Meme-coin lottery — forward testPAPER ONLY

Strategy under test: many small bets, paid for by the rare monster. A win rate as low as 1% is fine — at $2.25 a bet, one true 1000x ($2.1M market cap from a ~$2.1k launch) returns roughly $1,900 before depth caps and carries ~859 losers on its back. Stakes: pump $2, every other venue $5; every entry pays constant-product impact against its entry pool. Everything below is measured through that lens: keep the future monsters in the net, cut everything else. No real money. Built 2026-07-27 14:41 UTC.

Drift between our entry quote and the next quote — the realistic band a ~1-minute-latency fill lands in. Price impact at $2 size is negligible (~0.003% into a $60k pool); timing volatility is the whole story. We hold rather than snipe, so entry drift mostly shifts the base, not the thesis.

Each cell: ROI (net $) for that book's tranches on coins from that source. Anonymous pool births (the GeckoTerminal catcher) are listed by their birth AMM — fluxbeam, pumpswap, uniswap-*, orca — where 'pumpswap' means births the WS mint stream missed, distinct from the 'pump' venue.

Fluxbeam cut 2026-07-27 — honeypot. Jupiter round-trip test on live "deep" pools: buy $25 → sell back $0.00 (sell impact 99.99%; one-way token-2022 valves). Prices printed, exits did not exist — its "+29–84%" was unrealizable. The pilot retargeted to pumpswap-birth graduations, which round-trip at ~−0.7% (verified). Sellability is now checked pre-trade on every live buy.

MONSTER-HOLD, the DAY-1H TOP PICKS basket (each day trades only the sources whose 1H book was profitable the PREVIOUS UTC day — no history, no bet), and one tracking box per non-pump birth source (auto-enrolled at 20+ graduations, ≥5 gate-passing bets) running the sell-at-1h rule. Session policy: EU (06–14 UTC) trades every venue; US (14–22) trades robinhood rails only — the Solana sources lose in US hours (pumpswap −20%) while robinhood held up; ASIA (22–06) trades nowhere. Every excluded combination keeps being logged and graded nightly as a "benched cell", so a policy exclusion can be reversed on evidence. MONSTER-HOLD: $2 lottery tickets on every pump.fun graduation with ≥$5k of sane pool depth, never sold — open positions valued at the depth-capped last quote, drained pools and rugs at zero. The tail thesis in its purest form: monsters take 19–140 days, so this box is judged in weeks. The DAY exit research (timed sells, targets, per-source books) lives in exits.py and can rejoin the race any time. The grey sub-line under each 4h block splits net by mint source.

A filter earns its place by cutting bets faster than it cuts winners — "keep" is the share of the cohort's 218 big winners it retained; a filter that boosts EV but drops the monsters is a trap. Green EV clears the cost line ($2.25 at the $2 reference stake — race stakes differ per venue). ORACLE = needs future knowledge, ceiling only. ONE-TOKEN = result rides on a single coin.

2026-07-27: TIME-1h(EU+US) +6.4% n=186 | MONSTER -71.6% | winners 341 rugs 936 | Δ1h -9.1pp vs 2026-07-27

2026-07-27: TIME-1h(EU+US) +15.5% n=289 | MONSTER -71.6% | winners 341 rugs 935 | Δ1h -23.4pp vs 2026-07-27 | BENCHED-GREEN fluxbeam|US +21% n=30

2026-07-27: TIME-1h(EU+US) +38.9% n=261 | MONSTER -71.5% | winners 321 rugs 840

One-shot crawl of pump.fun's listing, survivor-biased by design (listed by CURRENT mcap, so collapsed winners are invisible). ATHs are pump-tracker prints with no realizability leg — wash-printable, and one confirmed $60M farm print is in the set. "First-launch" cannot see wallet rotation (URI-lineage lower bound: 2 of 365 rotated). The duration findings are robust to both: monsters grind for weeks, they do not spike.

Instead of one ticket at mint, tickets are added as a coin passes survival gates: graduation, then still-grinding a day later (≥$60k, within 50% of its high, live quote), then again at day 3. Winners grind for weeks — persistence is the signal. Gate tranches activate once coins age past each gate; each tranche settles at its own entry price.