I have already obtained comprehensive distribution statistics for all numeric columns in the dataset. Let me present these findings in a clear, organized way. The dataset contains numeric columns across multiple tables. Here are the distribution summaries: ## Analysis Run LLM Usage **LLM Call Metrics:** - **calls**: min=2, max=7, mean=4.49, median=4, stddev=2.49 - **input_tokens**: min=0, max=6,566, mean=2,045, median=403, stddev=2,409 - **output_tokens**: min=0, max=10,958, mean=1,286, median=286, stddev=1,626 - **cost_usd**: min=$0, max=$0.14, mean=$0.024, median=$0.0018, stddev=$0.031 ## Analysis Runs **Resource Consumption:** - **cost_compute_seconds**: min=0.40, max=545, mean=42.6, median=44.8, stddev=53.6 - **total_cost_usd**: min=$0, max=$0.14, mean=$0.046, median=$0.059, stddev=$0.031 - **total_input_tokens**: min=0, max=6,967, mean=3,985, median=5,341, stddev=2,560 - **total_output_tokens**: min=0, max=11,119, mean=2,506, median=3,118, stddev=1,873 ## Batches **Processing Metrics:** - **elapsed_ms**: min=0, max=1,560,492, mean=59,620, median=1,293, stddev=192,026 (highly right-skewed) - **total_bytes**: min=0, max=2.7B, mean=113.5M, median=1,086, stddev=419.7M (extremely right-skewed) - **total_rows**: min=0, max=19.4M, mean=941,670, median=20, stddev=3.3M (extremely right-skewed) **Key Observations:** 1. **Token usage shows high variability**: The large standard deviations relative to means (especially for input/output tokens) indicate substantial variation in LLM interaction complexity across runs. 2. **Batch processing is highly skewed**: The median values for batches are orders of magnitude smaller than the means, indicating most batches are small but a few are extremely large (the median batch has 20 rows vs. mean of 941K rows). 3. **Cost distribution**: LLM costs are relatively modest (max $0.14) with most runs clustering at the lower end (median $0.0018 for individual usage, $0.059 for total runs). The complete statistics for all 50+ numeric columns are shown in the rendered table above, covering metrics across analysis runs, batches, classifications, catalogs, jobs, and other operational tables.