Unkerien/PLL-Sensor-Cyber-Guard

1name: PLL-Sensor-Cyber-Guard-v12version: "1.0.0"3description: >4  Meta OpenEnv environment simulating a Phase-Locked Loop (PLL) sensor5  under cyber-attack.  Agents must detect and classify adversarial signal6  injections (step attacks, frequency ramps, stealthy FDI) from noisy7  PLL telemetry in real-time.8 9protocol: openenv-2026-sync10action_format: wrapped          # {"action": {"action_id": int, ...}}11port: 786012 13sync_metadata:14  requires_player_id: true15  requires_session_id: true16 17api:18  reset: POST /reset19  step:  POST /step20  state: GET  /state21 22observation_space:23  type: Box24  description: >25    Continuous PLL signals: phase_error (rad), freq_deviation (Hz),26    vco_voltage (V), plus a rolling buffer of the last 10 phase-error27    readings for trend analysis.28  fields:29    - name: phase_error30      dtype: float6431      low: -10.032      high: 10.033    - name: freq_deviation34      dtype: float6435      low: -5.036      high: 5.037    - name: vco_voltage38      dtype: float6439      low: -5.040      high: 5.041    - name: buffer42      dtype: float6443      shape: [10]44      low: -10.045      high: 10.046 47action_space:48  type: Discrete49  description: >50    Classification labels indicating the detected attack type.51  labels:52    - none53    - step_attack54    - freq_ramp55    - stealthy_fdi56 57tasks:58  - id: task_159    difficulty: easy60    description: >61      Step Attack — detect a sudden +2.0 rad spike in phase error62      injected at a random step in the episode.63    grader: >64      Score 1.0 if correct type detected within 5 steps of injection.65      Latency penalty β=0.3 for delayed detection.66      -0.2 per false positive before attack onset.67 68  - id: task_269    difficulty: medium70    description: >71      Frequency Ramp — detect a gradual +0.1 Hz/sec frequency offset72      that starts at a random step.73    grader: >74      Score 1.0 if correct type detected.  0.5 if detected but75      misclassified.  Latency penalty β=0.3.76 77  - id: task_378    difficulty: hard79    description: >80      Stealthy False Data Injection — detect a subtle bias (≤ 1σ of81      normal noise) causing long-term phase drift without obvious spikes.82    grader: >83      Score 1.0 if correct type detected.  Requires trend analysis84      over the 10-step buffer.  High latency tolerance but strict85      false-positive penalty.86 87reward:88  formula: "R = (A × Success) - (β × Latency) - (F × FalsePositive)"89  success_weight: 1.090  latency_beta: 0.391  false_positive_penalty: 0.292 93