@@ -18,7 +18,7 @@ DESCRIPTION >
1818 without security data averaged 85.0 vs 80.8 for measured peers, and 2,742 repos were rated
1919 "excellent" with zero security assessment. Imputing at the population median instead removes
2020 that reward: missing weight earns the typical rate, not the repo's inflated rate. Population
21- medians are computed from the same *_ds tables via population_priors NODE (quantileExact (0.5)
21+ medians are computed from the same *_ds tables via population_priors NODE (quantileExactOrNull (0.5)
2222 over measured-only rows). Each category datasource already carries NULL when its own coverage
2323 fell below 40% (Layer 1, done per-category pipe). If fewer than 2 of the 3 categories are
2424 available, healthScoreV2 becomes NULL — a single-category score cannot anchor a reliable median
@@ -37,23 +37,23 @@ SQL >
3737 SELECT
3838 coalesce(
3939 (
40- SELECT quantileExact (0.5)(maintainerHealthScoreV2 / 40.0)
40+ SELECT quantileExactOrNull (0.5)(maintainerHealthScoreV2 / 40.0)
4141 FROM health_score_v2_maintainer_ds
4242 WHERE maintainerHealthScoreV2 IS NOT NULL
4343 ),
4444 0.0
4545 ) AS priorMaintainerRate,
4646 coalesce(
4747 (
48- SELECT quantileExact (0.5)(securitySupplyChainScoreV2 / 35.0)
48+ SELECT quantileExactOrNull (0.5)(securitySupplyChainScoreV2 / 35.0)
4949 FROM health_score_v2_security_ds
5050 WHERE securitySupplyChainScoreV2 IS NOT NULL
5151 ),
5252 0.0
5353 ) AS priorSecurityRate,
5454 coalesce(
5555 (
56- SELECT quantileExact (0.5)(developmentActivityScoreV2 / 25.0)
56+ SELECT quantileExactOrNull (0.5)(developmentActivityScoreV2 / 25.0)
5757 FROM health_score_v2_development_ds
5858 WHERE developmentActivityScoreV2 IS NOT NULL
5959 ),
0 commit comments