groups:
  - name: dsh-prometheus
    rules:
      - alert: DshMetricsTargetDown
        expr: up{job="deepseek-harness"} == 0
        for: 2m
        labels:
          severity: critical
        annotations:
          summary: DeepSeek Harness metrics target is down
          description: Prometheus has failed to scrape the DSH target for two minutes.

      - alert: DshAgentTurnFailureRatioHigh
        expr: |
          (
            sum(rate(dsh_agent_turns_total{status=~"error|blocked|aborted"}[10m]))
            /
            clamp_min(sum(rate(dsh_agent_turns_total[10m])), 0.000001)
          ) > 0.10
          and sum(increase(dsh_agent_turns_total[10m])) >= 10
        for: 10m
        labels:
          severity: warning
        annotations:
          summary: DSH agent turn failure ratio is high
          description: More than 10% of turns failed over a meaningful ten-minute sample.

      - alert: DshLlmFailureRatioHigh
        expr: |
          (
            sum(rate(dsh_llm_requests_total{status=~"error|aborted|thrown|incomplete"}[10m]))
            /
            clamp_min(sum(rate(dsh_llm_requests_total[10m])), 0.000001)
          ) > 0.10
          and sum(increase(dsh_llm_requests_total[10m])) >= 10
        for: 10m
        labels:
          severity: warning
        annotations:
          summary: DSH LLM failure ratio is high
          description: More than 10% of observed LLM attempts failed or ended incompletely.

      - alert: DshAgentTurnP95Slow
        expr: |
          histogram_quantile(
            0.95,
            sum by (le) (rate(dsh_agent_turn_duration_seconds_bucket[15m]))
          ) > 120
        for: 15m
        labels:
          severity: warning
        annotations:
          summary: DSH P95 turn duration exceeds two minutes
          description: The fifteen-minute P95 agent turn duration is above 120 seconds.

      - alert: DshJobFailuresDetected
        expr: sum(increase(dsh_jobs_failed_total[15m])) > 0
        labels:
          severity: warning
        annotations:
          summary: DSH background jobs failed
          description: At least one background job failed in the last fifteen minutes.

      - alert: DshMetricLabelOverflow
        expr: sum(increase(dsh_metrics_label_overflow_total[15m])) > 0
        labels:
          severity: info
        annotations:
          summary: DSH metric labels were collapsed
          description: One or more dynamic label values were mapped to __other__; review naming or cardinality limits.