Each Period is a distinct, independently-traded scenario — so every tile is one company in one period, a self-contained six-factor regression over the 2021–2026 window. Metadata (name, sector, industry, market cap) is reference only.
—
Agents · canonical universe
10
Independent periods
6
Factors controlled
—
Evaluation window
—
No agents match these filters.
Selection Frontier
The universe mean is honestly ~zero. Alpha is a selection phenomenon — drag the conviction slider and watch it concentrate. Ranking: mean alpha t-statistic across periods, in-sample. Evaluate the selection forward on your own names.
—
Basket mean α · bps/period
—
vs universe mean
α term structure · by period
Aggregate factor loadings
Sector composition · top 6
All figures in-sample over 2021-06-30 – 2026-07-31. Basket = equal-weight top-N by mean α t-stat on the selected strategy's primary leg.
Agent Atlas
Every agent, mapped by its six-factor fingerprint (60 dimensions → 2). No agent was told its sector — the map recovers it anyway: nearest-neighbor same-sector rate 41.8% vs a 12.3% class-frequency baseline. Sectors emerge as continents; alpha shows as heat. Paste your book to see where it sits.
Embedding: t-SNE of standardized per-period factor loadings (Naive-10 long leg; Pure Signal fallback). Positions are structural, not predictive. In-sample. Nearest-neighbor same-sector rate = share of agents whose closest neighbor in the 2-D embedding shares its sector (n = 2,476 with a sector); baseline = Σ p² over sector shares = 12.3%.