MODEL LAB · CANDIDATE RECORD

Cross-sectional ranker

A candidate model that contributes evidence to the desk read. It ranks a liquid spot universe by expected 14-day return relative to that day's median. It is not a buy list.

Candidate model
Current record: rejected. Not a promotion.
Rejected gate reasons
net_spread_not_above_relative_strength
regime_edge_concentration
episode_edge_concentration
Data cutoff2026-07-23T00:00:00Z Generated2026-07-24T02:51:45Z Verdictrejected Live board statuscurrent Folds18

Current-state board

Ranks remain visible when an independent filter blocks eligibility. They are research context, not an instruction. Reliability uses prior LightGBM OOS hit rate and forward-return dispersion in the same rank decile and regime. It is not a probability or trade instruction.

SymbolRank percentilePrior OOS reliability bucketStateVeto / filterFeature evidence
ALLOUSDT+100.00%lower historical reliability (OOS hit 39%; dispersion 27.2%; n=2362)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
RIFUSDT+97.83%lower historical reliability (OOS hit 39%; dispersion 27.2%; n=2362)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
KAITOUSDT+95.65%lower historical reliability (OOS hit 39%; dispersion 27.2%; n=2362)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
ZECUSDT+93.48%lower historical reliability (OOS hit 39%; dispersion 27.2%; n=2362)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
TRXUSDT+91.30%lower historical reliability (OOS hit 39%; dispersion 27.2%; n=2362)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
BTCUSDT+89.13%lower historical reliability (OOS hit 40%; dispersion 19.3%; n=2442)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
WLDUSDT+86.96%lower historical reliability (OOS hit 40%; dispersion 19.3%; n=2442)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
BNBUSDT+84.78%lower historical reliability (OOS hit 40%; dispersion 19.3%; n=2442)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
NEARUSDT+82.61%lower historical reliability (OOS hit 40%; dispersion 19.3%; n=2442)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
XLMUSDT+80.43%lower historical reliability (OOS hit 40%; dispersion 19.3%; n=2442)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
XRPUSDT+78.26%lower historical reliability (OOS hit 37%; dispersion 18.1%; n=2038)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
FETUSDT+76.09%lower historical reliability (OOS hit 37%; dispersion 18.1%; n=2038)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
DOTUSDT+73.91%lower historical reliability (OOS hit 37%; dispersion 18.1%; n=2038)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
SOLUSDT+71.74%lower historical reliability (OOS hit 37%; dispersion 18.1%; n=2038)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
APTUSDT+69.57%lower historical reliability (OOS hit 37%; dispersion 18.1%; n=2063)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank
KITEUSDT+67.39%lower historical reliability (OOS hit 37%; dispersion 18.1%; n=2063)research onlyregime is not positiverealized_vol_30d
breadth
ret_90d_rank

Method

Closed UTC daily candles from liquid Binance spot USDT pairs. LightGBM is compared with XGBoost and the Relative Strength baseline on identical time-ordered folds. The target is each coin's 14-day return less the eligible-universe median.

Independent controls

RegimeRegime at or below zero keeps rows research-only.
ToxicityTop-decile Toxicity keeps a row vetoed regardless of rank.
toxicity_top_decile
LiquidityTrailing 30-day median quote volume below USD 1m makes a row ineligible.

Out-of-sample passport

Same rows, same chronological folds, fixed 40 bps round-trip cost on each rebalance. Net figures are historical measurements, not forecasts.

ModelMedian ICTop-minus-bottom spreadTop-decile netMax drawdownRebalances
LightGBM0.053+1.40%-1.26%-88.02%50
XGBoost0.040+0.74%-0.34%-82.94%50
Relative Strength0.003+1.33%-0.41%-79.94%50
FoldTest periodMedian ICTop-decile net
12024-08-22 — 2025-02-170.047+3.38%
22024-09-21 — 2025-03-190.041-0.27%
32024-10-21 — 2025-04-180.076-0.93%
42024-11-20 — 2025-05-180.084-1.05%
52024-12-20 — 2025-06-170.099-4.54%
62025-01-19 — 2025-07-170.062-0.96%
72025-02-18 — 2025-08-160.069+1.58%
82025-03-20 — 2025-09-150.090-0.04%
92025-04-19 — 2025-10-150.096+0.20%
102025-05-19 — 2025-11-140.123-2.43%
112025-06-18 — 2025-12-140.084-0.37%
122025-07-18 — 2026-01-130.071-2.40%
132025-08-17 — 2026-02-120.074-5.25%
142025-09-16 — 2026-03-140.056-2.49%
152025-10-16 — 2026-04-130.068-2.01%
162025-11-15 — 2026-05-130.030-3.20%
172025-12-15 — 2026-06-120.047-3.21%
182026-01-14 — 2026-07-120.022-2.56%

Cost: 40 bps round-trip per rebalance.

Limitations

What would change our mind

A failed promotion gate, regime concentration, data revision, schema mismatch or live degradation sends this record back for review or rejection. Candidate is not a promotion to Indicators or Daily Read.

Research-only record. No execution path on this page.

x.com/toptraders0x · toptraders0x.t.me · home·terms·privacy·risk·compliance