{
  "version": "recent_performance_v1",
  "purpose": "Monthly recent-performance watchlists; not a live order list. Compare12m and6m; do not select a window from its best backtest PF.",
  "training": "Only the latest12 or6 completed monthly SQ dates strictly before the target SQ and no later than its previous-session cutoff. Require all12/6 valid SQ observations and20-session liquidity coverage, median turnover>=500millionJPY. Train on GD>=0.5% observations only, minimum3. No outside-window observations, including no older winning stocks retained as fallback.",
  "recency": "Later half of the SQ calendar window has weight2, earlier half weight1 (latest6of12 or3of6). Weight is based on calendar SQ age, not the last3/6 GD trades. Weights fixed before evaluation, not optimized.",
  "score": "weighted_mean_net_pct * effective_n/(effective_n+10); effective_n=(sum(weights))^2/sum(weights^2). Positive scores only; top100 maximum, no padding. Rank on score rounded6decimals, tie code ascending. Equal-weight sensitivity uses the same rule with all weights1.",
  "sample": "Minimum3 is a data floor, not evidence of reliability. Show raw GD count, effective count, win rate, PF and best-trade-removed mean; flag n<6 and best-trade-dependent rows. Do not claim3wins proves an edge.",
  "test": "Chronological reconstruction on the saved2023-01..last completed SQ cache, not prospective OOS. Each list tested on its next SQ only. Same actual Open to Close, equal notional, commission0, adverse0.05% each side. Report GD0(no filter)/0.5/1/1.5/2, annual and major/minor results, date-equal mean, best3dates-removed PF, tail risk. Not11:30 exit, no TP/SL/CFD gate or auction-fill guarantee.",
  "controls": "Same dated bars for frozen annual performance100, movement12m/6m, recent eligible pools and equal-weight recent rankings. Annual100 has its original eligibility (no added liquidity gate); recent eligible pools require3GD but no positive-score gate. Different selection rules are labeled, never silently equated.",
  "limitations": "Current-listed/collected1605-stock survivorship bias remains. Price response reflects recent markets but does not identify AI/semiconductor themes or causal news. Current industry is descriptive only. Daily OHLC rounding uncertainty gets the existing1JPY conservative sensitivity. Major/minor subdivisions contain very few dates.",
  "current": "The current list targets the SQ named in the source manifest, using its asof date. Do not rename it to a later SQ without new completed data. Current is not evaluated. Old sources and annual membership stay unchanged."
}