+  [
+   19,
+   21,
+   22
+  ],
+  [
+   0,
+   5
+  ],
+  [
+   10,
+   12,
+   15,
+   25
+  ],
+  [
+   3,
+   6
+  ],
+  [
+   27
+  ]
+ ],
+ "battery": [
+  20.8,
+  23.1,
+  18.6,
+  17.8,
+  27.8,
+  27.9,
+  23.9,
+  25.3
+ ],
+ "demand": [
+  0.734,
+  0.625,
+  0.386,
+  0.706,
+  0.509,
+  0.914,
+  0.48,
+  0.709,
+  0.008,
+  0.867,
+  0.29,
+  0.087,
+  0.226,
+  0.071,
+  0.032,
+  0.056,
+  0.78,
+  0.927,
+  0.613,
+  0.915,
+  0.442,
+  0.072,
+  0.513,
+  0.504,
+  0.04,
+  0.578,
+  0.157,
+  0.981,
+  0.175,
+  0.27,
+  0.886,
+  0.262,
+  0.528,
+  0.454,
+  0.695,
+  0.164,
+  0.487,
+  0.124,
+  0.729,
+  0.061,
+  0.309,
+  0.064,
+  0.334,
+  0.244,
+  0.542,
+  0.108,
+  0.241,
+  0.694,
+  0.86,
+  0.273,
+  0.755,
+  0.804,
+  0.472,
+  0.88,
+  0.715,
+  0.5,
+  0.821,
+  0.207,
+  0.733,
+  0.151,
+  0.372,
+  0.106,
+  0.322,
+  0.66,
+  0.422,
+  0.583,
+  0.891,
+  0.813,
+  0.954,
+  0.87,
+  0.1,
+  0.522,
+  0.362,
+  0.822,
+  0.143,
+  0.735,
+  0.806,
+  0.331,
+  0.216,
+  0.719,
+  0.866,
+  0.678,
+  0.463,
+  0.21,
+  0.555,
+  0.906,
+  0.172,
+  0.93,
+  0.295,
+  0.764,
+  0.616,
+  0.146,
+  0.124,
+  0.543,
+  0.09,
+  0.371,
+  0.245,
+  0.65,
+  0.82,
+  0.162,
+  0.201,
+  0.811,
+  0.971,
+  0.907,
+  0.323,
+  0.852,
+  0.262,
+  0.685,
+  0.322,
+  0.702,
+  0.639,
+  0.771,
+  0.618,
+  0.592,
+  0.777,
+  0.398,
+  0.672,
+  0.503,
+  0.674,
+  0.165,
+  0.106,
+  0.33,
+  0.258,
+  0.682,
+  0.995,
+  0.713,
+  0.621,
+  0.305,
+  0.49,
+  0.554,
+  0.194,
+  0.275,
+  0.118,
+  0.875,
+  0.879,
+  0.037,
+  0.115,
+  0.081,
+  0.078,
+  0.682,
+  0.599,
+  0.586,
+  0.767,
+  0.327,
+  0.574,
+  0.878,
+  0.36,
+  0.121,
+  0.259,
+  0.809,
+  0.084,
+  0.121,
+  0.23,
+  0.752,
+  0.506,
+  0.529,
+  0.098,
+  0.943,
+  0.624,
+  0.076,
+  0.501,
+  0.825,
+  0.697,
+  0.904,
+  0.218,
+  0.607,
+  0.216,
+  0.977,
+  0.395,
+  0.652,
+  0.314,
+  0.652,
+  0.795,
+  0.689,
+  0.118,
+  0.959,
+  0.105,
+  0.965,
+  0.588,
+  0.179,
+  0.879,
+  0.978,
+  0.041,
+  0.91,
+  0.547,
+  0.969,
+  0.784,
+  0.565,
+  0.823,
+  0.553,
+  0.136,
+  0.845,
+  0.688,
+  0.041,
+  0.4,
+  0.722,
+  0.152,
+  0.008,
+  0.873,
+  0.346,
+  0.145,
+  0.726,
+  0.578,
+  0.849,
+  0.235,
+  0.214,
+  0.536,
+  0.497,
+  0.6,
+  0.854,
+  0.655,
+  0.52,
+  0.197,
+  0.572,
+  0.445,
+  0.679,
+  0.085,
+  0.483,
+  0.041,
+  0.402,
+  0.666,
+  0.821,
+  0.575,
+  0.283,
+  0.824,
+  0.359,
+  0.021,
+  0.502,
+  0.128,
+  0.335,
+  0.433,
+  0.861,
+  0.335,
+  0.074,
+  0.688,
+  0.36,
+  0.575,
+  0.606,
+  0.546,
+  0.472,
+  0.487,
+  0.796,
+  0.352,
+  0.462,
+  0.04,
+  0.789,
+  0.138,
+  0.777,
+  0.327,
+  0.205,
+  0.814,
+  0.607,
+  0.387,
+  0.424,
+  0.015,
+  0.244,
+  0.936,
+  0.763,
+  0.626,
+  0.922,
+  0.485,
+  0.937,
+  0.481,
+  0.56,
+  0.036,
+  0.773,
+  0.773,
+  0.303,
+  0.724,
+  0.621,
+  0.257,
+  0.452,
+  0.869,
+  0.923,
+  0.063,
+  0.223,
+  0.395,
+  0.779,
+  0.068,
+  0.502,
+  0.204,
+  0.68,
+  0.635,
+  0.069,
+  0.899,
+  0.694,
+  0.906,
+  0.479,
+  0.459,
+  0.815,
+  0.845,
+  0.037,
+  0.664,
+  0.805,
+  0.168,
+  0.75,
+  0.618,
+  0.265,
+  0.911,
+  0.983,
+  0.562,
+  0.67,
+  0.075,
+  0.565,
+  0.245,
+  0.27,
+  0.83,
+  0.488,
+  0.011,
+  0.472,
+  0.117,
+  0.155,
+  0.778,
+  0.178,
+  0.644,
+  0.5,
+  0.59,
+  0.486,
+  0.669,
+  0.122,
+  0.476,
+  0.109,
+  0.717,
+  0.203,
+  0.679,
+  0.482,
+  0.241,
+  0.368,
+  0.101,
+  0.384,
+  0.337,
+  0.265,
+  0.981,
+  0.813,
+  0.356,
+  0.076,
+  0.177,
+  0.795,
+  0.967,
+  0.136,
+  0.147,
+  0.809,
+  0.558,
+  0.477,
+  0.345,
+  0.989,
+  0.902,
+  0.676,
+  0.51,
+  0.52,
+  0.157,
+  0.923,
+  0.963,
+  0.482,
+  0.277,
+  0.274,
+  0.873,
+  0.645,
+  0.353,
+  0.011,
+  0.719,
+  0.481,
+  0.525,
+  0.943,
+  0.558,
+  0.059,
+  0.151,
+  0.149,
+  0.531,
+  0.265,
+  0.449,
+  0.017,
+  0.96,
+  0.914,
+  0.103,
+  0.057,
+  0.505,
+  0.894,
+  0.605,
+  0.501,
+  0.169,
+  0.211,
+  0.611,
+  0.9,
+  0.815,
+  0.767,
+  0.957,
+  0.455,
+  0.888,
+  0.731,
+  0.382,
+  0.553,
+  0.878,
+  0.131,
+  0.789,
+  0.43,
+  0.477,
+  0.718,
+  0.815,
+  0.571
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_4.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_4.json
new file mode 100644
index 00000000..ed831ec5
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_4.json
@@ -0,0 +1,617 @@
+{
+ "seed": 4,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   54.2,
+   156.1
+  ],
+  [
+   34.3,
+   145.0
+  ],
+  [
+   154.4,
+   74.4
+  ],
+  [
+   57.0,
+   13.2
+  ],
+  [
+   60.2,
+   94.7
+  ],
+  [
+   54.1,
+   37.5
+  ],
+  [
+   40.1,
+   89.0
+  ],
+  [
+   73.3,
+   64.1
+  ],
+  [
+   37.5,
+   149.5
+  ],
+  [
+   41.0,
+   195.0
+  ],
+  [
+   66.6,
+   50.8
+  ],
+  [
+   63.8,
+   30.5
+  ],
+  [
+   35.4,
+   157.6
+  ],
+  [
+   59.0,
+   164.5
+  ],
+  [
+   169.5,
+   60.9
+  ],
+  [
+   21.3,
+   46.6
+  ],
+  [
+   68.3,
+   73.7
+  ],
+  [
+   138.5,
+   57.0
+  ],
+  [
+   46.9,
+   168.5
+  ],
+  [
+   41.2,
+   40.0
+  ],
+  [
+   55.7,
+   54.2
+  ],
+  [
+   145.5,
+   176.0
+  ],
+  [
+   183.6,
+   182.4
+  ],
+  [
+   44.3,
+   51.5
+  ],
+  [
+   189.5,
+   19.7
+  ],
+  [
+   66.4,
+   51.9
+  ],
+  [
+   153.2,
+   129.7
+  ],
+  [
+   146.5,
+   62.6
+  ],
+  [
+   46.0,
+   164.6
+  ],
+  [
+   148.7,
+   152.4
+  ],
+  [
+   131.7,
+   55.3
+  ],
+  [
+   76.0,
+   49.5
+  ]
+ ],
+ "groups": [
+  [
+   9
+  ],
+  [
+   4,
+   6
+  ],
+  [
+   0,
+   13,
+   18,
+   28
+  ],
+  [
+   2,
+   14,
+   17,
+   24,
+   27,
+   30
+  ],
+  [
+   3
+  ],
+  [
+   1,
+   8,
+   12
+  ],
+  [
+   21,
+   22,
+   26,
+   29
+  ],
+  [
+   7,
+   10,
+   16,
+   20,
+   25,
+   31
+  ],
+  [
+   15,
+   19,
+   23
+  ],
+  [
+   5,
+   11
+  ]
+ ],
+ "battery": [
+  23.9,
+  29.6,
+  24.1,
+  23.2,
+  23.7,
+  19.8,
+  15.5,
+  27.5,
+  19.4,
+  30.0
+ ],
+ "demand": [
+  0.652,
+  0.2,
+  0.202,
+  0.439,
+  0.492,
+  0.219,
+  0.444,
+  0.647,
+  0.428,
+  0.216,
+  0.188,
+  0.037,
+  0.253,
+  0.526,
+  0.772,
+  0.417,
+  0.262,
+  0.325,
+  0.891,
+  0.315,
+  0.117,
+  0.57,
+  0.963,
+  0.653,
+  0.871,
+  0.717,
+  0.041,
+  0.387,
+  0.43,
+  0.93,
+  0.572,
+  0.944,
+  0.165,
+  0.296,
+  0.471,
+  0.901,
+  0.315,
+  0.42,
+  0.215,
+  0.801,
+  0.801,
+  0.339,
+  0.392,
+  0.496,
+  0.859,
+  0.28,
+  0.628,
+  0.191,
+  0.395,
+  0.62,
+  0.767,
+  0.269,
+  0.856,
+  0.87,
+  0.688,
+  0.465,
+  0.472,
+  0.404,
+  0.39,
+  0.798,
+  0.211,
+  0.157,
+  0.61,
+  0.257,
+  0.396,
+  0.77,
+  0.381,
+  0.222,
+  0.053,
+  0.951,
+  0.162,
+  0.608,
+  0.818,
+  0.562,
+  0.784,
+  0.915,
+  0.44,
+  0.079,
+  0.695,
+  0.113,
+  0.561,
+  0.257,
+  0.774,
+  0.041,
+  0.08,
+  0.879,
+  0.905,
+  0.297,
+  0.347,
+  0.075,
+  0.96,
+  0.454,
+  0.205,
+  0.977,
+  0.389,
+  0.759,
+  0.855,
+  0.4,
+  0.077,
+  0.916,
+  0.798,
+  0.366,
+  1.0,
+  0.416,
+  0.711,
+  0.444,
+  0.629,
+  0.941,
+  0.637,
+  0.93,
+  0.479,
+  0.118,
+  0.558,
+  0.372,
+  0.163,
+  0.707,
+  0.327,
+  0.86,
+  0.259,
+  0.974,
+  0.005,
+  0.169,
+  0.648,
+  0.121,
+  0.111,
+  0.786,
+  0.601,
+  0.526,
+  0.521,
+  0.411,
+  0.293,
+  0.229,
+  0.18,
+  0.864,
+  0.002,
+  0.053,
+  0.908,
+  0.544,
+  0.909,
+  0.782,
+  0.568,
+  0.937,
+  0.507,
+  0.441,
+  0.617,
+  0.392,
+  0.252,
+  0.597,
+  0.986,
+  0.34,
+  0.433,
+  0.606,
+  0.675,
+  0.622,
+  0.285,
+  0.94,
+  0.197,
+  0.794,
+  0.675,
+  0.883,
+  0.077,
+  0.646,
+  0.904,
+  0.33,
+  0.372,
+  0.175,
+  0.883,
+  0.604,
+  0.524,
+  0.755,
+  0.089,
+  0.359,
+  0.813,
+  0.098,
+  0.182,
+  0.987,
+  0.661,
+  0.993,
+  0.077,
+  0.926,
+  0.986,
+  0.172,
+  0.479,
+  0.886,
+  0.222,
+  0.787,
+  0.66,
+  0.853,
+  0.688,
+  0.881,
+  0.237,
+  0.49,
+  0.221,
+  0.368,
+  0.746,
+  0.537,
+  0.53,
+  0.452,
+  0.803,
+  0.506,
+  0.97,
+  0.313,
+  0.438,
+  0.588,
+  0.251,
+  0.897,
+  0.459,
+  0.561,
+  0.826,
+  0.404,
+  0.389,
+  0.028,
+  0.511,
+  0.458,
+  0.965,
+  0.631,
+  0.347,
+  0.796,
+  0.91,
+  0.756,
+  0.483,
+  0.64,
+  0.477,
+  0.604,
+  0.066,
+  0.924,
+  0.069,
+  0.133,
+  0.047,
+  0.398,
+  0.582,
+  0.621,
+  0.713,
+  0.538,
+  0.019,
+  0.245,
+  0.751,
+  0.205,
+  0.949,
+  0.082,
+  0.271,
+  0.879,
+  0.643,
+  0.61,
+  0.632,
+  0.172,
+  0.149,
+  0.415,
+  0.13,
+  0.709,
+  0.081,
+  0.364,
+  0.003,
+  0.881,
+  0.156,
+  0.412,
