On its first plot, this map exposed a dead pillar: the Earnings Quality half of the composite (40%) was inert for 444 of 501 names — quality_score pinned at 50 because the 5-year KPI bundle was never built, so every quality component was null. The screen had effectively been ranking on valuation + momentum alone. Now fixed: the bulk Börsdata KPI fetch (ROIC 5y, margins, growth, incremental ROIC) is wired in, and quality is differentiated for 913 of 923 names. The remaining 10 lack Börsdata 5-yr coverage and sit on the mid-line.

The point of plotting the data: a 17-column table hid a defect in 40% of the score; the map surfaced it in one glance, and it was fixed the same day. The quality axis below is now real.

Every name positioned on quality (Y) against cheapness (X). Among the 913 names with a real quality read, 262 sit in the top-right prize quadrant (quality & cheap) and 175 are bottom-right value traps. The faint hollow dots on the mid-line are the 10 names whose quality isn't computed yet. Outlined dots are deep-dived; red-ringed labelled dots are the conviction portfolio.

Two scores you already compute (quality_score, valuation_score), finally plotted. A value investor reads this in one glance: hunt the top-right, interrogate anything red-ringed that has drifted left, and avoid the bottom-right however cheap it looks.

The composite weights and the conviction probabilities are priors until proven. This logs every rating and every call, then measures them against realized forward returns. It has logged 47946 rating rows across 54 dates and 571 conviction calls. History begins 2026-05-31.

No backtest is shown because an honest one needs elapsed time, not look-ahead-fitted current fundamentals at past prices. The clock above is the truth: the first 3-month information coefficient lands 2026-08-31.