Alphy

A momentum rotation in S&P 500 and Nasdaq-100 stocks with accelerating revenue, plus a gated 3x Nasdaq-100 ETF sleeve, run in a US margin account. Paper trading on the current rules (from 2026-10-31). Backtest, risks and a daily trade log.

1. What it does

• Universe: stocks that are S&P 500 or Nasdaq-100 members on that date. One share class per company.
• Trend filter: the close is above its long-term moving average, and that average is rising.
• Growth filter: year-over-year revenue growth is higher than in the prior quarter.
• Balance-sheet filters: each name is tested for cash and for financial distress against other recent filers, using only filings available at the time.
• Ranking: a medium-term over long-term moving-average ratio, highest first.
• Picks: a small fixed number of names per sleeve. A held name stays while it passes the filters and stays near the top of the ranking. Empty slots go to the highest-ranked names not held.
• Daily check: on every session between rebalances, a held name that no longer passes the filters is sold at the next open. A held name that has fallen in the ranking is sold only when the best name not held is ahead of it by a wide statistical margin. When a sale happens, the sleeve refills and re-weights that day; otherwise it does not trade.
• Admission: on every session, after the rebalance or daily check, a name that has reached the top of the ranking and is larger than every name a sleeve holds can replace that sleeve's weakest held name at the next open, so a new leader can enter without waiting for a held name to drop out.
• Pre-release boost: on every session, when the stock the strategy holds most of (by market value, over all sleeves) is in the run-up to its estimated next earnings release, the stock book is held at a larger size from the next open; it goes back to normal at the open of the estimated release day. Each sleeve keeps its picks and weights and only resizes. The release date is estimated from the company's earnings-release filings (SEC Form 8-K, item 2.02). With no estimate there is no boost.
• Weighting: by company market value raised to a high power, then scaled to a target level of portfolio volatility, with the scale capped. In practice most of each sleeve goes to its largest company (median largest-pick weight 99.9% at the 64 January rebalances, 2011-2026). Market cap (daily, point-in-time, all share classes of the company) is used only for weighting, never for selection.
• Rebalancing: several sleeves of equal size. Each sleeve re-ranks (re-selects its picks) on its own staggered schedule, and re-weights more often: on the re-weight days between two re-ranks it keeps its picks, and its weights, volatility scaling and quality strip are re-applied. The re-weight days are the same for all sleeves. The daily check and admission still run every session. The backtest trades at the next open.
• Leverage: the stock book and the ETF sleeve together stay within a fixed gross limit at a rebalance, more while the pre-release boost is on, where the broker's initial-margin limit scales the extra purchases down. Positions drift between rebalances.
• TQQQ sleeve: a fixed share of equity while a volatility gate is on, resized monthly. The gate turns the sleeve off when QQQ volatility rises past a set level. It turns back on after volatility falls below a lower level, or sooner once volatility has fallen far enough from its peak of the off period (faster re-entry). A faster re-entry is undone if volatility climbs back toward that peak.
• Quality strip: after the sleeves are formed, a hard strip removes names that fail a balance-sheet quality screen. A second, softer screen can cut weight on names that fail it; a name the book already holds is immune to that soft cut. The soft screen combines balance-sheet and accrual quality with a year-over-year fall in gross margin. From 2026-10-31 it no longer includes the inventory-versus-revenue screen (see the changelog).
• Delisting: a held stock whose series ends is cashed at its last close if it was acquired. Any other cause counts as a total loss.

Exact parameter values are not published.

Each trading day

One run before the open:

• About 9:00 AM ET: update the data to the prior close and check it; compute the TQQQ gate, which sleeves rebalance today, the daily check of every held name, the admission check of every sleeve and the pre-release boost (the largest holding and its estimated release date); read the account; build the day's order list before the 9:30 open. The run happens every session, because the daily check can sell on any day. If the data are stale or invalid, the order list is not ready by 9:30, or any check raises an alert, there is no trading that day.
• 9:30 to 10:00 AM ET: buys and sells are worked during the first 30 minutes to beat the opening price, within fixed price and size limits. Sells go first. Anything still open at 10:00 is completed then.
• About 10:30 AM ET: fills are compared with the official open and the trade log is updated.
• Stock sleeves: each sleeve re-ranks on a staggered schedule; all sleeves re-weight together on the days between re-ranks. Every sleeve is also checked on every other session (the daily check above), so eligibility and scores for every held name are computed every day; a check that fires sells at that day's open.
• TQQQ sleeve: while the gate is on, resized to its target share of equity once a month. The gate is checked every session.
• A missed day does not change the rules; the next run works from the account as it stands.