+  0.215,
+  0.298,
+  0.508,
+  0.376,
+  0.163,
+  0.89,
+  0.515,
+  0.537,
+  0.303,
+  0.401,
+  0.17,
+  0.062,
+  0.98,
+  0.279,
+  0.676,
+  0.304,
+  0.789,
+  0.086,
+  0.116,
+  0.609,
+  0.015,
+  0.231,
+  0.009,
+  0.911,
+  0.532,
+  0.885,
+  0.452,
+  0.732,
+  0.87,
+  0.684,
+  0.697,
+  0.336,
+  0.173,
+  0.53,
+  0.188,
+  0.475,
+  0.903,
+  0.357,
+  0.93,
+  0.832,
+  0.413,
+  0.738,
+  0.821,
+  0.212,
+  0.579,
+  0.802,
+  0.591,
+  0.784,
+  0.477,
+  0.834,
+  0.807,
+  0.083,
+  0.236,
+  0.316,
+  0.104,
+  0.976,
+  0.038,
+  0.322,
+  0.476,
+  0.955,
+  0.928,
+  0.177,
+  0.043,
+  0.519,
+  0.827,
+  0.711,
+  0.039,
+  0.676,
+  0.336,
+  0.602,
+  0.138,
+  0.309,
+  0.508,
+  0.889,
+  0.825,
+  0.839,
+  0.071,
+  0.226,
+  0.274,
+  0.954,
+  0.891,
+  0.973,
+  0.011,
+  0.698,
+  0.213,
+  0.249,
+  0.738,
+  0.345,
+  0.983,
+  0.858,
+  0.632,
+  0.403,
+  0.173,
+  0.494,
+  0.064,
+  0.511,
+  0.509,
+  0.56,
+  0.283,
+  0.208,
+  0.703,
+  0.322,
+  0.215,
+  0.479,
+  0.437,
+  0.972,
+  0.331,
+  0.909,
+  0.95,
+  0.928,
+  0.41,
+  0.75,
+  0.21,
+  0.28,
+  0.087,
+  0.403,
+  0.609,
+  0.676,
+  0.642,
+  0.613,
+  0.313,
+  0.935,
+  0.782,
+  0.62,
+  0.701,
+  0.934,
+  0.254,
+  0.216,
+  0.642,
+  0.764,
+  0.645,
+  0.008,
+  0.57,
+  0.204,
+  0.094,
+  0.519,
+  0.095,
+  0.798,
+  0.316,
+  0.869,
+  0.582,
+  0.151,
+  0.746,
+  0.662
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_5.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_5.json
new file mode 100644
index 00000000..1504665c
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_5.json
@@ -0,0 +1,640 @@
+{
+ "seed": 5,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   52.5,
+   35.1
+  ],
+  [
+   58.0,
+   144.3
+  ],
+  [
+   63.3,
+   35.6
+  ],
+  [
+   47.1,
+   148.5
+  ],
+  [
+   139.4,
+   156.9
+  ],
+  [
+   69.2,
+   80.1
+  ],
+  [
+   134.0,
+   135.4
+  ],
+  [
+   190.5,
+   150.5
+  ],
+  [
+   157.5,
+   99.8
+  ],
+  [
+   161.5,
+   138.1
+  ],
+  [
+   120.8,
+   142.7
+  ],
+  [
+   171.9,
+   195.0
+  ],
+  [
+   191.4,
+   141.2
+  ],
+  [
+   170.5,
+   52.8
+  ],
+  [
+   148.4,
+   167.9
+  ],
+  [
+   24.2,
+   132.2
+  ],
+  [
+   36.4,
+   184.4
+  ],
+  [
+   45.8,
+   37.9
+  ],
+  [
+   48.1,
+   43.4
+  ],
+  [
+   67.9,
+   144.1
+  ],
+  [
+   85.0,
+   54.0
+  ],
+  [
+   122.6,
+   76.9
+  ],
+  [
+   62.1,
+   53.7
+  ],
+  [
+   161.0,
+   160.2
+  ],
+  [
+   39.5,
+   65.5
+  ],
+  [
+   151.5,
+   40.6
+  ],
+  [
+   153.5,
+   45.8
+  ],
+  [
+   56.2,
+   173.2
+  ],
+  [
+   165.9,
+   41.0
+  ],
+  [
+   43.3,
+   166.1
+  ],
+  [
+   158.5,
+   123.2
+  ],
+  [
+   161.9,
+   49.8
+  ],
+  [
+   73.7,
+   163.2
+  ],
+  [
+   82.2,
+   56.5
+  ],
+  [
+   160.2,
+   30.4
+  ],
+  [
+   83.9,
+   94.1
+  ]
+ ],
+ "groups": [
+  [
+   0,
+   2,
+   17,
+   18
+  ],
+  [
+   28,
+   31
+  ],
+  [
+   21
+  ],
+  [
+   13
+  ],
+  [
+   8,
+   30
+  ],
+  [
+   7,
+   9,
+   11,
+   12,
+   23
+  ],
+  [
+   22,
+   24
+  ],
+  [
+   4,
+   6,
+   10,
+   14
+  ],
+  [
+   5,
+   20,
+   33,
+   35
+  ],
+  [
+   25,
+   26,
+   34
+  ],
+  [
+   1,
+   3,
+   15,
+   16,
+   19,
+   27,
+   29,
+   32
+  ]
+ ],
+ "battery": [
+  16.6,
+  21.5,
+  17.2,
+  27.7,
+  19.4,
+  21.8,
+  30.0,
+  27.8,
+  29.6,
+  21.8,
+  22.3
+ ],
+ "demand": [
+  0.73,
+  0.479,
+  0.291,
+  0.404,
+  0.147,
+  0.377,
+  0.988,
+  0.96,
+  0.627,
+  0.499,
+  0.338,
+  0.089,
+  0.272,
+  0.782,
+  0.867,
+  0.361,
+  0.786,
+  0.775,
+  0.695,
+  0.664,
+  0.76,
+  0.363,
+  0.704,
+  0.281,
+  0.486,
+  0.77,
+  0.691,
+  0.294,
+  0.946,
+  0.65,
+  0.581,
+  0.012,
+  0.547,
+  0.251,
+  0.672,
+  0.463,
+  0.817,
+  0.647,
+  0.798,
+  0.348,
+  0.644,
+  0.738,
+  0.828,
+  0.35,
+  0.843,
+  0.87,
+  0.688,
+  0.976,
+  0.957,
+  0.518,
+  0.529,
+  0.166,
+  0.837,
+  0.937,
+  0.477,
+  0.691,
+  0.72,
+  0.73,
+  0.172,
+  0.78,
+  0.581,
+  0.666,
+  0.421,
+  0.624,
+  0.775,
+  0.637,
+  0.72,
+  0.028,
+  0.16,
+  0.441,
+  0.65,
+  0.219,
+  0.686,
+  0.631,
+  0.042,
+  0.472,
+  0.226,
+  0.054,
+  0.134,
+  0.317,
+  0.182,
+  0.193,
+  0.036,
+  0.465,
+  0.38,
+  0.612,
+  0.59,
+  0.238,
+  0.903,
+  0.001,
+  0.405,
+  0.279,
+  0.41,
+  0.115,
+  0.831,
+  0.374,
+  0.036,
+  0.614,
+  0.095,
+  0.545,
+  0.339,
+  0.581,
+  0.958,
+  0.819,
+  0.419,
+  0.813,
+  0.642,
+  0.369,
+  0.142,
+  0.596,
+  0.564,
+  0.957,
+  0.968,
+  0.609,
+  0.351,
+  0.893,
+  0.001,
+  0.108,
+  0.566,
+  0.615,
+  0.141,
+  0.629,
+  0.891,
+  0.376,
+  0.432,
+  0.226,
+  0.291,
+  0.972,
+  0.38,
+  0.961,
+  0.914,
+  0.596,
+  0.26,
+  0.981,
+  0.496,
+  0.415,
+  0.319,
+  0.984,
+  0.492,
+  0.286,
+  0.477,
+  0.122,
+  0.622,
+  0.443,
+  0.293,
+  0.782,
+  0.827,
+  0.013,
+  0.533,
+  0.274,
+  0.935,
+  0.782,
+  0.246,
+  0.268,
+  0.155,
+  0.989,
+  0.293,
+  0.608,
+  0.475,
+  0.645,
+  0.604,
+  0.743,
+  0.118,
+  0.76,
+  0.301,
+  0.533,
+  0.336,
+  0.297,
+  0.53,
+  0.464,
+  0.361,
+  0.745,
+  0.591,
+  0.036,
+  0.252,
+  0.456,
+  0.917,
+  0.888,
+  0.546,
+  0.015,
+  0.778,
+  0.428,
+  0.576,
+  0.708,
+  0.632,
+  0.482,
+  0.912,
+  0.385,
+  0.392,
+  0.852,
+  0.196,
+  0.296,
+  0.83,
+  0.066,
+  0.836,
+  0.695,
+  0.433,
+  0.286,
+  0.781,
+  0.911,
+  0.143,
+  0.478,
+  0.549,
+  0.498,
+  0.331,
+  0.154,
+  0.586,
+  0.812,
+  0.068,
+  0.23,
+  0.82,
+  0.792,
+  0.664,
+  0.026,
+  0.723,
+  0.979,
+  0.998,
+  0.701,
+  0.049,
+  0.842,
+  0.219,
+  0.646,
+  0.952,
+  0.712,
+  0.135,
+  0.292,
+  0.918,
+  0.15,
+  0.611,
+  0.414,
+  0.161,
+  0.622,
+  0.044,
+  0.108,
+  0.379,
+  0.072,
+  0.058,
+  0.575,
+  0.742,
+  0.878,
+  0.134,
+  0.432,
+  0.315,
+  0.6,
+  0.49,
+  0.939,
+  0.374,
+  0.056,
+  0.697,
+  0.151,
+  0.631,
+  0.506,
+  0.91,
+  0.555,
+  0.621,
+  0.263,
+  0.552,
+  0.254,
+  0.751,
+  0.517,
+  0.134,
+  0.234,
+  0.371,
+  0.737,
+  0.179,
+  0.713,
+  0.655,
+  0.085,
+  0.668,
+  0.091,
+  0.125,
+  0.594,
+  0.239,
+  0.877,
+  0.48,
+  0.323,
+  0.796,
+  0.029,
+  0.725,
+  0.054,
+  0.151,
+  0.952,
+  0.681,
+  0.223,
+  0.116,
+  0.973,
+  0.665,
+  0.821,
+  0.14,
+  0.625,
+  0.354,
+  0.235,
+  0.333,
+  0.614,
+  0.349,
+  0.386,
+  0.136,
+  0.831,
+  0.648,
+  0.805,
+  0.433,
+  0.852,
+  0.517,
+  0.593,
+  0.573,
+  0.74,
+  0.396,
+  0.097,
+  0.033,
+  0.202,
+  0.039,
+  0.889,
+  0.481,
+  0.76,
+  0.0,
+  0.47,
+  0.89,
+  0.619,
+  0.429,
+  0.466,
+  0.1,
+  0.155,
+  0.159,
+  0.375,
+  0.386,
+  0.88,
+  0.152,
+  0.254,
+  0.277,
+  0.162,
+  0.287,
+  0.235,
+  0.482,
+  0.032,
+  0.925,
+  0.369,
+  0.938,
+  0.688,
+  0.674,
+  0.472,
+  0.945,
+  0.118,
+  0.669,
+  0.291,
+  0.674,
+  0.729,
+  0.163,
+  0.201,
+  0.025,
+  0.23,
+  0.078,
+  0.401,
+  0.974,
+  0.364,
+  0.312,
+  0.468,
+  0.283,
+  0.732,
+  0.718,
+  0.163,
+  0.241,
+  0.672,
+  0.94,
+  0.646,
+  0.431,
+  0.975,
+  0.006,
+  0.061,
+  0.779,
+  0.41,
+  0.045,
+  0.548,
+  0.99,
+  0.519,
+  0.35,
+  0.094,
+  0.071,
+  0.899,
+  0.491,
+  0.936,
+  0.054,
+  0.243,
+  0.05,
+  0.397,
+  0.06,
+  0.255,
+  0.407,
+  0.306,
+  0.051,
+  0.038,
+  0.972,
+  0.179,
+  0.509,
+  0.402,
+  0.531,
+  0.084,
+  0.314,
+  0.108,
+  0.542,
+  0.921
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_6.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_6.json
new file mode 100644
index 00000000..ba9d253d
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_6.json
@@ -0,0 +1,663 @@
+{
+ "seed": 6,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   66.2,
+   71.1
+  ],
+  [
+   48.2,
+   153.8
+  ],
+  [
+   160.4,
+   119.3
+  ],
+  [
+   171.1,
+   136.7
+  ],
+  [
+   58.9,
+   73.8
+  ],
+  [
+   35.7,
+   126.6
+  ],
+  [
+   60.1,
+   35.0
+  ],
+  [
+   36.7,
+   65.7
+  ],
+  [
+   41.6,
+   153.0
+  ],
+  [
+   139.1,
+   16.0
+  ],
+  [
+   75.4,
+   42.0
+  ],
+  [
+   170.8,
+   11.0
+  ],
+  [
+   190.0,
+   132.0
+  ],
+  [
+   40.0,
+   38.9
+  ],
+  [
+   141.1,
+   175.1
+  ],
+  [
+   19.9,
+   50.0
+  ],
+  [
+   171.9,
+   69.8
+  ],
+  [
+   46.2,
+   130.8
+  ],
+  [
+   158.9,
+   60.5
+  ],
+  [
+   54.6,
+   61.3
+  ],
+  [
+   22.1,
+   139.4
+  ],
+  [
+   32.7,
+   167.4
+  ],
+  [
+   109.4,
+   195.0
+  ],
+  [
+   60.6,
+   74.7
+  ],
+  [
+   167.8,
+   162.2
+  ],
+  [
+   73.2,
+   31.3
+  ],
+  [
+   114.3,
+   162.3
+  ],
+  [
+   27.6,
+   47.6
+  ],
