AJ AI · Bets Desk · The honest answer · 07.12.2026

Can we make money following cappers? No.
Did we build something that can find what does? Yes.

6 months · 5,730 graded picks · 428 cappers · 30 strategies walk-forward tested · every number independently re-verified · replay-tested against real days

01 — The main resultEvery strategy died. All 30.

We tested every way you'd think to profit from capper picks — tail the career winners, fade the losers, chase hot streaks, fade cold ones, bet the consensus — walk-forward, so no strategy ever saw the future. Zero survived. Most lost money outright.

Strategy	Variants	Best result	Verdict
Tail proven winners	9	all negative	DEAD
Chase hot streaks (the 15-0 guys)	6	≈ zero, none significant	DEAD
Fade the losers	5	−5.5% ROI (vig eats the fade)	DEAD
Consensus (from old DBs)	4	untestable — 0.6% coverage	→ REPLAY
Bet everything	6	−1.2% ROI	DEAD

The one mechanical "survivor" (NBA subset, +14%) failed the fragility screen: 82% of its profit came from ONE month (May playoffs), the edge was gone by June, and it straddles the multiple-testing threshold. Flagged artifact by its own finder AND the independent verifier. Not tradeable.

Why everyone loses, in one line of arithmetic: the pool needs 52.62% to beat its own juice and hits 50.95%. Cappers aren't dumb — they're average, and average pays the book −2.35% forever.

02 — Your replay ideaWe rebuilt real days blind. The cards went 4–1.

Your idea, executed: pull what cappers actually posted on past days, build the card with zero knowledge of outcomes (results firewalled away from the card-builder), then grade against real scores.

Day	Card	Result	P&L
Jul 3	Fery +123 · Bublik +105	W W	+2.28u
Jul 5	J. Young U6.5 ast	prop ungradeable — excluded	0
Jul 7	Mets ML · Astros TT o3.5 · Red Sox ML	L W W	+0.82u
Jul 8	NO CARD — zero signal, honest skip	—	0
Jul 10	NO CARD — no picks in feed	—	0
Total	5 graded tickets	4W–1L	+3.10u

Honesty tax: 4–1 on five tickets happens ~1-in-6 by coin flip. This proves the MACHINE works end-to-end — not the money. The forward archive now records every day automatically, so this test grows daily until it can speak with real statistical weight.

03 — What actually showed signalWeak, real, and honestly sized

Conviction carries info: cappers' high-unit plays beat their sprinkles (+8.3% ROI, CI just clears zero). Already wired into the card weighting.
4 of 428 cappers are career-positive — but certifying a capper statistically needs ~1,488 graded picks and nobody has that. They inform weighting; nobody gets anointed.
The lie detector works: your Telegram leaderboard screenshots — our tape caught inflated staking bases (one row 7.6×), a "profitable" 37.7% win rate, and 15 "hot" names that are career-negative on our grades.

04 — The money pathSo how does this make money? Honest board:

LIVECLV tracking — closing lines now captured on every card night. Beating the close is the only edge metric your own research ever validated. Verdict in ~4 weeks of card nights.
LIVEReplay accumulation — every day auto-archives; the consensus-card test grows toward a real sample for free.
28-DAY GATEFade-the-inflated-touts — in-sample it prints; paper-tracked from today, no real money until the gate passes.
GATEDConviction filter — weak but real; rides inside the card weighting, tracked by CLV.
DEADStreak-chasing · blind tailing · blind fading · trusting any leaderboard anyone publishes, including ours.

The realistic frame: betting income from picks alone = not supported by the data. Discipline (band, sizing, no-chase) + CLV-verified line shopping = where a real but small edge could live. The system now measures exactly that, daily, for free.

05 — What runs without youThe machine, daily

8:00 — yesterday's picks auto-graded → careers/streaks updated
8:15 / 12:00 / 18:00 — X pull → board regenerates → deploys if changed
10:45 / 17:15 — headless digest builds the day's card (no human needed)
Card nights — one command: live scores + real line moves + Telegram pings + closing-line capture
Public board: aj-board.pages.dev · kill switch: BOARD_REFRESH_OFF

06 — Make it betterThe "go crazy" button

The upgrade script is staged. Paste one line into Claude any time:

ultracode — run UPGRADE-PROMPT.md from bets-data, full power

What it fires: replay scale-up to 30+ days · CLV verdict analysis · fade-gate scoring · new-source scouting (Discord/Reddit signal lanes) · board UX wave · fresh adversarial audit of everything since the last one. Details in UPGRADE-PROMPT.md.

Every number on this page traces to a file: BACKTEST-VERDICT · REPLAY-PILOT.md · INSIGHTS.md · CLAIMS-VS-TAPE.md · CAPPER-DOSSIERS.md · verified by independent recomputation + adversarial review.
Research and entertainment only — not financial advice. Consensus measures popularity, not value. Expect most nights to lose. 21+ · bet what you can lose · 1-800-GAMBLER.