2. Why it works

Momentum and growth. Stocks in strong trends tend to keep going for months. Requiring accelerating revenue drops trends that the business does not back. The cash and distress filters drop fragile balance sheets. Weights by company market value, raised to a high power, put each sleeve almost entirely in its largest leader.

The post-2008 regime. The backtest starts on 2010-02-11, TQQQ's first trading day, and covers only this regime. Rules and policies in effect during it:

• Short-sale circuit breaker (Reg SHO Rule 201, the alternative uptick rule). After a stock falls 10% in a day, short sales must be above the national best bid for the rest of that day and the next. Adopted 2010-02-24; compliance from 2011-02-28. SEC release, compliance date.
• Single-stock pauses after the 2010-05-06 Flash Crash: a 5-minute pause after a 10% move in 5 minutes, approved 2010-06-10. Limit Up-Limit Down bands replaced them from 2013-04-08. SEC 2010, SEC 2012, SEC bulletin.
• Market-wide circuit breakers at 7%, 13% and 20% declines in the S&P 500, approved 2012-05-31 and in effect from 2013-04-08. NYSE, FINRA filing.
• Zero rates and a Fed backstop: the Fed cut the federal funds target range to 0-0.25% on 2008-12-16 and kept it there until 2015-12-16. It cut to 0-0.25% again on 2020-03-15 and kept it there until 2022-03-16. On 2020-03-23 it added open-ended asset purchases and corporate credit facilities. Fed 2008-12-16, Fed 2015-12-16, Fed 2020-03-15, Fed 2020-03-23, Fed 2022-03-16.

A daily-reset 3x ETF loses most in long, choppy declines, where volatility decay compounds. The trading rules above are designed to slow disorderly selling in single stocks and the whole market. The gate takes the sleeve out when QQQ's volatility rises, and the faster re-entry puts it back once volatility has clearly come off its peak instead of waiting for it to fall all the way back.

Limits: these rules do not stop bear markets. In 2022 the Nasdaq-100 fell about 33% and TQQQ about 79%. Regimes can change.

3. Results

Current rules (paper, from 2026-10-31): fixed-January walk-forward 231.31% a year on the research basis (mean over 20 start-day shifts of the schedule 223.89%), CAGR 214.0%, maximum drawdown -59.8% (-57.2% on the headline margin model), on data through 2026-10-28; the median of its rolling four-year CAGRs is 166.2% and a fresh account from 2022 compounds at 338.5% a year. Every rule change, and what it added or cost, is on the Components page. The first row of the table below is the current rules. Research basis, 5 bps per side. The CAGR and MaxDD columns of the first row run through 2026-10-28. Sharpe, the four-year windows, and the buy-and-hold rows are 2010-02-11 to 2026-10-02, the window in which holding QQQ gives a Sharpe of 0.90. Strategy rules were selected using these data, so the results carry selection bias. The 151 four-year windows start on the first session of each month from 2010-03 to 2022-09. They overlap. A cell written as median / 10th percentile / worst is that distribution. Calmar is CAGR divided by the depth of the worst drawdown. Sortino uses monthly returns and counts only the down months. Sharpe is daily return above the fed funds rate. Two columns give time underwater in each window: the share of sessions spent below the window's running peak, and the longest single spell below it, in months; both are written as median / worst. The trend line holds QQQ while its close is above its 200-day moving average and holds cash otherwise, at no trading cost, like the buy-and-hold rows. The added rows (the trend line, VGT, Bitcoin, NVIDIA and Broadcom) are computed from public price histories on the same conventions, and they reprint the rows above before they are added. Bitcoin's daily series begins in September 2014, so its figures cover 2014-09 to 2026-10-02 and its windows are the 97 starting from October 2014. NVIDIA and Broadcom are the two best-performing stocks of the window, with total returns of about 598x and 296x.

On the current rules, in all 151 windows the strategy's CAGR and its Calmar are higher than buy-and-hold QQQ and higher than buy-and-hold TQQQ. Buy-and-hold TQQQ beats QQQ on CAGR in 89% of the windows and on Calmar in 21%. The trend rule on QQQ finishes with a lower Sharpe than holding QQQ outright (0.83 against 0.90) and a longer median underwater spell (16.5 months against 7.8), so the one-line overlay does not reproduce the strategy's drawdown control.

