{
    "action": "final",
    "content": "Only two columns across the five monthly tables can be treated as numeric: **Founded** (a year) and **Amount** (stored as text like '$1,000,000' and parsed to USD). I pooled all 446 rows. The summary table from the last query (step 5) is rendered above.\n\n**Founded (year, 427 usable of 446 rows)**\n- Min 1871, max 2022\n- Mean ≈ 2015.3, median 2017\n- Spread: standard deviation ≈ 11.0 years; interquartile range 2014–2020\n- The mean is below the median, so the distribution is left-skewed. A few very old companies (the minimum is 1871) pull the mean down, while most startups were founded between 2014 and 2020.\n- The 19 unusable rows are missing values or misaligned entries. The May table is affected most, because its Founded column is VARCHAR and some of its rows are misaligned or blank.\n\n**Amount (USD, 385 usable of 446 rows)**\n- Min about $11.3, max $18.3 billion\n- Mean ≈ $145.6 million, median $9.4 million\n- Spread: standard deviation ≈ $991.5 million; interquartile range $1.3 million to $45 million\n- The distribution is extremely right-skewed. The mean is about 15 times the median, and the standard deviation is about 7 times the mean. A handful of very large deals drive the mean and spread. For typical funding, the median and quartiles are the better guide.\n- The $11.3 minimum and the $18.3 billion maximum look like parsing or data-entry artifacts, so they should be checked before relying on them. The 61 non-usable rows include 'Undisclosed' amounts (6 of them appear in the value-frequency query at step 0) plus blanks or misaligned rows.\n\n**Takeaway:** Founded year is fairly tightly clustered, with a long tail of older firms. Funding amount is heavy-tailed, so use the median ($9.4M) and the IQR rather than the mean and standard deviation."
}