+  [
+   162.2,
+   145.2
+  ],
+  [
+   19.5,
+   159.6
+  ],
+  [
+   174.7,
+   52.9
+  ],
+  [
+   34.0,
+   139.9
+  ],
+  [
+   32.1,
+   149.7
+  ],
+  [
+   95.4,
+   144.7
+  ],
+  [
+   138.0,
+   139.8
+  ],
+  [
+   137.1,
+   130.6
+  ],
+  [
+   121.6,
+   150.6
+  ],
+  [
+   136.9,
+   176.3
+  ],
+  [
+   144.6,
+   135.2
+  ],
+  [
+   181.9,
+   144.9
+  ]
+ ],
+ "groups": [
+  [
+   6,
+   10,
+   25
+  ],
+  [
+   16,
+   18,
+   30
+  ],
+  [
+   34,
+   35,
+   38
+  ],
+  [
+   0,
+   4,
+   19,
+   23
+  ],
+  [
+   24,
+   28
+  ],
+  [
+   2,
+   3,
+   12,
+   39
+  ],
+  [
+   9,
+   11
+  ],
+  [
+   1,
+   5,
+   8,
+   17,
+   20,
+   31,
+   32
+  ],
+  [
+   7,
+   13,
+   15,
+   27
+  ],
+  [
+   21,
+   29
+  ],
+  [
+   14,
+   22,
+   37
+  ],
+  [
+   26,
+   33,
+   36
+  ]
+ ],
+ "battery": [
+  20.4,
+  29.1,
+  27.8,
+  15.9,
+  17.9,
+  29.2,
+  24.2,
+  21.3,
+  26.7,
+  27.2,
+  22.4,
+  22.0
+ ],
+ "demand": [
+  0.1,
+  0.362,
+  0.145,
+  0.626,
+  0.607,
+  0.144,
+  0.628,
+  0.097,
+  0.105,
+  0.646,
+  0.463,
+  0.9,
+  0.629,
+  0.701,
+  0.127,
+  0.833,
+  0.084,
+  0.618,
+  0.911,
+  0.908,
+  0.893,
+  0.452,
+  0.694,
+  0.791,
+  0.137,
+  0.943,
+  0.909,
+  0.441,
+  0.778,
+  0.447,
+  0.276,
+  0.712,
+  0.064,
+  0.8,
+  0.978,
+  0.286,
+  0.94,
+  0.084,
+  0.858,
+  0.28,
+  0.966,
+  0.431,
+  0.922,
+  0.032,
+  0.086,
+  0.816,
+  0.639,
+  0.536,
+  0.033,
+  0.173,
+  0.37,
+  0.052,
+  0.993,
+  0.765,
+  0.95,
+  0.701,
+  0.343,
+  0.078,
+  0.732,
+  0.114,
+  0.547,
+  0.464,
+  0.495,
+  0.075,
+  0.074,
+  0.52,
+  0.03,
+  0.525,
+  0.15,
+  0.282,
+  0.151,
+  0.568,
+  0.388,
+  0.223,
+  0.222,
+  0.564,
+  0.911,
+  0.419,
+  0.163,
+  0.316,
+  0.24,
+  0.572,
+  0.786,
+  0.569,
+  0.188,
+  0.325,
+  0.381,
+  0.441,
+  0.398,
+  0.998,
+  0.551,
+  0.662,
+  0.716,
+  0.645,
+  0.805,
+  0.372,
+  0.667,
+  0.973,
+  0.164,
+  0.625,
+  0.385,
+  0.061,
+  0.04,
+  0.759,
+  0.115,
+  0.711,
+  0.1,
+  0.368,
+  0.032,
+  0.543,
+  0.786,
+  0.226,
+  0.565,
+  0.922,
+  0.293,
+  0.471,
+  0.724,
+  0.864,
+  0.169,
+  0.661,
+  0.12,
+  0.112,
+  0.983,
+  0.05,
+  0.92,
+  0.75,
+  0.676,
+  0.729,
+  0.651,
+  0.287,
+  0.443,
+  0.202,
+  0.859,
+  0.779,
+  0.788,
+  0.214,
+  0.592,
+  0.255,
+  0.791,
+  0.745,
+  0.995,
+  0.708,
+  0.684,
+  0.777,
+  0.094,
+  0.595,
+  0.033,
+  0.367,
+  0.117,
+  0.948,
+  0.028,
+  0.921,
+  0.948,
+  0.436,
+  0.832,
+  0.052,
+  0.139,
+  0.01,
+  0.068,
+  0.003,
+  0.828,
+  0.088,
+  0.591,
+  0.273,
+  0.132,
+  0.419,
+  0.182,
+  0.977,
+  0.384,
+  0.594,
+  0.564,
+  0.627,
+  0.48,
+  0.27,
+  0.137,
+  0.888,
+  0.193,
+  0.448,
+  0.808,
+  0.576,
+  0.948,
+  0.81,
+  0.514,
+  0.12,
+  0.593,
+  0.474,
+  0.862,
+  0.531,
+  0.694,
+  0.812,
+  0.47,
+  0.842,
+  0.899,
+  0.154,
+  0.919,
+  0.87,
+  0.855,
+  0.338,
+  0.252,
+  0.133,
+  0.371,
+  0.855,
+  0.363,
+  0.005,
+  0.305,
+  0.083,
+  0.623,
+  0.768,
+  0.927,
+  0.089,
+  0.38,
+  0.126,
+  0.91,
+  0.837,
+  0.52,
+  0.954,
+  0.52,
+  0.776,
+  0.488,
+  0.364,
+  0.994,
+  0.842,
+  0.592,
+  0.867,
+  0.649,
+  0.798,
+  0.067,
+  0.132,
+  0.109,
+  0.603,
+  0.72,
+  0.155,
+  0.505,
+  0.114,
+  0.688,
+  0.779,
+  0.087,
+  0.98,
+  0.665,
+  0.125,
+  0.351,
+  0.158,
+  0.728,
+  0.917,
+  0.175,
+  0.184,
+  0.629,
+  0.86,
+  0.739,
+  0.458,
+  0.111,
+  0.614,
+  0.456,
+  0.804,
+  0.572,
+  0.734,
+  0.971,
+  0.626,
+  0.912,
+  0.274,
+  0.002,
+  0.882,
+  0.03,
+  0.994,
+  0.456,
+  0.061,
+  0.275,
+  0.721,
+  0.384,
+  0.624,
+  0.102,
+  0.836,
+  0.373,
+  0.833,
+  0.147,
+  0.667,
+  0.469,
+  0.404,
+  0.019,
+  0.532,
+  0.838,
+  0.104,
+  0.446,
+  0.012,
+  0.994,
+  0.97,
+  0.678,
+  0.48,
+  0.07,
+  0.121,
+  0.705,
+  0.795,
+  0.989,
+  0.862,
+  0.438,
+  0.255,
+  0.542,
+  0.346,
+  0.545,
+  0.884,
+  0.453,
+  0.981,
+  0.718,
+  0.046,
+  0.747,
+  0.662,
+  0.622,
+  0.003,
+  0.879,
+  0.495,
+  0.537,
+  0.17,
+  0.035,
+  0.014,
+  0.21,
+  0.304,
+  0.154,
+  0.058,
+  0.727,
+  0.702,
+  0.882,
+  0.218,
+  0.728,
+  0.594,
+  0.09,
+  0.875,
+  0.357,
+  0.185,
+  0.183,
+  0.924,
+  0.221,
+  0.384,
+  0.547,
+  0.359,
+  0.656,
+  0.088,
+  0.43,
+  0.585,
+  0.619,
+  0.609,
+  0.269,
+  0.638,
+  0.972,
+  0.482,
+  0.644,
+  0.837,
+  0.252,
+  0.266,
+  0.97,
+  0.301,
+  0.426,
+  0.371,
+  0.993,
+  0.545,
+  0.896,
+  0.162,
+  0.278,
+  0.65,
+  0.318,
+  0.663,
+  0.508,
+  0.078,
+  0.112,
+  0.439,
+  0.534,
+  0.177,
+  0.263,
+  0.001,
+  0.266,
+  0.309,
+  0.388,
+  0.593,
+  0.853,
+  0.717,
+  0.741,
+  0.283,
+  0.496,
+  0.727,
+  0.894,
+  0.24,
+  0.015,
+  0.051,
+  0.515,
+  0.872,
+  0.153,
+  0.784,
+  0.641,
+  0.864,
+  0.901,
+  0.107,
+  0.794,
+  0.135,
+  0.119,
+  0.399,
+  0.489,
+  0.07,
+  0.276,
+  0.175,
+  0.135,
+  0.099
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_7.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_7.json
new file mode 100644
index 00000000..80637450
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_7.json
@@ -0,0 +1,568 @@
+{
+ "seed": 7,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   53.9,
+   142.0
+  ],
+  [
+   61.7,
+   142.5
+  ],
+  [
+   43.0,
+   153.1
+  ],
+  [
+   43.4,
+   147.8
+  ],
+  [
+   76.7,
+   140.8
+  ],
+  [
+   32.2,
+   140.6
+  ],
+  [
+   87.9,
+   61.4
+  ],
+  [
+   128.2,
+   62.1
+  ],
+  [
+   121.9,
+   38.8
+  ],
+  [
+   25.1,
+   136.0
+  ],
+  [
+   175.8,
+   68.1
+  ],
+  [
+   32.8,
+   134.1
+  ],
+  [
+   153.8,
+   127.9
+  ],
+  [
+   140.2,
+   161.6
+  ],
+  [
+   45.3,
+   35.8
+  ],
+  [
+   109.7,
+   43.6
+  ],
+  [
+   136.8,
+   195.0
+  ],
+  [
+   38.0,
+   149.1
+  ],
+  [
+   163.4,
+   68.9
+  ],
+  [
+   57.8,
+   148.1
+  ],
+  [
+   140.0,
+   65.7
+  ],
+  [
+   130.6,
+   139.4
+  ],
+  [
+   71.3,
+   142.3
+  ],
+  [
+   78.9,
+   161.6
+  ]
+ ],
+ "groups": [
+  [
+   7,
+   8,
+   15,
+   20
+  ],
+  [
+   10,
+   18
+  ],
+  [
+   13,
+   16
+  ],
+  [
+   6,
+   14
+  ],
+  [
+   2,
+   3,
+   5,
+   9,
+   11,
+   17
+  ],
+  [
+   12,
+   21
+  ],
+  [
+   0,
+   1,
+   4,
+   19,
+   22,
+   23
+  ]
+ ],
+ "battery": [
+  15.3,
+  21.9,
+  17.5,
+  16.8,
+  15.9,
+  26.5,
+  16.9
+ ],
+ "demand": [
+  0.248,
+  0.391,
+  0.871,
+  0.081,
+  0.449,
+  0.549,
+  0.883,
+  0.819,
+  0.864,
+  0.278,
+  0.415,
+  0.359,
+  0.884,
+  0.958,
+  0.151,
+  0.176,
+  0.232,
+  0.233,
+  0.485,
+  0.589,
+  0.263,
+  0.004,
+  0.419,
+  0.369,
+  0.566,
+  0.953,
+  0.69,
+  0.515,
+  0.618,
+  0.676,
+  0.054,
+  0.9,
+  0.78,
+  0.875,
+  0.798,
+  0.392,
+  0.399,
+  0.104,
+  0.634,
+  0.062,
+  0.067,
+  0.209,
+  0.162,
+  0.34,
+  0.053,
+  0.0,
+  0.151,
+  0.101,
+  0.364,
+  0.026,
+  0.874,
+  0.614,
+  0.149,
+  0.252,
+  0.347,
+  0.364,
+  0.123,
+  0.849,
+  0.993,
+  0.466,
+  0.484,
+  0.086,
+  0.102,
+  0.343,
+  0.265,
+  0.829,
+  0.161,
+  0.023,
+  0.951,
+  0.528,
+  0.147,
+  0.543,
+  0.027,
+  0.528,
+  0.979,
+  0.863,
+  0.696,
+  0.261,
+  0.367,
+  0.167,
+  0.772,
+  0.533,
+  0.779,
+  0.33,
+  0.223,
+  0.812,
+  0.985,
+  0.853,
+  0.806,
+  0.818,
+  0.74,
+  0.227,
+  0.518,
+  0.356,
+  0.029,
+  0.028,
+  0.279,
+  0.259,
+  0.693,
+  0.957,
+  0.447,
+  0.937,
+  0.988,
+  0.955,
+  0.365,
+  0.22,
+  0.227,
+  0.197,
+  0.204,
+  0.624,
+  0.9,
+  0.84,
+  0.479,
+  0.653,
+  0.8,
+  0.085,
+  0.661,
+  0.91,
+  0.782,
+  0.75,
+  0.478,
+  0.179,
+  0.789,
+  0.333,
+  0.801,
+  0.972,
+  0.396,
+  0.401,
+  0.947,
+  0.725,
+  0.17,
+  0.127,
+  0.151,
+  0.905,
+  0.807,
+  0.146,
+  0.827,
+  0.98,
+  0.657,
+  0.35,
+  0.549,
+  0.131,
+  0.014,
+  0.971,
+  0.65,
+  0.527,
+  0.934,
+  0.434,
+  0.872,
+  0.826,
+  0.211,
+  0.252,
+  0.293,
+  0.241,
+  0.586,
+  0.259,
+  0.419,
+  0.131,
+  0.91,
+  0.354,
+  0.458,
+  0.583,
+  0.904,
+  0.421,
+  0.918,
+  0.502,
+  0.532,
+  0.524,
+  0.019,
+  0.44,
+  0.183,
+  0.004,
+  0.799,
+  0.172,
+  0.473,
+  0.725,
+  0.556,
+  0.326,
+  0.518,
+  0.555,
+  0.784,
+  0.106,
+  0.56,
+  0.248,
+  0.277,
+  0.772,
+  0.508,
+  0.562,
+  0.76,
+  0.912,
+  0.443,
+  0.613,
+  0.506,
+  0.512,
+  0.693,
+  0.452,
+  0.533,
+  0.478,
+  0.942,
+  0.699,
+  0.877,
+  0.942,
+  0.26,
+  0.56,
+  0.943,
+  0.84,
+  0.137,
+  0.122,
+  0.442,
+  0.073,
+  0.241,
+  0.073,
+  0.669,
+  0.784,
+  0.897,
+  0.154,
+  0.716,
+  0.66,
+  0.143,
+  0.883,
