{
  "mtp_observation": "A true delta requires a matched MTP-off run.",
  "limitation": "Without a controlled baseline run under identical conditions but with MTP disabled, observed performance changes cannot be reliably attributed to the MTP configuration alone, as they may stem from environmental variance, prompt ordering, or stochastic model behavior.",
  "metric_needed": "Baseline performance score from a matched MTP-off run to compute the absolute or relative delta."
}