Track Record

Every call the look-alike finder would have made on IWM, QQQ, SPY since 2022-01-01, replayed at 10:30 ET and 12:00 ET with the corpus cut to what existed before each date — a 2022 call only ever saw pre-2022 history, so hindsight is structurally impossible. The finder’s outcome stats are descriptive (“what past look-alikes did — not a forecast”); this page scores them anyway, because a public record either builds trust or teaches us something. Whatever the number is, it ships.

Checkpoints

6,754

2022 → 2026

Leaned

22%

pctUp ≥ 60 or ≤ 40; the rest abstain and are counted

Hit rate — clean

49%

270 leans without a loose/generic cap

Hit rate — capped

49%

calls where the finder itself flagged a weak read

The verdict so far

The leans carry no direction edge beyond drift: always leaning up over this period scores 52.9%, and the finder’s leans score 49%. That matches the tail-agreement rig’s finding — the engine’s measurable signal is in shape (how the rest of the day traded), not in sign (which way it closed) — and it is exactly why every outcome panel on this site says “what past look-alikes did — not a forecast.” This page exists to keep that claim checkable in public.

By slice

The honest split to watch: clean reads vs the ones the finder capped as loose or generic. If the caps mean anything, the two columns should diverge. The per-grade rows are the grade calibration: what each letter actually delivered, measured on calls the finder never saw coming.

slicecheckpointsleanedhit rate
all675422%49% (708/1453)
clean (uncapped)121422%49% (132/270)
capped (loose/generic)554021%49% (576/1183)
at 10:30 ET337721%48% (344/720)
at 12:00 ET337722%50% (364/733)
IWM225421%46% (212/464)
QQQ224623%48% (247/511)
SPY225421%52% (249/478)
grade A78122%46% (80/173)
grade B267822%47% (276/589)
grade C212121%50% (220/436)
grade D103222%52% (116/223)
grade F14223%50% (16/32)
2022149820%46% (139/303)
2023150021%50% (156/313)
2024151223%47% (167/355)
2025150022%52% (172/331)
202674420%49% (74/151)

Recent calls

The newest checkpoints, exactly as the replay scored them. Every row links to the same query in the time machine so the call can be audited.

datecutbest twingradeleanactualresult
2026-07-0112:00 ETINTU 2021-09-03B-89 bpabstain
2026-07-0112:00 ETJNJ 2023-04-11cappedC-60 bpabstain
2026-07-0112:00 ETV 2020-02-07A-41 bpabstain
2026-07-0110:30 ETCVX 2023-04-24cappedBup (60%)-73 bpmiss
2026-07-0110:30 ETORCL 2024-08-16cappedB-70 bpabstain
2026-07-0110:30 ETUSDJPY 2017-01-11cappedB-15 bpabstain
2026-06-3012:00 ETV 2021-10-19cappedBup (62%)+32 bphit
2026-06-3012:00 ETDIA 2020-09-30cappedB+21 bpabstain
2026-06-3012:00 ETEEM 2021-10-19cappedB+7 bpabstain
2026-06-3010:30 ETSPY 2026-06-26cappedB+14 bpabstain
2026-06-3010:30 ETSPY 2021-05-13cappedB+38 bpabstain
2026-06-3010:30 ETEURJPY 2023-07-28cappedB+25 bpabstain
2026-06-2912:00 ETJNJ 2022-05-10cappedC+84 bpabstain
2026-06-2912:00 ETFXI 2022-01-06cappedB+70 bpabstain
Method: at each checkpoint the finder ran with beforeDate set to the session’s own date, so its pool was exactly what a trader would have had that morning. A “lean” is the outcome sample the product displayed (≥60% of twins closed up → leaned up; ≤40% → down; otherwise abstain). “Actual” is checkpoint close → session close. Capped calls are the ones the finder itself labeled loose or generic — scored separately because the caps are the product’s own confidence statement. The ledger appends incrementally; new sessions land as they close.