+  0.968,
+  0.22,
+  0.953,
+  0.398,
+  0.487,
+  0.99,
+  0.832,
+  0.161,
+  0.432,
+  0.516,
+  0.339,
+  0.196,
+  0.319,
+  0.722,
+  0.019,
+  0.554,
+  0.44,
+  0.018,
+  0.331,
+  0.624,
+  0.512,
+  0.064,
+  0.985,
+  0.788,
+  0.972,
+  0.105,
+  0.266,
+  0.04,
+  0.779,
+  0.27,
+  0.13,
+  0.422,
+  0.911,
+  0.819,
+  0.259,
+  0.149,
+  0.919,
+  0.571,
+  0.7,
+  0.089,
+  0.058,
+  0.688,
+  0.425,
+  0.072,
+  0.938,
+  0.634,
+  0.802,
+  0.084,
+  0.856,
+  0.067,
+  0.863,
+  0.454,
+  0.339,
+  0.553,
+  0.927,
+  0.268,
+  0.129,
+  0.527,
+  0.238,
+  0.109,
+  0.161,
+  0.05,
+  0.202,
+  0.312,
+  0.305,
+  0.759,
+  0.29,
+  0.5,
+  0.178,
+  0.347,
+  0.018,
+  0.25,
+  0.015,
+  0.733,
+  0.551,
+  0.189,
+  0.475,
+  0.935,
+  0.106,
+  0.819,
+  0.432,
+  0.495,
+  0.835,
+  0.393,
+  0.507,
+  0.688,
+  0.982,
+  0.343,
+  0.832,
+  0.707,
+  0.636,
+  0.405,
+  0.348,
+  0.054,
+  0.13,
+  0.071,
+  0.741,
+  0.256,
+  0.163,
+  0.084,
+  0.841,
+  0.871,
+  0.671,
+  0.282,
+  0.242,
+  0.293,
+  0.459,
+  0.158,
+  0.446,
+  0.263,
+  0.962,
+  0.973,
+  0.547,
+  0.244,
+  0.966,
+  0.31,
+  0.357,
+  0.001,
+  0.382,
+  0.475,
+  0.503,
+  0.201,
+  0.505,
+  0.005,
+  0.264,
+  0.09,
+  0.4,
+  0.042,
+  0.022,
+  0.304,
+  0.233,
+  0.586,
+  0.529,
+  0.751,
+  0.658,
+  0.716,
+  0.879,
+  0.39,
+  0.326,
+  0.985,
+  0.149,
+  0.724,
+  0.643,
+  0.044,
+  0.835,
+  0.892,
+  0.627,
+  0.734,
+  0.812,
+  0.139,
+  0.524,
+  0.504,
+  0.835,
+  0.805,
+  0.826,
+  0.584,
+  0.893,
+  0.683,
+  0.693,
+  0.23,
+  0.031,
+  0.133,
+  0.361,
+  0.105,
+  0.836,
+  0.559,
+  0.628,
+  0.626,
+  0.681,
+  0.489,
+  0.003,
+  0.798,
+  0.748,
+  0.503,
+  0.535,
+  0.659,
+  0.066,
+  0.737,
+  0.252,
+  0.074
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_8.json b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_8.json
new file mode 100644
index 00000000..550d197e
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/data/instances/instance_8.json
@@ -0,0 +1,617 @@
+{
+ "seed": 8,
+ "grid": {
+  "width": 200.0,
+  "height": 200.0,
+  "nx": 20,
+  "ny": 20
+ },
+ "sites": [
+  [
+   147.6,
+   141.9
+  ],
+  [
+   158.9,
+   128.0
+  ],
+  [
+   36.1,
+   32.8
+  ],
+  [
+   29.8,
+   66.6
+  ],
+  [
+   62.4,
+   57.8
+  ],
+  [
+   142.7,
+   26.9
+  ],
+  [
+   65.9,
+   38.7
+  ],
+  [
+   145.3,
+   42.3
+  ],
+  [
+   121.6,
+   74.6
+  ],
+  [
+   140.3,
+   34.6
+  ],
+  [
+   121.0,
+   50.4
+  ],
+  [
+   140.3,
+   144.6
+  ],
+  [
+   144.6,
+   48.6
+  ],
+  [
+   173.7,
+   181.8
+  ],
+  [
+   44.6,
+   9.1
+  ],
+  [
+   13.2,
+   136.1
+  ],
+  [
+   72.5,
+   75.5
+  ],
+  [
+   133.0,
+   46.6
+  ],
+  [
+   86.6,
+   138.1
+  ],
+  [
+   143.2,
+   167.5
+  ],
+  [
+   42.3,
+   149.5
+  ],
+  [
+   125.6,
+   186.2
+  ],
+  [
+   41.3,
+   163.1
+  ],
+  [
+   49.5,
+   66.8
+  ],
+  [
+   59.1,
+   67.9
+  ],
+  [
+   185.6,
+   133.9
+  ],
+  [
+   160.2,
+   82.6
+  ],
+  [
+   76.5,
+   173.8
+  ],
+  [
+   43.3,
+   60.2
+  ],
+  [
+   59.1,
+   43.9
+  ],
+  [
+   43.2,
+   54.5
+  ],
+  [
+   147.8,
+   58.7
+  ]
+ ],
+ "groups": [
+  [
+   5,
+   7,
+   9,
+   10,
+   12,
+   17
+  ],
+  [
+   13,
+   19,
+   21
+  ],
+  [
+   3,
+   28,
+   30
+  ],
+  [
+   4,
+   23,
+   24
+  ],
+  [
+   16
+  ],
+  [
+   18,
+   27
+  ],
+  [
+   0,
+   1,
+   11,
+   25
+  ],
+  [
+   2,
+   6,
+   14,
+   29
+  ],
+  [
+   8,
+   26,
+   31
+  ],
+  [
+   15,
+   20,
+   22
+  ]
+ ],
+ "battery": [
+  25.6,
+  29.7,
+  15.9,
+  15.3,
+  17.1,
+  20.8,
+  23.6,
+  26.2,
+  20.3,
+  29.5
+ ],
+ "demand": [
+  0.003,
+  0.836,
+  0.21,
+  0.875,
+  0.033,
+  0.965,
+  0.759,
+  0.99,
+  0.58,
+  0.178,
+  0.97,
+  0.442,
+  0.083,
+  0.605,
+  0.497,
+  0.473,
+  0.155,
+  0.122,
+  0.581,
+  0.453,
+  0.126,
+  0.439,
+  0.726,
+  0.715,
+  0.471,
+  0.394,
+  0.361,
+  0.226,
+  0.198,
+  0.446,
+  0.108,
+  0.725,
+  0.708,
+  0.612,
+  0.023,
+  0.364,
+  0.432,
+  0.093,
+  0.968,
+  0.756,
+  0.503,
+  0.705,
+  0.201,
+  0.229,
+  0.526,
+  0.744,
+  0.647,
+  0.007,
+  0.613,
+  0.909,
+  0.743,
+  0.928,
+  0.982,
+  0.224,
+  0.296,
+  0.547,
+  0.977,
+  0.369,
+  0.941,
+  0.85,
+  0.516,
+  0.625,
+  0.957,
+  0.783,
+  0.296,
+  0.846,
+  0.196,
+  0.37,
+  0.959,
+  0.847,
+  0.312,
+  0.29,
+  0.613,
+  0.843,
+  0.684,
+  0.936,
+  0.728,
+  0.821,
+  0.605,
+  0.362,
+  0.428,
+  0.49,
+  0.529,
+  0.838,
+  0.269,
+  0.069,
+  0.733,
+  0.271,
+  0.42,
+  0.117,
+  0.829,
+  0.274,
+  0.292,
+  0.804,
+  0.504,
+  0.277,
+  0.888,
+  0.412,
+  0.361,
+  0.965,
+  0.79,
+  0.747,
+  0.25,
+  0.813,
+  0.597,
+  0.83,
+  0.368,
+  0.781,
+  0.793,
+  0.388,
+  0.655,
+  0.78,
+  0.993,
+  0.92,
+  0.786,
+  0.137,
+  0.61,
+  0.685,
+  0.087,
+  0.463,
+  0.869,
+  0.905,
+  0.925,
+  0.093,
+  0.607,
+  0.865,
+  0.653,
+  0.32,
+  0.039,
+  0.92,
+  0.859,
+  0.654,
+  0.366,
+  0.828,
+  0.815,
+  0.725,
+  0.6,
+  0.319,
+  0.297,
+  0.357,
+  0.327,
+  0.52,
+  0.114,
+  0.721,
+  0.348,
+  0.16,
+  0.255,
+  0.389,
+  0.431,
+  0.782,
+  0.1,
+  0.223,
+  0.293,
+  0.177,
+  0.749,
+  0.109,
+  0.051,
+  0.556,
+  0.736,
+  0.825,
+  0.559,
+  0.327,
+  0.882,
+  0.736,
+  0.479,
+  0.563,
+  0.936,
+  0.997,
+  0.931,
+  0.133,
+  0.327,
+  0.408,
+  0.169,
+  0.762,
+  0.061,
+  0.887,
+  0.586,
+  0.614,
+  0.415,
+  0.937,
+  0.578,
+  0.897,
+  0.923,
+  0.692,
+  0.983,
+  0.49,
+  0.367,
+  0.844,
+  0.528,
+  0.309,
+  0.68,
+  0.476,
+  0.128,
+  0.838,
+  0.5,
+  0.972,
+  0.534,
+  0.937,
+  0.289,
+  0.217,
+  0.527,
+  0.767,
+  0.825,
+  0.647,
+  0.124,
+  0.394,
+  0.035,
+  0.371,
+  0.945,
+  0.347,
+  0.699,
+  0.621,
+  0.143,
+  0.688,
+  0.08,
+  0.185,
+  0.58,
+  0.127,
+  0.964,
+  0.426,
+  0.63,
+  0.965,
+  0.108,
+  0.999,
+  0.04,
+  0.601,
+  0.656,
+  0.986,
+  0.358,
+  0.947,
+  0.65,
+  0.477,
+  0.17,
+  0.236,
+  0.365,
+  0.346,
+  0.372,
+  0.808,
+  0.375,
+  0.629,
+  0.039,
+  0.141,
+  0.71,
+  0.891,
+  0.313,
+  0.032,
+  0.233,
+  0.36,
+  0.366,
+  0.561,
+  0.302,
+  0.647,
+  0.786,
+  0.783,
+  0.741,
+  0.584,
+  0.261,
+  0.841,
+  0.29,
+  0.86,
+  0.447,
+  0.629,
+  0.499,
+  0.654,
+  0.341,
+  0.578,
+  0.737,
+  0.361,
+  0.6,
+  0.456,
+  0.47,
+  0.384,
+  0.547,
+  0.709,
+  0.174,
+  0.146,
+  0.823,
+  0.088,
+  0.793,
+  0.759,
+  0.218,
+  0.636,
+  0.253,
+  0.882,
+  0.197,
+  0.116,
+  0.037,
+  0.719,
+  0.212,
+  0.314,
+  0.977,
+  0.736,
+  0.738,
+  0.576,
+  0.641,
+  0.011,
+  0.813,
+  0.329,
+  0.855,
+  0.415,
+  0.911,
+  0.0,
+  0.98,
+  0.096,
+  0.511,
+  0.482,
+  0.986,
+  0.717,
+  0.982,
+  0.906,
+  0.425,
+  0.148,
+  0.146,
+  0.709,
+  0.951,
+  0.665,
+  0.346,
+  0.597,
+  0.657,
+  0.253,
+  0.694,
+  0.199,
+  0.028,
+  0.064,
+  0.488,
+  0.696,
+  0.226,
+  0.397,
+  0.693,
+  0.263,
+  0.522,
+  0.603,
+  0.406,
+  0.208,
+  0.693,
+  0.701,
+  0.189,
+  0.807,
+  0.655,
+  0.866,
+  0.78,
+  0.494,
+  0.716,
+  0.314,
+  0.654,
+  0.082,
+  0.346,
+  0.506,
+  0.886,
+  0.684,
+  0.909,
+  0.869,
+  0.971,
+  0.618,
+  0.617,
+  0.469,
+  0.221,
+  0.797,
+  0.61,
+  0.162,
+  0.24,
+  0.221,
+  0.723,
+  0.597,
+  0.201,
+  0.688,
+  0.563,
+  0.267,
+  0.445,
+  0.993,
+  0.141,
+  0.237,
+  0.594,
+  0.84,
+  0.922,
+  0.399,
+  0.769,
+  0.119,
+  0.386,
+  0.857,
+  0.665,
+  0.447,
+  0.599,
+  0.487,
+  0.548,
+  0.884,
+  0.976,
+  0.2,
+  0.755,
+  0.945,
+  0.607,
+  0.5,
+  0.373,
+  0.484,
+  0.581,
+  0.611,
+  0.114,
+  0.999,
+  0.276,
+  0.548
+ ],
+ "pt_dbm": 20.0,
+ "n_exp": 6.0,
+ "threshold": -105.0,
+ "coverage_ratio": 0.8,
+ "delta_min": 5.0,
+ "horizon": 96,
+ "site_cap": 60.0,
+ "p_silent": 0.05,
+ "p_work_base": 3.0,
+ "p_work_coef": 3.0
+}
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/docker/Dockerfile b/benchmarks/PowerSystems/TelecomBackup/verification/docker/Dockerfile
new file mode 100644
index 00000000..7c94d870
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/docker/Dockerfile
@@ -0,0 +1,13 @@
+FROM python:3.11-slim
+
+WORKDIR /workspace
+
+# The evaluator, simulator, reference solver and baseline use only the Python
+# standard library. The unified runtime mounts the benchmark sandbox into the
+# container, so no benchmark files are baked into the image. The candidate may
+# read `verification/simulator.py` (it is part of the scoring objective and is
+# allowed by the constraints), but no instance data is baked in.
