🧭 The Project-10x Method: eight layers of discipline

The base rate of a 10x in 1–2 years is tiny. The only structure that has done it repeatedly is a high-purity, high-leverage name bought at a cycle bottom (our own audited sample: SanDisk, ~50x). So the system hunts cycle windows, not tickers. The honest premise comes first: any structure that can 10x can also drop 70%.

Ballast ≥80% (cash-flow compounders like NVDA/CEG — never used for 10x hunting) + a satellite pool ≤20%. Each satellite is sized so that a total wipe-out costs ≤3% of assets. The ballast exists so the rules can still be followed on the days the satellites are down 50%.

Purity × scarcity × small cap × cycle-bottom start × operating leverage — all five, or it is not a 10x structure. Then the expectations question: what is already priced in? Our 52-week positioning readout is the crude gauge; anything +600% off its low is presumed already-priced until proven otherwise. Low consensus first: the day sell-side names a stock, the purity premium is gone.

Our six-item top-zone checklist (leader making new highs after a multi-bagger run / record volume / retail slogans / sold-out narrative / mega-IPO / second-tier catch-up rally) has verified three times: SKHY→memory top, SPCX→space top, Innolight's HK IPO→optics top. Inverted, it is the bottom checklist: contract prices turning negative → capacity-cut announcements → PS≈1x with weekly price stabilization. Everything is a pre-registered trigger: it fires, you execute; it doesn't, you sit.

Full Kelly destroys accounts when probabilities are misjudged, so ≤1/4 Kelly, simplified into executable notches: high conviction = 1/3 of the satellite pool, medium = 1/5, low = watchlist only. Buys happen on panic days (−5%+), in 3 batches, ≥10% pullback or ≥4 weeks apart; never on limit-up days or inside pre-declared no-buy windows. Never average down a loser (Druckenmiller's and Livermore's shared rule): add only to positions that have proven themselves.

Exit rules are written before entry: at event tops (mega-IPO, listing week) distribute mechanically down to a core stub — sell to the people who just turned bullish. Sell-side upgrades during a run are logged as distribution counterparty signals, not confirmation.

Satellite pool down 30% → circuit breaker: all new buying stops and the framework itself gets reviewed (not averaged into). Every call enters the public ledger with date, original wording, falsification condition and review date; misses stay published with their lesson. When two plans conflict, the newer framework wins and the conflict itself is logged — a silently overridden rule is no rule.

Before a major call enters the ledger it must survive multiple bull-vs-bear rounds: each side attacks the other's thesis at full strength, and a referee assigns survival odds — anything under 50% is rejected or downgraded to watch-only. Open calls get re-reviewed the day material counter-evidence appears (e.g. price approaching a falsification line), with confidence cuts published. A system-level tripwire guards against correlated error: when several top calls share one macro premise, any two falsifications suspend the whole pattern for framework review instead of piecemeal apologies. The ledger's hit rate can only come from honest grading — deleting misses burns the entire asset.

Every call is logged before the outcome (public track record); trigger lines are machine-watched every 30 minutes; falsification conditions and review dates are registered together with the call. When the method is revised, the old rules and the conflict ruling stay on file — this page is part of the ledger too.