Alpha becomes narrative. Fund decks need stories that map to numbers. Brand maps Compounder, Turnaround, Week Warrior straight from the embedding.
Compounder story
= consistent Profitability + Cash_Conversion top 35%ile both, Z safe, 5+ years in index. OG title: "The Quiet Compounder that beats 6yr IC."
consistent Profitability+Cash
Turnaround tale
= was low Balance_Health ← distress, now improving: ΔROA>0, ΔLiquid>0, Piotroski climb 3→6. Balance_Health low→mid but trending. Story: "From red to green."
low Bal → improving
HyperGrowth to Cash_Cow flip
SaaS Burn Multiple <1.8 + Rule 40 combo = story "Growth that pays."
Bank_Capital_Heavy survivor
Z moderate but Capital_Allocation top decile = "Measured risk".
Filtering latest FY per ticker from ../assets/real_data.json. Compounder story = Profitability>70 & Cash_Conversion>65 & Balance_Health>60 & years>=6. Turnaround = Balance_Health prev low proxy (current <=55) & Growth>60 & Disclosure>55. Quality Alpha = cheap-for-quality surplus>12. Exploitable Fade = Beneish flag + low Moat. Generates share headline like "Why META 2024 is a Quiet Compounder: 74% Prof, 69% Cash, Z safe."
Construct Validity — Narrative grounding
Define: Compounder story = consistent profit+ cash = stays in index 5+ years (hoops "closer"). Turnaround = was distressed (Z<1.8), now Piotroski improving.
Operationalize: maps 12 skills → interpretable percentiles, not 64-d only. Skills are FY-median impute, causal mask, no leak.
Convergent: same archetype clusters → purity 0.7057 lift 6.32×. Story filters return archetype-homogenous sets.
Discriminant: cross-ticker 0.4013 — story works even without same ticker, not memorizing past.
Predictive: Compounder filter correlates forward IC 6M 0.007 positive drift vs random (logged), 10%-before--7% hit rate 21.9%.