+
+ENV PYTHONUNBUFFERED=1
+
+CMD ["python"]
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/evaluate.py b/benchmarks/PowerSystems/TelecomBackup/verification/evaluate.py
new file mode 100644
index 00000000..b65c84d9
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/evaluate.py
@@ -0,0 +1,283 @@
+"""区域备电评测入口。
+
+CLI：python verification/evaluate.py <solver.py> [--time-budget 60] [--data-dir ...]
+对每个实例：subprocess 运行候选求解器（超时 = 时间预算），解析输出 on_intervals，
+调 simulator 模拟打分；超时/格式错/越界 -> 该实例 0 分。
+分数 = 各实例备电时长的平均值（分钟）。
+
+完整性 / 防作弊（对齐 CVRP 基准的经验）：
+  * 评分前静态检查候选（EVOLVE-BLOCK 标记与标记外代码、禁引用评测/生成模块、
+    禁绝对路径、禁按实例名硬编码）——见 verification/validator.py；
+  * 候选子进程环境剥离 FRONTIER_*/TELECOM_EVAL_*（candidate_env），封侧信道；
+  * 运行时生成实例（TELECOM_EVAL_GENERATE_SEED 设置时，评测现场按种子生成新实例，
+    候选无法预先记忆；生成的实例只存在于临时目录，不落仓库/沙箱）。
+  * 确定性探针：跨规模选若干实例（小/中/大 + 一个生成实例）各跑两次，输出必须一致。
+
+环境变量：
+    TELECOM_EVAL_GENERATE_SEED  设置后开启运行时生成（防硬编码）
+    TELECOM_EVAL_GENERATE_COUNT 生成实例数（默认 8）
+
+对外接口（供 frontier_eval/evaluator.py 包装）：
+    evaluate(program_path, *, time_budget=60.0) -> {"combined_score": float,
+            "valid": float, "per_instance": {...}}
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import subprocess
+import sys
+import tempfile
+import time
+from pathlib import Path
+from typing import Any
+
+# 保证无论从哪个 cwd/以何种方式加载，都能 import 到同目录的模块
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+from simulator import simulate, load_instance  # noqa: E402
+from validator import candidate_env, check_candidate, check_determinism  # noqa: E402
+
+INVALID_SCORE = 0.0
+DATA_DIR = Path(__file__).resolve().parent / "data" / "instances"
+DEFAULT_GENERATE_COUNT = 8
+# 运行时生成实例的规模循环（与 generator 默认一致）
+GEN_SIZES = (20, 24, 28, 32, 36, 40)
+
+
+def _source_benchmark_dir() -> Path | None:
+    """宿主基准目录（unified 沙箱通过该环境变量暴露），用于读取初始 baseline 做比对。"""
+    raw = os.environ.get("FRONTIER_EVAL_UNIFIED_SOURCE_BENCHMARK_DIR", "").strip()
+    if raw:
+        path = Path(raw)
+        if path.is_dir():
+            return path
+    return None
+
+
+def _parse_on_intervals(raw: str, k: int, horizon: int) -> list[list[list[int]]] | None:
+    """解析候选 stdout 为 on_intervals；非法返回 None。
+
+    格式：{"on": [[[a,b),...], ...]}，on[k] 为电源 k 的开启区间（半开 [a,b)）。
+    """
+    try:
+        text = raw.strip()
+        if not text:
+            return None
+        obj = json.loads(text)
+        if isinstance(obj, dict):
+            obj = obj.get("on")
+        if not isinstance(obj, list) or len(obj) != k:
+            return None
+        out: list[list[list[int]]] = []
+        for ivs in obj:
+            if not isinstance(ivs, list):
+                return None
+            cur: list[list[int]] = []
+            for iv in ivs:
+                if not isinstance(iv, list) or len(iv) != 2:
+                    return None
+                a, b = iv
+                if isinstance(a, bool) or isinstance(b, bool):
+                    return None
+                if not isinstance(a, int) or not isinstance(b, int):
+                    return None
+                if a < 0 or b > horizon or a > b:
+                    return None
+                cur.append([a, b])
+            out.append(cur)
+        return out
+    except Exception:
+        return None
+
+
+def _run_one(program_path: Path, inst_path: Path, time_budget: float,
+             python: str) -> tuple[float, dict[str, Any]]:
+    inst = load_instance(inst_path)
+    k = len(inst["groups"])
+    horizon = int(inst["horizon"])
+    started = time.time()
+    try:
+        proc = subprocess.run(
+            [python, str(program_path), str(inst_path)],
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
+            timeout=time_budget,
+            cwd=str(program_path.parent),
+            env=candidate_env(),
+        )
+        elapsed = time.time() - started
+        if proc.returncode != 0:
+            return 0.0, {"status": "crash", "stderr_tail": proc.stderr[-500:]}
+        on = _parse_on_intervals(proc.stdout, k, horizon)
+        if on is None:
+            return 0.0, {"status": "bad_output", "stdout_tail": proc.stdout[-500:]}
+        minutes = simulate(inst, on)
+        return minutes, {"status": "ok", "on": on, "minutes": round(minutes, 1),
+                         "elapsed_s": round(elapsed, 2)}
+    except subprocess.TimeoutExpired:
+        return 0.0, {"status": "timeout", "budget_s": time_budget}
+    except Exception as exc:
+        return 0.0, {"status": "error", "message": str(exc)}
+
+
+def _load_host_module(mod_name: str):
+    """从宿主 benchmark 目录加载 `verification/<mod>.py`（沙箱内不含该模块时）。
+
+    沙箱内不复制 generator.py（候选不可见）；评测器需要生成器时经
+    FRONTIER_EVAL_UNIFIED_SOURCE_BENCHMARK_DIR 从宿主加载（CVRP 同款模式）。
+    """
+    import importlib.util
+
+    src = os.environ.get("FRONTIER_EVAL_UNIFIED_SOURCE_BENCHMARK_DIR", "").strip()
+    if src and Path(src).is_dir():
+        path = Path(src) / "verification" / f"{mod_name}.py"
+        if path.is_file():
+            spec = importlib.util.spec_from_file_location(f"_tb_{mod_name}", path)
+            mod = importlib.util.module_from_spec(spec)
+            spec.loader.exec_module(mod)
+            return mod
+    return None
+
+
+def _generate_instances(base_seed: int, count: int, out_dir: Path) -> list[Path]:
+    """按种子现场生成新实例（防硬编码）。派生种子 base_seed*1000+i，可复现。
+
+    生成器从宿主加载；生成实例逐个过 `_stagger_ok`（错峰 ≥ 全程开启 25%），
+    不合格跳过——保证生成实例同样奖励调度（与固定实例的验收一致）。
+    """
+    gen_mod = _load_host_module("generator")
+    if gen_mod is None:
+        import generator as gen_mod  # 直跑（非沙箱）时的本地回退
+
+    paths: list[Path] = []
+    i = 0
+    attempts = 0
+    while len(paths) < count and attempts < count * 64:
+        inst = gen_mod.generate(base_seed * 1000 + i, GEN_SIZES[i % len(GEN_SIZES)])
+        i += 1
+        attempts += 1
+        if not gen_mod._stagger_ok(inst):
+            continue
+        path = out_dir / f"gen_{base_seed}_{len(paths) + 1}.json"
+        path.write_text(json.dumps(inst, ensure_ascii=False) + "\n", encoding="utf-8")
+        paths.append(path)
+    return paths
+
+
+def _select_probes(instances: list[Path], n: int = 3) -> list[Path]:
+    """跨规模选确定性探针：按排序取 小/中/大 各一（外加一个生成实例，若有）。"""
+    if len(instances) <= n:
+        return list(instances)
+    idxs = sorted({0, len(instances) // 2, len(instances) - 1})
+    probes = [instances[i] for i in idxs]
+    gen = [p for p in instances if p.name.startswith("gen_")]
+    if gen:
+        probes.append(gen[0])
+    return probes
+
+
+def evaluate(program_path: str, *, time_budget: float = 60.0,
+             data_dir: str | Path | None = None,
+             python: str | None = None) -> dict[str, Any]:
+    """评测候选求解器：返回 metrics 字典。"""
+    prog = Path(program_path).resolve()
+    if not prog.exists():
+        return {"combined_score": 0.0, "valid": 0.0, "per_instance": {},
+                "error": f"program not found: {prog}"}
+
+    # 静态完整性检查（EVOLVE-BLOCK 标记/只读区比对、禁引用、禁绝对路径、禁硬编码）
+    baseline_path = None
+    src_dir = _source_benchmark_dir()
+    if src_dir is not None:
+        candidate_baseline = src_dir / "baseline" / "solver.py"
+        if candidate_baseline.is_file():
+            baseline_path = candidate_baseline
+    violations = check_candidate(prog, baseline_path=baseline_path)
+
+    # 实例池 = 固定实例（本地 data-dir 存在时）+ 运行时生成（设了生成种子时）
+    inst_dir = Path(data_dir).resolve() if data_dir else DATA_DIR
+    instances: list[Path] = []
+    if inst_dir.is_dir():
+        instances = sorted(inst_dir.glob("instance_*.json"))
+
+    gen_seed_raw = os.environ.get("TELECOM_EVAL_GENERATE_SEED", "").strip()
+    tmp_dir: Path | None = None
+    if gen_seed_raw:
+        try:
+            base_seed = int(gen_seed_raw)
+        except ValueError:
+            base_seed = 0
+        gen_count = DEFAULT_GENERATE_COUNT
+        try:
+            gen_count = max(0, int(os.environ.get("TELECOM_EVAL_GENERATE_COUNT", "").strip()))
+        except ValueError:
+            pass
+        if gen_count > 0:
+            tmp_dir = Path(tempfile.mkdtemp(prefix="telecom_eval_"))
+            instances.extend(_generate_instances(base_seed, gen_count, tmp_dir))
+
+    if not instances:
+        return {"combined_score": 0.0, "valid": 0.0, "per_instance": {},
+                "error": f"no instances (data_dir={inst_dir}, generate_seed={gen_seed_raw!r})"}
+
+    py = python or sys.executable
+
+    # 确定性探针：跨规模选若干实例（小/中/大 + 一个生成实例，若有）各跑两次，
+    # 输出必须一致（候选不能只在最小实例上确定）。
+    if not violations:
+        for probe in _select_probes(instances):
+            det_ok, det_note = check_determinism(py, prog, probe, time_budget)
+            if not det_ok:
+                violations = [f"determinism check failed on {probe.name}: {det_note}"]
+                break
+
+    per_instance: dict[str, Any] = {}
+    total = 0.0
+    all_valid = True
+    for inst_path in instances:
+        if violations:
+            per_instance[inst_path.name] = {"status": "preflight_failed",
+                                            "reasons": violations}
+            continue
+        minutes, info = _run_one(prog, inst_path, time_budget, py)
+        per_instance[inst_path.name] = info
+        if minutes <= 0 and info.get("status") != "ok":
+            all_valid = False
+        total += minutes
+
+    score = total / len(instances) if instances else 0.0
+    return {
+        "combined_score": round(score, 2),
+        "valid": 1.0 if all_valid and not violations else 0.0,
+        "per_instance": per_instance,
+        "num_instances": len(instances),
+        "time_budget_s": time_budget,
+        "generate_seed": gen_seed_raw or None,
+    }
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description="区域备电评测")
+    parser.add_argument("solver", help="候选求解器脚本路径")
+    parser.add_argument("--time-budget", type=float, default=60.0,
+                        help="求解时间预算（秒），默认 60")
+    parser.add_argument("--data-dir", type=str, default=None,
+                        help="实例目录（默认 verification/data/instances）")
+    parser.add_argument("--generate-seed", type=int, default=None,
+                        help="运行时生成实例的种子（防硬编码；等价于设 TELECOM_EVAL_GENERATE_SEED）")
+    args = parser.parse_args(argv)
+
+    if args.generate_seed is not None:
+        os.environ["TELECOM_EVAL_GENERATE_SEED"] = str(args.generate_seed)
+    result = evaluate(args.solver, time_budget=args.time_budget, data_dir=args.data_dir)
+    print(json.dumps(result, ensure_ascii=False, indent=2))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/generator.py b/benchmarks/PowerSystems/TelecomBackup/verification/generator.py
new file mode 100644
index 00000000..2f9de45f
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/generator.py
@@ -0,0 +1,154 @@
+"""实例生成器：seed 固定，生成可复现的区域备电实例（JSON）。
+
+参数语义：
+- 区域 200m x 200m，栅格化为 nx*ny 栅格（默认 20x20 = 400）。
+- 站点围绕 4 个聚类中心高斯散布（城区基站分布），坐标裁剪到区域内。
+- 电源分组：对站点坐标做 K-Means（纯 Python），K = ceil(N/3.5)。
+- 电池电量：每个电源 uniform(10, 25) kWh。
+- 栅格需求：每个栅格 uniform(0, 1)（相对负载单位）。
+- 覆盖参数：pt_dbm=20, n_exp=6（阈值 -105dBm 下站点良好覆盖半径约 120m）。
+"""
+
+from __future__ import annotations
+
+import json
+import math
+import random
+from pathlib import Path
+from typing import Any
+
+GRID = {"width": 200.0, "height": 200.0, "nx": 20, "ny": 20}
+CLUSTER_CENTERS = [(50.0, 50.0), (150.0, 50.0), (50.0, 150.0), (150.0, 150.0)]
+SITE_SIGMA = 20.0
+BATTERY_RANGE = (15.0, 30.0)
+P_SILENT = 0.05       # 静默功耗 (kW/站点)：便宜，休息保电收益大
+P_WORK_BASE = 3.0    # 工作功耗基值 (kW/站点)：昂贵，全开惩罚明显
+P_WORK_COEF = 3.0    # 工作功耗负载系数 (kW/负载比)
+DEMAND_MAX = 1.0
+SITE_CAP = 60.0
+HORIZON = 96
+DELTA_MIN = 5.0
+PT_DBM = 20.0
+N_EXP = 6.0
+THRESHOLD = -105.0
+COVERAGE_RATIO = 0.8
+
+
+def _kmeans(points: list[tuple[float, float]], k: int, rng: random.Random,
+            iters: int = 20) -> list[list[int]]:
+    n = len(points)
+    assert n >= k, "not enough sites for k groups"
+    init = rng.sample(range(n), k)
+    centers = [points[i] for i in init]
+    assign: list[int] = [0] * n
+    for _ in range(iters):
