VEDKUBER / TRANSPARENCY / RATINGS
VEDKUBER RATING SYSTEM
Every rating has its denominator in public.
A rating whose math is hidden is a marketing claim, not a rating. That is why everything here is public: 9 rateable items, percentile-band star mapping, VK-grades on credit-style cutoffs, and /100 cross-sectional scores. NR means insufficient sample — no stars are awarded from thin data. The bottom decile is also published. Every figure is backward-looking; past performance is not predictive.
RATEABLE ITEMS
9
models · symbols · signals · weekly audits · data sources · school modules · country cards · platform · brokersSCALES
★1–5 · /100 · VK-A…D
Bell-curve bands 10/22.5/35/22.5/10 · cross-sectional deciles · credit-style cutoffs 85/70/50REFRESH
Weekly + on-demand
Sunday 20:00 IST snapshot cron (rating_snapshots history) + live compute on viewMeasured ratings load karte hue… Source: live compute, GET /api/public/ratings (app.vedkuber.in)
Methodology — poora math, public
- Models — ★1–5: forward accuracy (voter_performance, 30-day window, ≥20 graded votes) is mapped onto a peer-set percentile bell-curve (top 10% → 5★, 22.5% → 4★, middle 35% → 3★, 22.5% → 2★, bottom 10% → 1★). The delta versus backtest cv_acc (meta.json) is reported alongside.
- Symbols — score /100 + decile: per-symbol realized hit-rate on graded prediction_logs (≥10 graded), cross-sectional rank; top AND bottom 10 published.
- Signals — ★1–5 + NR: per signal-type realized hit-rate, 30/90-day rolling windows, ≥30 graded needed for stars.
- Weekly audits, data sources, platform, brokers — VK-A…VK-D: cutoffs ≥85 / ≥70 / ≥50 / else; NOT_READY or >7-day-stale shows an "On Watch" chip (credit-rating style watch list) — no surprise downgrades.
- Country cards — score /100: 50% 30-day index coverage (world_prices) + 30% freshness (26h rule) + 20% daily-candle depth (world_candles). Incomplete data = lower score, never hidden.
- School modules — ★1–5 + NR: per-level quiz pass-rate percentile (≥10 attempts).
- NR is honest — insufficient sample shows NR with the exact count, never a star from thin data.
- Rating rationale on every action — each rating carries its inputs + source in the "why" line; history is kept in rating_snapshots (rank-change arrows come from there).
- Backward-looking disclaimer on every display — past performance is not predictive.
Machine-readable: /api/public/ratings (JSON, public) · Daily Public Report · Weekly Audit Report · Monthly Transparency Report · Track Record