Gate 4 — Severity decomposition: divergence audit

Generated 2026-05-22 11:47 UTC by `scripts/py/intelligence/severity_divergence_audit.py`. Re-run anytime; pure-Python, no network.

Gate-4 test: Spearman ρ between systematic-fund and corporate-strategy rankings of the proof-slice should be < 0.6 — and both orderings should be defensible by inspection of the underlying decomposed scores.

Population. 973 actions parsed from docs/iptm/actions/*.md. Of these, 16 carry explicit (directness, durability) values via the SEVERITY_OVERRIDES table in lib/iptm-severity.ts (the proof slice). The remaining actions score with v1 fallback (axes default to magnitude), so ranking divergence is dominated by tte/magnitude weighting and should not be interpreted as proof of the decomposition.

Proof-slice scores

Spearman ρ matrix — proof slice

Pairwise rank correlation between every viewer-profile pair on the proof slice. The gate's headline test is the systematic-fund × corporate-strategy cell — pass condition is ρ < 0.6.

Verdict: ρ(systematic-fund, corporate-strategy) = +0.208 < 0.6 — gate test PASSES on the proof slice. Defensibility still requires manual inspection of the top-ranked actions per profile (below).

Full-register top-25 per profile (v1 fallback)

_With v1 fallback, an action missing explicit directness/durability scores under magnitude on those axes — so for unaxis-mapped actions the ranking collapses toward the existing severity scalar. The proof-slice actions (marked by their presence in the override table) are the interesting departures from that baseline._

systematic-fund

corporate-strategy

commodity-trader

policy-analyst

compliance-officer