+        for i, (x, y) in enumerate(points):
+            assign[i] = min(
+                range(k), key=lambda c: math.hypot(x - centers[c][0], y - centers[c][1])
+            )
+        new_centers = []
+        for c in range(k):
+            members = [points[i] for i in range(n) if assign[i] == c]
+            if members:
+                new_centers.append(
+                    (sum(p[0] for p in members) / len(members),
+                     sum(p[1] for p in members) / len(members))
+                )
+            else:
+                new_centers.append(centers[c])
+        centers = new_centers
+    groups: list[list[int]] = [[] for _ in range(k)]
+    for i, c in enumerate(assign):
+        groups[c].append(i)
+    return [g for g in groups if g]
+
+
+def generate(seed: int, n_sites: int, n_clusters: int = 4) -> dict[str, Any]:
+    rng = random.Random(seed)
+    centers = rng.sample(CLUSTER_CENTERS, n_clusters)
+    sites: list[list[float]] = []
+    for _ in range(n_sites):
+        cx, cy = rng.choice(centers)
+        x = min(195.0, max(5.0, cx + rng.gauss(0.0, SITE_SIGMA)))
+        y = min(195.0, max(5.0, cy + rng.gauss(0.0, SITE_SIGMA)))
+        sites.append([round(x, 1), round(y, 1)])
+
+    k = max(1, math.ceil(n_sites / 3.5))
+    groups = _kmeans([(s[0], s[1]) for s in sites], k, rng=rng)
+    battery = [round(rng.uniform(*BATTERY_RANGE), 1) for _ in groups]
+
+    nx, ny = GRID["nx"], GRID["ny"]
+    demand = [round(rng.uniform(0.0, DEMAND_MAX), 3) for _ in range(nx * ny)]
+
+    inst: dict[str, Any] = {
+        "seed": seed,
+        "grid": GRID,
+        "sites": sites,
+        "groups": groups,
+        "battery": battery,
+        "demand": demand,
+        "pt_dbm": PT_DBM,
+        "n_exp": N_EXP,
+        "threshold": THRESHOLD,
+        "coverage_ratio": COVERAGE_RATIO,
+        "delta_min": DELTA_MIN,
+        "horizon": HORIZON,
+        "site_cap": SITE_CAP,
+        "p_silent": P_SILENT,
+        "p_work_base": P_WORK_BASE,
+        "p_work_coef": P_WORK_COEF,
+    }
+    return inst
+
+
+
+
+def _stagger_ok(inst: dict[str, Any], min_gap: float = 0.25) -> bool:
+    """验收：错峰调度（同时休息多路）须比全程开启好 >= min_gap。
+
+    全程开启电池并行耗尽；让多路电源同时休息（其余保持负载分散）能大幅保电
+    并延长总备电时长。实例几何/电池若不满足，则此实例对"调度"无意义，应重造。
+    """
+    from simulator import simulate
+
+    horizon = int(inst["horizon"])
+    k = len(inst["groups"])
+    on_all = [[[0, horizon]] for _ in range(k)]
+    ao = simulate(inst, on_all)
+    best = ao
+    for n_rest in (1, 2, 3):
+        for r in (4, 8, 12, 16):
+            on = [[] for _ in range(k)]
+            slot = seg = 0
+            while slot < horizon:
+                for i in range(k):
+                    if (seg + i) % k >= n_rest:
+                        on[i].append([slot, min(slot + r, horizon)])
+                slot += r
+                seg += 1
+            best = max(best, simulate(inst, on))
+    return best >= ao * (1.0 + min_gap)
+
+def main() -> None:
+    out_dir = Path(__file__).resolve().parent / "data" / "instances"
+    out_dir.mkdir(parents=True, exist_ok=True)
+    specs = [
+        (1, 20), (2, 24), (3, 28), (4, 32),
+        (5, 36), (6, 40), (7, 24), (8, 32),
+    ]
+    for idx, (seed, n_sites) in enumerate(specs):
+        inst = None
+        for trial in range(300):
+            s = seed + trial * 101
+            cand = generate(s, n_sites)
+            if _stagger_ok(cand):
+                inst = cand
+                break
+        assert inst is not None, f"no acceptable instance for slot {idx}"
+        path = out_dir / f"instance_{seed}.json"
+        path.write_text(json.dumps(inst, ensure_ascii=False, indent=1) + "\n", encoding="utf-8")
+        print(f"instance_{seed}.json  N={n_sites}  K={len(inst['groups'])}  seed={inst['seed']}  (stagger ok)")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/ref_solver.py b/benchmarks/PowerSystems/TelecomBackup/verification/ref_solver.py
new file mode 100644
index 00000000..88b8e621
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/ref_solver.py
@@ -0,0 +1,55 @@
+#!/usr/bin/env python3
+"""参考求解器：轮流休息一路的错峰调度（"好解"参考，供 agent 对比/超越）。
+
+思路：全程开启会让所有电池并行耗尽；"轮流休息一路"（其余保持负载分散）
+能让被休息的电源以低静默功耗保电，从而延长总备电时长。对 R（休息窗口）
+做小范围搜索取最优。
+
+用法：python verification/ref_solver.py <instance.json>
+输出：stdout 打印 {"on": [[[a,b),...], ...]}
+"""
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+_here = Path(__file__).resolve().parent
+if str(_here) not in sys.path:
+    sys.path.insert(0, str(_here))
+from simulator import simulate  # noqa: E402
+
+
+def solve(inst: dict) -> list[list[list[int]]]:
+    horizon = int(inst["horizon"])
+    k = len(inst["groups"])
+    best_on = [[[0, horizon]] for _ in range(k)]
+    best_s = simulate(inst, best_on)
+    for n_rest in (1, 2, 3):
+        for r in (2, 4, 6, 8, 12, 16):
+            on = [[] for _ in range(k)]
+            slot = seg = 0
+            while slot < horizon:
+                for i in range(k):
+                    if (seg + i) % k >= n_rest:
+                        on[i].append([slot, min(slot + r, horizon)])
+                slot += r
+                seg += 1
+            s = simulate(inst, on)
+            if s > best_s:
+                best_s, best_on = s, on
+    return best_on
+
+
+def main() -> int:
+    if len(sys.argv) < 2:
+        print("usage: python ref_solver.py <instance.json>", file=sys.stderr)
+        return 2
+    inst_path = Path(sys.argv[1])
+    inst = json.loads(inst_path.read_text(encoding="utf-8"))
+    print(json.dumps({"on": solve(inst)}))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/requirements.txt b/benchmarks/PowerSystems/TelecomBackup/verification/requirements.txt
new file mode 100644
index 00000000..4e66700e
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/requirements.txt
@@ -0,0 +1,2 @@
+# The evaluator uses only the Python standard library.
+# Runtime requirement: Python >= 3.10 (no third-party dependencies).
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/simulator.py b/benchmarks/PowerSystems/TelecomBackup/verification/simulator.py
new file mode 100644
index 00000000..7c929cec
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/simulator.py
@@ -0,0 +1,155 @@
+"""计分模拟器：验证候选电源开关策略，返回备电时长（分钟）。
+
+规则（自编简化模型，物理语义见 Task.md / README）：
+- 区域为 nx*ny 个栅格；N 个站点；K 个电源，每组站点共用一个电池。
+- 覆盖电平 P(g, s) = pt_dbm - 10*n_exp*log10(d(g,s)+1)（dBm，d 为米）。
+- 栅格接入"最强存活站点"；栅格"良好"当且仅当 max P(g,s) > threshold。
+- 站点功耗（由实例提供，默认见下）：工作 P_work = p_work_base + p_work_coef*min(load_s/site_cap, 1.0) kW；静默 p_silent kW。
+  实例自带校准（当前数据：p_silent=0.05, p_work_base=3.0, p_work_coef=3.0），保证错峰调度有明显收益。
+- 负载迁移：电源关闭 -> 站点静默（不提供覆盖，仍耗静默电）-> 栅格即时接入最强存活站点。
+- 电源 k 在时隙 t > close_time[k] 时关闭；电量 E_k 每时隙扣 dt*sum(P)，耗尽 -> 停服（不耗电不覆盖）。
+- 备电时长 = 首次覆盖比例 < coverage_ratio 的时隙的上一时刻（分钟）；撑满 horizon 则 T*delta_min。
+"""
+
+from __future__ import annotations
+
+import json
+import math
+from pathlib import Path
+from typing import Any
+
+# 默认参数（与 generator.py 保持一致）
+PT_DBM = 20.0
+N_EXP = 6.0
+THRESHOLD = -105.0
+COVERAGE_RATIO = 0.8
+DELTA_MIN = 5.0
+SITE_CAP = 60.0
+
+
+def load_instance(path: str | Path) -> dict[str, Any]:
+    with open(path, "r", encoding="utf-8") as f:
+        return json.load(f)
+
+
+def build_power(inst: dict[str, Any]) -> list[list[float]]:
+    """返回 (N, G) 的覆盖电平矩阵 P(s, g)（dBm）。"""
+    grid = inst["grid"]
+    nx, ny = grid["nx"], grid["ny"]
+    w, h = grid["width"], grid["height"]
+    pt = float(inst.get("pt_dbm", PT_DBM))
+    n_exp = float(inst.get("n_exp", N_EXP))
+    centers = [
+        ((ix + 0.5) * w / nx, (iy + 0.5) * h / ny) for iy in range(ny) for ix in range(nx)
+    ]
+    rows: list[list[float]] = []
+    for sx, sy in inst["sites"]:
+        row = [
+            pt - 10.0 * n_exp * math.log10(math.hypot(sx - cx, sy - cy) + 1.0)
+            for cx, cy in centers
+        ]
+        rows.append(row)
+    return rows
+
+
+def _assign(power: list[list[float]], served: list[int], demand: list[float],
+            threshold: float) -> tuple[list[float], int]:
+    """给每个栅格接入最强存活站点；返回 (每站点负载, 良好栅格数)。"""
+    n_sites = len(power)
+    n_cells = len(demand)
+    best_val = [-1e9] * n_cells
+    best_site = [-1] * n_cells
+    for s in served:
+        prow = power[s]
+        for g in range(n_cells):
+            v = prow[g]
+            if v > best_val[g]:
+                best_val[g] = v
+                best_site[g] = s
+    load = [0.0] * n_sites
+    good = 0
+    for g in range(n_cells):
+        if best_val[g] > threshold:
+            good += 1
+        s = best_site[g]
+        if s >= 0:
+            load[s] += demand[g]
+    return load, good
+
+
+def _normalize_intervals(intervals: list[list[int]], horizon: int) -> list[list[int]]:
+    """合并重叠/相邻区间并排序，返回升序不交区间 [a, b)。"""
+    ivs = sorted((a, b) for a, b in intervals if a < b)
+    merged: list[list[int]] = []
+    for a, b in ivs:
+        if merged and a <= merged[-1][1]:
+            merged[-1][1] = max(merged[-1][1], b)
+        else:
+            merged.append([a, b])
+    return merged
+
+
+def simulate(inst: dict[str, Any], on_intervals: list[list[list[int]]]) -> float:
+    """返回备电时长（分钟）。
+
+    on_intervals[k] = 电源 k 的开启时隙区间列表，如 [[0, 48], [60, 96]]，
+    表示时隙 0..47 与 60..95 开启（时隙 0-index，半开区间 [a, b)）。
+    on_intervals[k] = [[0, horizon]] 表示永不关闭；[] 表示一开始就关。
+    """
+    power = build_power(inst)
+    groups: list[list[int]] = inst["groups"]
+    demand: list[float] = inst["demand"]
+    # 功耗参数由实例提供（数据自带校准）：静默功耗、工作基值、工作负载系数。
+    p_silent = float(inst.get("p_silent", 1.0))
+    p_work_base = float(inst.get("p_work_base", 1.2))
+    p_work_coef = float(inst.get("p_work_coef", 2.0))
+    threshold = float(inst.get("threshold", THRESHOLD))
+    coverage_ratio = float(inst.get("coverage_ratio", COVERAGE_RATIO))
+    delta_min = float(inst.get("delta_min", DELTA_MIN))
+    site_cap = float(inst.get("site_cap", SITE_CAP))
+    horizon = int(inst["horizon"])
+    n_cells = len(demand)
+    n_sites = len(inst["sites"])
+    required_good = coverage_ratio * n_cells
+    dt_hour = delta_min / 60.0
+
+    # 预计算每时隙每电源开关状态（K x horizon 布尔）
+    on_states: list[list[bool]] = []
+    for k in range(len(groups)):
+        ivs = _normalize_intervals(on_intervals[k], horizon)
+        row = [False] * horizon
+        for a, b in ivs:
+            for s in range(max(0, a), min(b, horizon)):
+                row[s] = True
+        on_states.append(row)
+
+    def work_sites(slot: int, battery: list[float]) -> list[int]:
+        served: list[int] = []
+        for k in range(len(groups)):
+            if on_states[k][slot] and battery[k] > 0:
+                served.extend(groups[k])
+        return served
+
+    battery = [float(e) for e in inst["battery"]]
+    for slot in range(horizon):
+        # 本时隙开始时的工作站点（扣电前）
+        served_now = work_sites(slot, battery)
+        load, _ = _assign(power, served_now, demand, threshold)
+        # 电量推进
+        for k in range(len(groups)):
+            if battery[k] <= 0:
+                continue
+            e_k = 0.0
+            for s in groups[k]:
+                if on_states[k][slot]:
+                    load_ratio = min(load[s] / site_cap, 1.0)
+                    e_k += p_work_base + p_work_coef * load_ratio  # 工作功耗
+                else:
+                    e_k += p_silent  # 静默功耗
+            battery[k] -= dt_hour * e_k
+        # 扣电后覆盖检查
+        served_after = work_sites(slot, battery)
+        _, good = _assign(power, served_after, demand, threshold)
+        if good < required_good:
+            return slot * delta_min
+    return horizon * delta_min
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/test_evaluator.py b/benchmarks/PowerSystems/TelecomBackup/verification/test_evaluator.py
new file mode 100644
index 00000000..979f415a
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/test_evaluator.py
@@ -0,0 +1,117 @@
+"""Unit tests for verification/evaluate.py (stdlib unittest, no third-party deps).
