$ trace-eval run examples/hermes_bad.jsonl

============================================================
  TRACE-EVAL SCORECARD  Total: 32.4/100
============================================================

LIKELY ROOT CAUSES:
  - Use canonical retrieval entrypoint
  - Stop accessing deprecated files
  - Context pressure exceeded 90% — reduce prompt size

DIMENSION SCORES:
  reliability             5.0  (high)
  efficiency             77.4  (medium)
  retrieval               0.0  (high)
  tool_discipline        80.0  (high)
  context                32.0  (high)

FRICTION FLAGS (sorted by severity):
  [CRITICAL] retrieval_no_entrypoint
    -> Use canonical retrieval entrypoint
  [CRITICAL] retrieval_deprecated_file @event 9
    -> Stop accessing deprecated files
  [CRITICAL] context_pressure_critical
    -> Context pressure exceeded 90% — reduce prompt size
  [HIGH] retrieval_fallback_search
    -> Avoid fallback search -- use primary retrieval
  [HIGH] tool_timeout @event 5
    -> 1 tool call(s) timed out
  [MEDIUM] reliability_errors @event 3
    -> Review 3 error(s) at event indices [3, 4, 8]
  [MEDIUM] context_compression
    -> Context compression triggered 1 time(s)

ADAPTER CAPABILITY REPORT:
  has_span_ids: False
  has_parent_spans: False
  has_event_latency: True
  has_token_data: True
  has_tool_calls: True
  has_cost_data: True
  has_retrieval_fields: True
