**The strongest relationships are between the categorical columns, but the top scores are inflated by high cardinality.** The dataset has only one numeric measure, Amount in USD, and it is stored as text, so there is no classic numeric correlation to compute. The pairwise ranking I ran uses normalized mutual information (NMI, 0 to 1) between columns. The scores come from the step-1 result; amount was bucketed on a log10 scale. **Strongest pairs** - **SubVertical and year:** NMI 0.974. - **Industry Vertical and Remarks:** NMI 0.888. - **InvestmentnType and SubVertical:** NMI 0.769. - **City and SubVertical:** NMI 0.694. - **Amount bucket and SubVertical:** NMI 0.655. - **Industry Vertical and year:** NMI 0.582. - **Industry Vertical and SubVertical:** NMI 0.525, which is expected since one nests inside the other. - **City and Industry Vertical:** NMI 0.425. - **Industry Vertical and InvestmentnType:** NMI 0.419. **Weakest pairs** - Amount bucket and year: NMI 0.036. - Amount bucket and city: NMI 0.070. - Amount bucket and remarks: NMI 0.112. - City and remarks: NMI 0.173. - InvestmentnType and remarks: NMI 0.176. **How to read this** - **Cardinality inflates the top scores.** SubVertical is about 82% unique and Industry Vertical has roughly 900 distinct values, so almost any column looks "predictable" from them. The near-perfect SubVertical and year score (0.974) and the Industry Vertical and Remarks score (0.888) are mostly artifacts of this, not real dependence. - **The better-grounded links are among the lower-cardinality columns.** Funding round type, city, industry vertical and amount all show moderate association with each other, for example Industry Vertical with InvestmentnType (0.419) and with Amount bucket (0.299). Round type goes with the amount raised (0.267), and sector goes with round type. - **Funding amount is largely independent of time and place.** Its association with year (0.036) and city (0.070) is negligible. Deal size varies mostly by sector and round type, not by when or where the deal happened. - **The table has no pre-aggregated gold tables, and one table has no joins.** The relationships are therefore between columns of `raw.startup_funding` only. The next step would be to repeat the analysis after grouping industry verticals and sub-verticals into broader categories. That would give a cleaner picture of sector, round type, city and deal size.