+
+Run from the TelecomBackup task directory:
+    python verification/test_evaluator.py
+"""
+from __future__ import annotations
+
+import json
+import os
+import shutil
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+import evaluate as ev  # noqa: E402
+from generator import generate  # noqa: E402
+
+TASK_ROOT = Path(__file__).resolve().parents[1]
+BASELINE = TASK_ROOT / "baseline" / "solver.py"
+
+ALWAYS_ON = (
+    "# EVOLVE-BLOCK-START\n"
+    "def solve(inst):\n"
+    "    k = len(inst['groups']); T = inst['horizon']\n"
+    "    return [[[0, T]] for _ in range(k)]\n"
+    "# EVOLVE-BLOCK-END\n"
+    "import json, sys\n"
+    "inst = json.load(open(sys.argv[1], encoding='utf-8'))\n"
+    "print(json.dumps({'on': solve(inst)}))\n"
+)
+
+
+class EvaluatorTestCase(unittest.TestCase):
+    def setUp(self):
+        self.tmp = Path(tempfile.mkdtemp(prefix="telecom_eval_test_"))
+        self.inst_dir = self.tmp / "instances"
+        self.inst_dir.mkdir()
+        # 3 个固定实例（seed 派生），使评测稳定
+        for i in range(3):
+            inst = generate(1000 + i, 20 + 8 * i)
+            (self.inst_dir / f"instance_{i}.json").write_text(
+                json.dumps(inst), encoding="utf-8"
+            )
+
+    def tearDown(self):
+        shutil.rmtree(self.tmp, ignore_errors=True)
+        os.environ.pop("TELECOM_EVAL_GENERATE_SEED", None)
+        os.environ.pop("TELECOM_EVAL_GENERATE_COUNT", None)
+
+    def _write(self, name: str, src: str) -> Path:
+        p = self.tmp / name
+        p.write_text(src, encoding="utf-8")
+        return p
+
+    def _run(self, solver: Path) -> dict:
+        return ev.evaluate(str(solver), time_budget=30.0, data_dir=self.inst_dir)
+
+    def test_always_on_scores_positive_and_valid(self):
+        solver = self._write("always_on.py", ALWAYS_ON)
+        result = self._run(solver)
+        self.assertEqual(result["valid"], 1.0)
+        self.assertGreater(result["combined_score"], 0.0)
+        self.assertEqual(result["num_instances"], 3)
+
+    def test_malformed_output_scores_zero(self):
+        solver = self._write("bad.py", "print('not json')\n")
+        result = self._run(solver)
+        self.assertEqual(result["combined_score"], 0.0)
+        self.assertEqual(result["valid"], 0.0)
+
+    def test_timeout_scores_zero(self):
+        solver = self._write("slow.py", "import time; time.sleep(60)\n")
+        result = ev.evaluate(str(solver), time_budget=0.5, data_dir=self.inst_dir)
+        self.assertEqual(result["combined_score"], 0.0)
+        self.assertEqual(result["valid"], 0.0)
+
+    def test_hardcoded_candidate_fails_preflight(self):
+        src = ALWAYS_ON.replace(
+            "    return [[[0, T]] for _ in range(k)]",
+            "    table = {'instance_0': [[[0, 48]]]}\n"
+            "    return table.get('instance_0', [[[0, T]] for _ in range(k)])",
+            1,
+        )
+        solver = self._write("hardcode.py", src)
+        result = self._run(solver)
+        self.assertEqual(result["valid"], 0.0)
+        self.assertTrue(
+            any("preflight" in v.get("status", "") for v in result["per_instance"].values())
+        )
+
+    def test_missing_program_returns_error(self):
+        result = self._run(self.tmp / "nope.py")
+        self.assertEqual(result["valid"], 0.0)
+        self.assertIn("error", result)
+
+    def test_generate_seed_adds_instances(self):
+        os.environ["TELECOM_EVAL_GENERATE_SEED"] = "7"
+        os.environ["TELECOM_EVAL_GENERATE_COUNT"] = "4"
+        solver = self._write("always_on2.py", ALWAYS_ON)
+        result = self._run(solver)
+        self.assertEqual(result["num_instances"], 3 + 4)
+        self.assertEqual(result["generate_seed"], "7")
+        self.assertEqual(result["valid"], 1.0)
+
+    def test_generated_instances_are_reproducible(self):
+        os.environ["TELECOM_EVAL_GENERATE_SEED"] = "7"
+        os.environ["TELECOM_EVAL_GENERATE_COUNT"] = "2"
+        solver = self._write("always_on3.py", ALWAYS_ON)
+        a = self._run(solver)
+        b = self._run(solver)
+        self.assertEqual(a["combined_score"], b["combined_score"])
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/test_frontier_eval_evaluator.py b/benchmarks/PowerSystems/TelecomBackup/verification/test_frontier_eval_evaluator.py
new file mode 100644
index 00000000..8fa8a68c
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/test_frontier_eval_evaluator.py
@@ -0,0 +1,76 @@
+"""Sandbox-evaluator end-to-end tests (frontier_eval/evaluator.py).
+
+The unified runtime runs the eval from the sandbox: the sandbox evaluator
+loads `verification/evaluate.py` (copied in), and the generation path
+host-loads `generator.py` via FRONTIER_EVAL_UNIFIED_SOURCE_BENCHMARK_DIR.
+These tests lock that wiring so a refactor cannot silently break the sandbox
+path (the review finding that flagged no end-to-end sandbox coverage).
+"""
+from __future__ import annotations
+
+import importlib.util
+import json
+import os
+import shutil
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+TASK_ROOT = Path(__file__).resolve().parents[1]
+EVALUATOR = TASK_ROOT / "frontier_eval" / "evaluator.py"
+BASELINE = TASK_ROOT / "baseline" / "solver.py"
+
+_spec = importlib.util.spec_from_file_location("_sandbox_eval", EVALUATOR)
+_sb = importlib.util.module_from_spec(_spec)
+assert _spec.loader is not None
+_spec.loader.exec_module(_sb)
+
+
+class TestSandboxBaseline(unittest.TestCase):
+    def test_baseline_scores(self):
+        m = _sb.evaluate(str(BASELINE))["metrics"]
+        self.assertEqual(m["valid"], 1.0)
+        self.assertEqual(m["num_instances"], 8)
+        self.assertGreater(m["combined_score"], 100.0)
+
+
+class TestSandboxHostLoad(unittest.TestCase):
+    def test_generation_host_loads_generator(self):
+        # Sandbox view: the host benchmark dir is exposed via the env var.
+        os.environ["FRONTIER_EVAL_UNIFIED_SOURCE_BENCHMARK_DIR"] = str(TASK_ROOT)
+        os.environ["TELECOM_EVAL_GENERATE_SEED"] = "7"
+        os.environ["TELECOM_EVAL_GENERATE_COUNT"] = "4"
+        try:
+            m = _sb.evaluate(str(BASELINE))["metrics"]
+        finally:
+            os.environ.pop("FRONTIER_EVAL_UNIFIED_SOURCE_BENCHMARK_DIR", None)
+            os.environ.pop("TELECOM_EVAL_GENERATE_SEED", None)
+            os.environ.pop("TELECOM_EVAL_GENERATE_COUNT", None)
+        self.assertEqual(m["valid"], 1.0)
+        self.assertGreaterEqual(m["num_instances"], 12)  # 8 fixed + generated
+
+
+class TestSandboxRejectsRefSolver(unittest.TestCase):
+    def test_ref_solver_cheat_rejected(self):
+        src = BASELINE.read_text(encoding="utf-8")
+        start = src.find("# EVOLVE-BLOCK-START")
+        end = src.find("# EVOLVE-BLOCK-END")
+        injected = (
+            src[: start + len("# EVOLVE-BLOCK-START")]
+            + "\n    from verification.ref_solver import solve\n    return solve(inst)\n"
+            + src[end:]
+        )
+        tmp = Path(tempfile.mkdtemp(prefix="tb_cheat_"))
+        cand = tmp / "cheat.py"
+        cand.write_text(injected, encoding="utf-8")
+        try:
+            m = _sb.evaluate(str(cand))["metrics"]
+        finally:
+            shutil.rmtree(tmp, ignore_errors=True)
+        self.assertEqual(m["valid"], 0.0)
+        self.assertEqual(m["combined_score"], 0.0)
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/test_simulator.py b/benchmarks/PowerSystems/TelecomBackup/verification/test_simulator.py
new file mode 100644
index 00000000..abfa8b76
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/test_simulator.py
@@ -0,0 +1,110 @@
+"""Unit tests for verification/simulator.py (stdlib unittest, no third-party deps).
+
+Run from the TelecomBackup task directory:
+    python verification/test_simulator.py
+"""
+from __future__ import annotations
+
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+import simulator as sim  # noqa: E402
+
+
+def _tiny() -> dict:
+    """1 站 1 电源 1 栅格：手算用例。"""
+    return {
+        "grid": {"width": 200.0, "height": 200.0, "nx": 1, "ny": 1},
+        "sites": [[100.0, 100.0]],
+        "groups": [[0]],
+        "battery": [10.0],
+        "demand": [0.5],
+        "pt_dbm": 20.0, "n_exp": 6.0, "threshold": -105.0,
+        "coverage_ratio": 0.8, "delta_min": 5.0, "horizon": 8, "site_cap": 60.0,
+    }
+
+
+class TestTinyManual(unittest.TestCase):
+    def setUp(self):
+        self.inst = _tiny()
+
+    def test_always_on(self):
+        # 工作功耗 1.2+2*min(0.5/60,1)=1.2167kW，8 时隙耗电远小于 10kWh -> 撑满 40min
+        self.assertEqual(sim.simulate(self.inst, [[[0, 8]]]), 40.0)
+
+    def test_never_on(self):
+        # 一开始就关 -> t=0 覆盖 0 < 80% -> 0 分钟
+        self.assertEqual(sim.simulate(self.inst, [[]]), 0.0)
+
+    def test_half_horizon(self):
+        # 前 4 时隙开、后关 -> 第 5 时隙跌破 -> 20 分钟
+        self.assertEqual(sim.simulate(self.inst, [[[0, 4]]]), 20.0)
+
+
+class TestNormalizeIntervals(unittest.TestCase):
+    def test_merge_overlap(self):
+        self.assertEqual(sim._normalize_intervals([[0, 10], [5, 15]], 96), [[0, 15]])
+
+    def test_merge_adjacent(self):
+        self.assertEqual(sim._normalize_intervals([[0, 5], [5, 10]], 96), [[0, 10]])
+
+    def test_empty_interval_list(self):
+        self.assertEqual(sim._normalize_intervals([], 96), [])
+
+    def test_out_of_order(self):
+        # 乱序的合法区间：排序后升序返回
+        self.assertEqual(sim._normalize_intervals([[3, 8], [0, 2]], 96), [[0, 2], [3, 8]])
+
+    def test_invalid_interval_filtered(self):
+        # a >= b 的畸形区间被忽略
+        self.assertEqual(sim._normalize_intervals([[8, 3], [0, 2]], 96), [[0, 2]])
+
+
+class TestBatteryDepletion(unittest.TestCase):
+    def test_tiny_battery_dies(self):
+        inst = _tiny()
+        inst["battery"] = [0.05]  # 1 个时隙都撑不过
+        # 全开：第 1 时隙扣电后耗尽停服 -> 覆盖掉到 0 -> 该时隙跌破 -> 0 分钟
+        self.assertEqual(sim.simulate(inst, [[[0, 8]]]), 0.0)
+
+    def test_multi_power_some_dead(self):
+        inst = {
+            "grid": {"width": 200.0, "height": 200.0, "nx": 2, "ny": 2},
+            "sites": [[50, 50], [150, 150]],
+            "groups": [[0], [1]],
+            "battery": [0.05, 20.0],
+            "demand": [0.5] * 4,
+            "pt_dbm": 20.0, "n_exp": 6.0, "threshold": -105.0,
+            "coverage_ratio": 0.4, "delta_min": 5.0, "horizon": 10, "site_cap": 60.0,
+        }
+        # 电源 0 立即耗尽停服，电源 1 继续服务；覆盖要求 40% 很容易满足 -> 撑满
+        self.assertEqual(sim.simulate(inst, [[[0, 10]], [[0, 10]]]), 50.0)
+
+
+class TestDeterminism(unittest.TestCase):
+    def test_same_input_same_output(self):
+        inst = _tiny()
+        a = sim.simulate(inst, [[[0, 8]]])
+        b = sim.simulate(inst, [[[0, 8]]])
+        self.assertEqual(a, b)
+
+
+class TestCoverageConstraint(unittest.TestCase):
+    def test_coverage_falls_below_ratio_ends_early(self):
+        inst = _tiny()
+        inst["coverage_ratio"] = 0.9  # 单站时覆盖不足，需两站才满足
+        inst["sites"] = [[50, 50], [150, 150]]
+        inst["groups"] = [[0], [1]]
+        inst["battery"] = [20.0, 20.0]
+        inst["demand"] = [0.5] * 4  # 与 2x2 栅格匹配
+        inst["grid"] = {"width": 200.0, "height": 200.0, "nx": 2, "ny": 2}
+        # 只开电源 0：(150,150) 栅格无良好覆盖 -> 良好 3/4=75% < 90% -> t=0 跌破 -> 0 分钟
+        self.assertEqual(sim.simulate(inst, [[[0, 8]], []]), 0.0)
+        # 两个都开 -> 覆盖 4/4=100% >= 90% -> 撑满 8 时隙（horizon=8）= 40 分钟
+        self.assertEqual(sim.simulate(inst, [[[0, 8]], [[0, 8]]]), 40.0)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/test_validator.py b/benchmarks/PowerSystems/TelecomBackup/verification/test_validator.py
new file mode 100644
index 00000000..042cfe97
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/test_validator.py
@@ -0,0 +1,170 @@
+"""Unit tests for verification/validator.py (stdlib unittest, no third-party deps).
