Ava Chen

Let's forget what we have already invested in this idea, because executing Kolmogorov-Smirnov tests is not a decisive proof of statistical similarity.
Focusing on minor deviations without in-depth analysis of extreme risks can be dangerously misleading, as past financial crises have shown.
We should examine with fresh eyes whether these tests are sufficient for high-stakes areas, such as venture portfolio evaluation.
For example, deviations that seem small in lower ranks may hide significant asymmetries or thick distribution tails.

Let's forget what we invested in this idea: performing a Kolmogorov-Smirnov test is not proof in itself but a simple tool whose value depends on interpretation.
The alleged concretization is conditional on the rigor of analysis and the quality of underlying data.
If we do not look beyond raw results, such as ignoring distribution tails in risk assessments before 2008, we may miss crucial signals.
A technical test never replaces a thorough qualitative analysis, especially when initial hypotheses are fragile.
For example, a KS test on highly volatile market data may seem compliant, but if we do not contextualize extreme events, we risk serious judgment errors.

Forget what we've already invested in this methodology; data relevance is the real issue, not just the tool.
Simply executing Kolmogorov-Smirnov tests does not guarantee the validity of conclusions if the underlying assumptions are fragile or the thresholds poorly chosen.
An apparent statistical similarity can mask significant financial risks, especially if the venture portfolio multiples used are biased.
For example, a French bank ignoring the specifics of its credit portfolio by relying on generic benchmarks could face failures.
We need to evaluate whether we would re-analyze with fresh eyes facing such limitations.

I have an allergy to bright ideas claiming that gating networks oppose basic methods; this is a superficial distinction.
In reality, a gating network is just an optimization mechanism, a refinement to improve existing methods, not an alternative.
This does not fundamentally change the cash flow or how we evaluate investments, unless there is concrete evidence of a significant improvement in predictability, such as a 15% jump in identifying high-potential startups.
Without that, it is just added complexity to the model, with no proven benefit in the real world of pension funds.