Model health
The model, with its hood open.
These are the live production weights and signal diagnostics — not a marketing snapshot. Weights are re-tuned by the feedback controller from observed hit rates, and every change is logged below with its rationale.
Current factor weights (baseline defaults)
momentum
24%Short-horizon adoption velocity (stars, downloads, visits, hiring deltas)
growth
21%Sustained structural trajectory via log-linear trend regression
quality
18%Signal consistency and noise discipline across the stack
attention
12%Narrative velocity — HN, social, launch-platform attention
risk
10%Volatility, spike-revert patterns, macro regime drag (subtracted)
provenance
15%Verified primary-source footprint (filings, patents, grants)
Signal predictiveness (Information Coefficient) · latest run 2026-06-13
| Signal | IC | Lead time | Sample |
|---|---|---|---|
| vc_round_size_stepup | 0.374 | 548d | 12 |
| vc_investor_count | 0.114 | 548d | 19 |
| vc_cumulative_raised_usd | 0.100 | 548d | 19 |
| vc_round_amount_usd | 0.052 | 548d | 19 |
| vc_post_money_valuation_usd | -0.115 | 548d | 15 |
| vc_months_since_last_round | -0.129 | 548d | 12 |
Weight-change log
Every adjustment the feedback controller makes, with its reasoning. No silent re-tuning.
2026-06-07scoreauto
Score component rebalance: momentum=0.55 rarity=0.25 consistency=0.20
2026-06-07signalautoF1 0.00 → 0.00
IC-optimized: top signal github_stars_pct_30d (IC=0.000)