Written for someone who knows investing but doesn't speak quant. The philosophy first, then the real equations, then an honest account of what's proven and what isn't. Nothing here is a black box.
Every source is graded on its own track record: the average return of its past picks, minus what the market did over the same window. That's its realized edge (its "alpha").
Plain English: A YouTuber whose picks have beaten the market earns trust and a bigger vote. One whose picks have lagged gets ignored — or, if they're reliably wrong, faded. We don't assume anyone is smart; we make them prove it.
A stock becomes a pick when enough trusted, independent sources agree within a short window:
Plain English: One person liking a stock means nothing. But an analyst, a senator's trade, and a technical breakout all landing on the same stock in three weeks? That's three different kinds of informed money agreeing independently. That's what we act on.
Why "independent types" matters: we measured it — media sources (newsletters, YouTubers) copy each other and co-occur 12–15× more than chance, so three of them is fake agreement. Hard-data sources (insiders, Congress, analysts, price) are genuinely independent. We require the agreement to span independent types, not just three names.
We hold up to 20 positions at ~10% each. When more stocks qualify than we have slots, we fill by conviction priority — most independent types first, then strongest agreement — deterministically, so the book never depends on the random order signals arrived in.
Separately from convergence, we lean into the single most robust pattern in finance: stocks that have been winning tend to keep winning for 3–12 months.
Plain English & honest (corrected 2026-06-24): Momentum is a real, documented premium (since 1993). But our survivorship-free backtest of our long-only version was heavily inflated by data artifacts — once we winsorized the prices (capping glitch spikes), the book's 2010–26 return fell from +1,577% to +155%, which actually lagged buy-and-hold SPY (+781%). So: momentum the academic factor is real, but our particular implementation, cleanly measured, did not beat the market. Its weakness is volatility — it earns the premium in calm uptrends and gives it back in choppy/volatile markets. This is the kind of correction that only shows up when you clean the data and check.
From SEC 13F filings, we flag stocks where the number of professional funds holding them is rising quarter-over-quarter — smart money quietly building a position. Actionable 45 days after quarter-end, when filings become public (no peeking).
Plain English & honest: When the count of funds holding a stock jumps 10%+, that's institutions accumulating. On 25 years of survivorship-free data, this was the one signal that beat its factor-matched peers (+6% over 6 months) and held in every era. It's the most promising thing we've found — but it's a candidate, confirmed by one rigorous method, not two. See the tests →
Plain English: We never trim a winner because it's "gone up enough." We exit only on a real break (down 30%) or if six months pass without the thesis paying off.
Uninvested capital is never left in 0% cash. When we're not fully deployed, the money sits in the market itself (SPY/QQQ) until a signal redeploys it.
Plain English & honest: Cash sitting idle is a silent tax on returns. Keeping un-deployed money in the market is the single change that, in backtests, turned a strategy that lagged buy-and-hold into one that beat it. (We first learned this on the crypto side.)
We track the market's state — whether the S&P is above or below its 200-day average, and how volatile it is — and lean more contrarian when fearful, more momentum when calm. The contrarian board is a watch-list of stocks where a few very-high-trust sources are bullish against the crowd — often an early signal, before full convergence.
Adding to high-conviction names into market pullbacks was one of our first ideas. It exists only as a standalone experiment (`backtest-tactics.mjs`) and is not wired into the live picks right now. Given that the book draws down harder than the market on red days, this is the single most worthwhile thing to test next — see the honest note below.
| Claim | Status |
|---|---|
| Beats SPY/QQQ/Russell over 5–20 years, net of costs | ⚠️ Backtests said yes, but were inflated by price artifacts / survivorship; clean momentum actually lagged SPY |
| That outperformance is mostly factor harvesting (momentum + market), not secret alpha | ⚠️ True — FF5+MOM alpha +2.5%/yr, not significant |
| One genuine candidate edge beyond factors: institutional accumulation | ⚠️ Candidate (one method) |
| It's a "risk-reducer" that softens down days | ❌ Not shown — today it fell −4.9% vs SPY −1.5% |
What this is: a disciplined, transparent, multi-source strategy that harvests well-documented premia and acts on independent smart-money agreement, with one promising candidate edge and serious risk discipline. What it isn't: a black box, or a machine that prints secret alpha. The honest frontier right now is managing drawdowns — the book is concentrated in high-momentum names that fall hard in a crash, and taming that (regime sizing, dip-buying, hedging) is the live work.
Every morning the convergence engine runs over all sources, grades them, ranks by conviction, and fills the 20-slot book. That live board — plus the paper-traded account that follows these rules with a real P&L — is at /picks and /portfolio.