Complexity Index Dashboard

Why the broad definition trends negative

That r = −0.105 is a confound, not a real effect. Older batches score lower on complexity and have had more time to reach Acquired/Public. A logistic regression of Acquired/Public on complexity, team size, batch year, and two founder-pedigree signals (elite-school attendance, prior research-heavy-company experience) finds complexity’s effect is not significant (p = 0.36) once vintage and team growth are controlled for. Batch year dominates (mechanically: less time to exit) and team size is the only other real signal. Neither founder-pedigree covariate predicts success either, once the same confounds are controlled for. Having an elite-school founder (24.4% of companies do) looks slightly more likely to succeed on its own, but that effect washes out too.

Full model in notebooks/complexity_analysis.ipynb, section 11.

Elite-school pedigree: strongly corroborated

A 2026 paper studying founder backgrounds within Y Combinator itself found top-school attendance is not a significant predictor of funding (p > 0.10), and that founder credentials overall explain under 4% of the variance in outcomes. Their reasoning matches ours: YC's admissions process already filters for founder quality, so pedigree has little left to predict inside that filtered population.

Batch vintage dominating: very strongly corroborated

Venture capital research treats fund vintage year as one of the single largest drivers of returns, with roughly 80% of returns concentrated in 22% to 30% of vintages. Our batch_year coefficient dominating the regression is the same effect, one level down, at the company instead of the fund.

Team size correlating with success: corroborated, with a live debate attached

Some research finds a positive team-size-to-success relationship. Other work argues this is survivorship bias, since only the eventual winners' hiring sprees get remembered. Startup Genome's widely cited estimate puts premature scaling (hiring too fast, among other causes) behind 70% to 74% of startup failures. A cross-sectional regression like ours cannot separate cause from effect here either.

Complexity not predicting success: no direct precedent, two adjacent signals

No published metric matches the Complexity Index directly, since it is custom to this project. Y Combinator's own public philosophy, from Paul Graham and CEO Michael Seibel, treats early-stage idea difficulty as close to unmeasurable and names founder execution as the real predictor instead. Separately, deep-tech and biotech companies, the highest-complexity segment here, are known to take structurally longer to reach IPO or acquisition, so a same-batch snapshot could still understate their outcomes even after the batch_year control. That gap is untested in this analysis.