Alphalumina is a measurement system before it is a signal feed. Every token that enters the pipeline leaves with a decision and the reasons for it — passed, rejected, or watched — and every decision is graded later against what actually happened.
New pump.fun launches, market microstructure, wallet flows, and social attention stream in continuously — roughly seven hundred scored candidates a day. Nothing is pre-filtered by vibes: the junk enters the pipeline too, because the junk is the control group.
Momentum, liquidity quality, wallet intelligence (smart wallets and KOL presence, weighted by measured reliability — not follower counts), safety/structure, and attention convergence. Positive evidence is offset by explicit penalties: clone risk, dirty flow, bundled launches, source quality, crowding.
A score alone earns nothing. Execution gates check trusted-source quality, momentum reality (already dumping? already pumped?), liquidity depth, repeat-churn patterns, and token age. Most candidates die here — by design. When we audited the gates against outcomes, tokens they blocked were three times less likely to be tradeable than tokens that passed.
Qualified signals enter paper positions with a real exit state machine — entries, stops, takes, all simulated against live market data with no real funds. Then the grading engine scores what happened: did it move, was the move reachable, did it round-trip, would a stop have survived. The grade is permanent and public to members — including when the grade is "we were wrong."
Every token lives in exactly one lane, and each lane has hard limits on what it may trigger. Experimental logic can never leak into the live feed: shadow lanes write to their own ledgers and must earn promotion with forward evidence.
Fresh launches with clean structure, scored and gated as above. The proven core. This is the only lane allowed to produce alerts and paper positions.
profile: clean_early_discovery v0.8
Tokens vetoed for clone risk or dirty flow that nonetheless show real liquidity, smart wallets, and persistent attention. Measured silently — because our data shows these fires are usually traps at the moment they look most exciting. May trigger nothing.
profile: dirty_but_real_shadow v0.3
When a flagged token's liquidity dies and later genuinely recovers, that recovery is a new episode — tracked as its own event. Early evidence says the revival, not the first fire, is where the real moves live. Accruing forward data before it may trigger anything.
profile: revival_watch v0.1
Tokens the attention layer loves and the scorer vetoes — the bucket where hype tools shine their spotlight. We publish these with their outcomes, because "what looked hot but got vetoed" is where most portfolios die.
graded against the same rules as everything else
Everything correctly discarded — the vast majority. Roughly 85% of new tokens die. This lane is the system working, and it is never deleted: the noise stream is the control group for every claim we publish.
the denominator no one else shows you
External discovery tools enter through one door: a base-rate tracker that logs their full flag stream — winners and deaths — and grades it against our own outcome data. A screenshot of one winner is survivorship, not evidence.
tradeability gate: v0.2 · applied to every lane and every external flag
The rules that decide what counts as tradeable — and what never did.
Read the Methodology See Real Receipts