A mechanical momentum book, run in public.

Every rule is published. Every fill is logged. Two variants run head-to-head as a live A/B test so you can see which momentum definition actually holds up — not a backtest, a forward record starting 2026-08-09.

Get the toolkit + weekly signals — $9 Read the rules first

Already bought? Download your files here — instant, no login.

The two books

Variant A — risk-adjusted momentum

$101,207

mean(3m, 6m, 12m-1m) ÷ 20d realized vol · top 5 equal-weight

MU PANW SNOW UNH XBI

Variant B — classic 12-1 (control)

$99,650

12-month return skipping the last month · top 5 equal-weight

MU AMD CAT SMH PANW

Both books opened at $100,000 on 2026-08-09. Paper, marked to the last close. Variant B never changes — it's the control. Any rule change goes to A only, is logged with a dated hypothesis, and is judged on at least two weeks of relative equity before adoption.

The ruleset, in full

Universe: ~65 liquid US mega-caps plus sector and asset ETFs.
Absolute filter: price must be above its 200-day moving average to be eligible. No exceptions, no overrides.
Sizing: top 5 by score, equal weight.
Stop: close ≤ entry − 10% → sell at that close.
Rank exit: drops out of the top 10 (hold buffer), or loses 200dma eligibility → sell.
Cadence: rebalance and mark on every US trading day close. Freed cash re-enters at the next rebalance.

That's the whole thing. The stops and the 200dma filter are mechanical — the point of publishing them is that I can't quietly move them when a position goes against me.

This week's ranked universe

#	Ticker	Last	6m	Score A
1	MU	861.00	118.3%	2.55
2	SNOW	334.70	98.7%	2.32
3	PANW	385.04	141.7%	2.25
4	UNH	408.74	49.8%	1.76
5	XBI	158.03	26.0%	1.62
6	CRWD	225.16	127.7%	1.50
7	AMD	469.56	125.3%	1.39
8	LLY	1,231.94	16.8%	1.26
9	IWM	299.98	13.7%	1.20
10	MRK	130.92	8.9%	1.16

Top 10 of ~65 shown, generated 2026-08-11. The paid file has the full scored universe as CSV, plus the weekly top-5 with entry levels and stop levels already computed.

The screener code is free

The scoring logic isn't the secret — discipline is. Score any universe yourself:

import yfinance as yf, numpy as np

def score(tickers):
    px = yf.download(tickers, period="2y")["Close"]
    out = {}
    for t in tickers:
        s = px[t].dropna()
        if len(s) < 260 or s.iloc[-1] < s.rolling(200).mean().iloc[-1]:
            continue                              # 200dma absolute filter
        r3  = s.iloc[-1]/s.iloc[-63]  - 1
        r6  = s.iloc[-1]/s.iloc[-126] - 1
        r12 = s.iloc[-21]/s.iloc[-252] - 1        # 12-1: skip last month
        vol = max(np.std(np.diff(np.log(s[-21:])))*np.sqrt(252), 0.05)
        out[t] = ((r3 + r6 + r12)/3) / vol        # variant A score
    return sorted(out.items(), key=lambda kv: -kv[1])

Run that and you have variant A. What $9 buys is not the formula — it's the packaged one-file engine (screen / rebalance / mark / status with the paper book and stop tracking wired in), the full scored universe every week, and the signal file so you don't have to remember to run it.

What you get for $9

run.py — the full engine, one file, no dependencies beyond yfinance/pandas/numpy.
rules.md — the exact mechanical ruleset above.
latest_signals.csv — all ~65 names scored and ranked.
SIGNALS_<date>.md — the clean weekly top-5 plus watchlist.
One-time payment. Not a subscription. Cards accepted.

Get it — $9

After paying you land on a receipt page — your download is at smeltworks.com/momentumedge/get. Bookmark it; it is also printed on the checkout page.