+
+Run from the TelecomBackup task directory:
+    python verification/test_validator.py
+"""
+from __future__ import annotations
+
+import json
+import shutil
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+import validator as vd  # noqa: E402
+
+TASK_ROOT = Path(__file__).resolve().parents[1]
+BASELINE = TASK_ROOT / "baseline" / "solver.py"
+BASELINE_SRC = BASELINE.read_text(encoding="utf-8")
+
+
+class TestStaticChecks(unittest.TestCase):
+    def setUp(self):
+        self.tmp = Path(tempfile.mkdtemp(prefix="telecom_val_"))
+
+    def tearDown(self):
+        shutil.rmtree(self.tmp, ignore_errors=True)
+
+    def _check(self, src: str) -> list[str]:
+        path = self.tmp / "candidate.py"
+        path.write_text(src, encoding="utf-8")
+        return vd.check_candidate(path, baseline_path=BASELINE)
+
+    def test_baseline_is_clean(self):
+        self.assertEqual(vd.check_candidate(BASELINE, baseline_path=BASELINE), [])
+
+    def test_missing_markers(self):
+        src = BASELINE_SRC.replace(vd.EVOLVE_START, "").replace(vd.EVOLVE_END, "")
+        issues = self._check(src)
+        self.assertTrue(any("EVOLVE-BLOCK" in i for i in issues))
+
+    def test_outside_block_changed(self):
+        src = BASELINE_SRC.replace('"""区域备电 baseline', '"""HACKED baseline', 1)
+        issues = self._check(src)
+        self.assertTrue(any("outside EVOLVE-BLOCK" in i for i in issues))
+
+    def test_hardcode_by_instance_name(self):
+        src = BASELINE_SRC.replace(
+            'return [[[0, horizon]] for _ in range(k)]',
+            'table = {"instance_1": [[[0, 48]]]}\n    return table',
+            1,
+        )
+        issues = self._check(src)
+        self.assertTrue(any("hardcodes" in i for i in issues))
+
+    def test_import_generator(self):
+        src = BASELINE_SRC.replace(
+            'return [[[0, horizon]] for _ in range(k)]',
+            "from generator import generate\n    return []",
+            1,
+        )
+        issues = self._check(src)
+        self.assertTrue(any("forbidden" in i for i in issues))
+
+    def test_import_verification_evaluate_dotted(self):
+        src = BASELINE_SRC.replace(
+            'return [[[0, horizon]] for _ in range(k)]',
+            "import verification.evaluate\n    return []",
+            1,
+        )
+        issues = self._check(src)
+        self.assertTrue(any("forbidden" in i for i in issues))
+
+    def test_import_ref_solver_rejected(self):
+        # P0：候选不得白嫖参考求解器（ref_solver）。
+        src = BASELINE_SRC.replace(
+            'return [[[0, horizon]] for _ in range(k)]',
+            "from verification.ref_solver import solve\n    return solve(inst)",
+            1,
+        )
+        issues = self._check(src)
+        self.assertTrue(any("forbidden" in i or "ref_solver" in i for i in issues))
+
+    def test_absolute_path(self):
+        src = BASELINE_SRC.replace(
+            'return [[[0, horizon]] for _ in range(k)]',
+            'p = "C:\\\\Users\\\\secret"\n    return []',
+            1,
+        )
+        issues = self._check(src)
+        self.assertTrue(any("absolute" in i for i in issues))
+
+    def test_whitebox_simulator_import_is_allowed(self):
+        # 候选被允许用白盒计分器做内部搜索（Task.md 承诺的能力）
+        src = BASELINE_SRC.replace(
+            'return [[[0, horizon]] for _ in range(k)]',
+            "from simulator import simulate\n    return [[[0, horizon]] for _ in range(k)]",
+            1,
+        )
+        self.assertEqual(self._check(src), [])
+
+
+class TestCandidateEnv(unittest.TestCase):
+    def test_strips_frontier_and_telecom_vars(self):
+        import os
+
+        old = dict(os.environ)
+        os.environ["FRONTIER_ENGINEERING_ROOT"] = "C:\\repo"
+        os.environ["FRONTIER_EVAL_UNIFIED_PYTHON"] = "C:\\python"
+        os.environ["TELECOM_EVAL_GENERATE_SEED"] = "42"
+        os.environ["KEEP_ME"] = "x"
+        try:
+            env = vd.candidate_env()
+        finally:
+            os.environ.clear()
+            os.environ.update(old)
+        self.assertNotIn("FRONTIER_ENGINEERING_ROOT", env)
+        self.assertNotIn("FRONTIER_EVAL_UNIFIED_PYTHON", env)
+        self.assertNotIn("TELECOM_EVAL_GENERATE_SEED", env)
+        self.assertEqual(env.get("KEEP_ME"), "x")
+
+
+class TestDeterminism(unittest.TestCase):
+    def setUp(self):
+        self.tmp = Path(tempfile.mkdtemp(prefix="telecom_det_"))
+        self.inst = self.tmp / "inst.json"
+        self.inst.write_text(json.dumps({
+            "grid": {"width": 200.0, "height": 200.0, "nx": 2, "ny": 2},
+            "sites": [[50, 50], [150, 150]],
+            "groups": [[0], [1]],
+            "battery": [12.0, 8.0],
+            "demand": [0.5] * 4,
+            "pt_dbm": 20.0, "n_exp": 6.0, "threshold": -105.0,
+            "coverage_ratio": 0.4, "delta_min": 5.0, "horizon": 10, "site_cap": 60.0,
+        }), encoding="utf-8")
+
+    def tearDown(self):
+        shutil.rmtree(self.tmp, ignore_errors=True)
+
+    def _write(self, name: str, src: str) -> Path:
+        p = self.tmp / name
+        p.write_text(src, encoding="utf-8")
+        return p
+
+    def test_deterministic_candidate_passes(self):
+        solver = self._write("det.py", (
+            "import json, sys\n"
+            "inst = json.load(open(sys.argv[1], encoding='utf-8'))\n"
+            "k = len(inst['groups']); T = inst['horizon']\n"
+            "print(json.dumps({'on': [[[0, T]] for _ in range(k)]}))\n"
+        ))
+        ok, note = vd.check_determinism(sys.executable, solver, self.inst, 30.0)
+        self.assertTrue(ok, note)
+
+    def test_random_candidate_fails(self):
+        solver = self._write("rand.py", (
+            "import json, random, sys\n"
+            "inst = json.load(open(sys.argv[1], encoding='utf-8'))\n"
+            "k = len(inst['groups']); T = inst['horizon']\n"
+            "iv = [[0, random.randint(0, T)] for _ in range(1)]\n"
+            "print(json.dumps({'on': [iv for _ in range(k)]}))\n"
+        ))
+        ok, note = vd.check_determinism(sys.executable, solver, self.inst, 30.0)
+        self.assertFalse(ok)
+        self.assertIn("not deterministic", note)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/benchmarks/PowerSystems/TelecomBackup/verification/validator.py b/benchmarks/PowerSystems/TelecomBackup/verification/validator.py
new file mode 100644
index 00000000..093dd384
--- /dev/null
+++ b/benchmarks/PowerSystems/TelecomBackup/verification/validator.py
@@ -0,0 +1,159 @@
+"""TelecomBackup 候选完整性校验（借鉴 CVRP validator，简化版）。
+
+在评分前以"可执行检查"强制任务约束（不只是自然语言）：
+  1. EVOLVE-BLOCK 完整性：标记必须存在；标记外的代码必须与初始 baseline 逐字节一致。
+  2. 禁引用：候选不得 import verification 的评测/生成模块（evaluate/generator），
+     不得出现绝对路径，不得按实例名硬编码调度。
+  3. 确定性：同一实例跑两次必须输出一致（调度器需要可复现）。
+  4. candidate_env：候选子进程环境剥离 FRONTIER_*/TELECOM_EVAL_* 变量，
+     封死"通过宿主环境变量定位评测基线"的侧信道。
+
+候选允许 import `verification/simulator.py`（白盒计分器，任务有意暴露），
+但不允许 import 评测/生成逻辑。纯标准库。
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import re
+import subprocess
+import tempfile
+from pathlib import Path
+from typing import Any
+
+EVOLVE_START = "EVOLVE-BLOCK-START"
+EVOLVE_END = "EVOLVE-BLOCK-END"
+
+# 禁引用 token：强 token 子串匹配即可（正常求解器几乎不会出现）。
+STRONG_TOKENS = (
+    "generate_instance",
+    "TELECOM_EVAL_",
+    "from generator",
+    "import generator",
+    "ref_solver",
+)
+# "verification" 只在与评测/生成模块构成 import/路径上下文时禁止
+# （候选被允许 import verification/simulator.py）。
+FORBIDDEN_RE = (
+    re.compile(r"verification[\\/](?:evaluate|generator)"),
+    re.compile(r"verification\s*\.\s*(?:evaluate|generator)\b"),
+    re.compile(r"\b(?:from|import)\s+(?:evaluate|generator)\b"),
+)
+# Windows 盘符 / POSIX 家目录绝对路径。
+ABS_PATH_RE = re.compile(r"[A-Za-z]:[\\/]|/home/|/Users/")
+# 按实例名硬编码调度，如 "instance_1": [...] 或 "gen_42_1": [...]
+HARDCODE_RE = re.compile(r"[\"'](?:instance|gen)_\d+(?:_\d+)?[\"']\s*:")
+
+
+def split_evolve_blocks(src: str) -> tuple[str, str, str] | None:
+    start = src.find(EVOLVE_START)
+    end = src.find(EVOLVE_END)
+    if start == -1 or end == -1 or end <= start:
+        return None
+    return (
+        src[:start],
+        src[start + len(EVOLVE_START) : end],
+        src[end + len(EVOLVE_END) :],
+    )
+
+
+def fixed_region(parts: tuple[str, str, str]) -> str:
+    """EVOLVE-BLOCK 之外的只读部分。
+
+    行尾归一化：CRLF/LF 差异与结尾缺换行（保存/复制时的常见无损差异）不计为
+    "修改了固定区"——固定区是否被改看内容，不看行尾。
+    """
+    return (parts[0] + parts[2]).replace("\r\n", "\n").rstrip("\n")
+
+
+def static_check_source(src: str, baseline_src: str | None = None) -> list[str]:
+    issues: list[str] = []
+    parts = split_evolve_blocks(src)
+    if parts is None:
+        issues.append("missing EVOLVE-BLOCK-START / EVOLVE-BLOCK-END markers")
+    elif baseline_src is not None:
+        init_parts = split_evolve_blocks(baseline_src)
+        if init_parts is not None and fixed_region(init_parts) != fixed_region(parts):
+            issues.append("code outside EVOLVE-BLOCK differs from initial baseline")
+
+    for token in STRONG_TOKENS:
+        if token in src:
+            issues.append(f"candidate references forbidden token {token!r}")
+    for pat in FORBIDDEN_RE:
+        if pat.search(src):
+            issues.append("candidate references forbidden evaluation/generation module")
+
+    if ABS_PATH_RE.search(src):
+        issues.append("candidate contains an absolute filesystem path")
+    if HARDCODE_RE.search(src):
+        issues.append("candidate hardcodes per-instance schedules by name")
+
+    return issues
+
+
+def check_candidate(
+    solver_path: Path | str, baseline_path: Path | str | None = None
+) -> list[str]:
+    solver_path = Path(solver_path)
+    try:
+        src = solver_path.read_text(encoding="utf-8", errors="replace")
+    except Exception as exc:
+        return [f"cannot read candidate source: {exc}"]
+
+    baseline_src = None
+    if baseline_path is not None:
+        try:
+            baseline_src = Path(baseline_path).read_text(
+                encoding="utf-8", errors="replace"
+            )
+        except Exception:
+            baseline_src = None
+    return static_check_source(src, baseline_src)
+
+
+def candidate_env() -> dict[str, str]:
+    """候选子进程环境：剥离宿主路径与评测相关变量。"""
+    env = os.environ.copy()
+    for key in list(env):
+        upper = key.upper()
+        if upper.startswith("FRONTIER") or upper.startswith("TELECOM_EVAL"):
+            del env[key]
+    return env
+
+
+def check_determinism(
+    python: str,
+    solver_path: Path | str,
+    inst_path: Path | str,
+    timeout: float,
+) -> tuple[bool, str]:
+    """同一实例跑两次，输出必须字节一致。"""
+    solver_path = Path(solver_path)
+    inst_path = Path(inst_path)
+    outputs: list[Any] = []
+    for _ in range(2):
+        try:
+            proc = subprocess.run(
+                [python, str(solver_path), str(inst_path)],
+                capture_output=True,
+                text=True,
+                encoding="utf-8",
+                errors="replace",
+                timeout=timeout,
+                cwd=str(solver_path.parent),
+                env=candidate_env(),
+            )
+        except subprocess.TimeoutExpired:
+            return False, "timeout during determinism check"
+        except Exception as exc:
+            return False, f"error during determinism check: {exc}"
+        if proc.returncode != 0:
+            return False, f"candidate exited with code {proc.returncode}"
+        try:
+            outputs.append(json.loads(proc.stdout))
+        except Exception as exc:
+            return False, f"cannot parse determinism output: {exc}"
+    if outputs[0] != outputs[1]:
+        return False, "candidate is not deterministic (output differs across two runs)"
+    return True, ""