Current rules. In-sample backtest, 2010-02-11 to 2026-10-28. The charts and tables in this part come from the same engine run as the headline figures above. Returns include 5 bps per side trading costs, borrowing costs and the margin checks below. Strategy rules and the evaluation period were selected using these data, so results carry selection bias.

Two margin models. Headline: a concentration-stress margin model (maintenance requirements as measured with a broker's own margin calculator, plus a stress requirement on concentrated positions). Second: Robinhood-style standard margin, the research basis every rule change is tested on. Both check maintenance at every close and at the daily lows.

A margin breach is a close or daily low below maintenance; the backtest sells pro rata at the next open to restore maintenance and continues. Rolling windows: 151 four-year windows of the full-run path, one starting on the first session of each month from 2010-03 to 2022-09. They overlap. The mean over the same windows is 215.0% (222.8% Robinhood-style). Starting a fresh account at each window start instead gives a median of 166.2% and a worst window of +86.5% (Robinhood-style margin).

Daily closing values as a multiple of starting capital (start = 1×), headline model, current rules. ETF benchmarks show buy-and-hold total returns.

Drawdown: how far each daily close sits below its previous peak, headline model, current rules.

Calendar-year returns (headline model, current rules):

2022 onward. From 2022-01-03 to 2026-10-28 the full-run path compounds at 316.3% a year (Robinhood-style margin: 342.8%). A fresh account started on 2022-01-03 compounds at 338.5% a year on Robinhood-style margin (311.7% on the headline model); a fresh account starts without the positions and the gains the full run carries into 2022, so the two differ. The rules were chosen on the full history, including these years, so this is not a clean out-of-sample test.

Starting account: the backtest starts from a small notional account. Borrowing spreads step down as the loan grows, so a small account pays the higher spreads for longer. On the current rules, the same rules started from a ten times larger account return 214.4% a year instead of 214.0% on the research basis and 205.2% instead of 204.8% on the headline model, about 0.4 points a year more; the maximum drawdown is unchanged (-59.8% and -57.2%). Every figure on this page uses the small starting account.

Known weaknesses of the daily check (current rules against the same rules without the daily check and the rules that rely on it, Robinhood-style margin). Without it the full run compounds at 123.5% a year instead of 214.0%, but 4 of the 17 calendar years are worse with it: 2010 (+81.6% against +105.8%), 2014 (+126.8% against +130.8%), 2015 (+55.8% against +101.9%) and 2026 to 2026-10-28 (+381.3% against +496.9%). The walk-forward rests on two years: without 2020 the fixed walk-forward falls from 231.3% to 187.6%, and without 2024 to 202.9%. Delaying every trade by one session lowers the CAGR to 187.2%, about 27 points, and the walk-forward to 198.3%: the check has to run every session, on the prior close.

Admission rule: limits. On the current rules it adds 45.3 points a year over the full run (214.0% against 168.7% without it, Robinhood-style margin) and limits the maximum drawdown (-59.8% against -79.7%). It fires rarely: on 21 dates from 2010 to 2024 (81 sleeve admissions; on 19 of the 21 dates all four sleeves at once), and its gain is concentrated: calendar 2020 alone adds 508 points to that year's return, 2018 adds 119, 2023 111 and 2013 71; leaving out any one year still leaves 24.8 to 50.6 points a year (24.8 without 2018). Nearby settings for the admission rank and the market-cap margin were tested before adoption; the adopted rank sits next to a cliff, where a tighter or a looser rank gives back most of the gain, while the market-cap margin is flat. It can buy a late leader: on 2024-12-24 all four sleeves moved into TSLA, which cost 7.5 points of return in the last week of 2024, and calendar 2024 is 94 points lower than without the rule. Its gain over the rules without it has a Sharpe ratio of 0.58 against a luck bar of 0.86 at the 2,579 variants counted now, a deflated Sharpe ratio of 0.11, so it does not pass the deflated test; it rests on the pre-registered test, the walk-forward, the held-out halves and the fresh starts.

Pre-release boost: limits. On the current rules it adds 45.8 points a year over the full run (214.0% against 168.2% without it, Robinhood-style margin). It was on for 54.8% of sessions, over 243 boost windows. Its gain is concentrated: the five largest windows carry 27.6% of it (MU before its 2026-06-24 release, MRNA before 2021-08-04, PLTR before 2025-05-05, NVDA before 2020-05-14, NFLX before 2025-04-17), and calendar 2020 alone adds 509 points to that year's return (2024: 201 points); leaving out any one year still leaves 37.8 to 50.2 points a year. It is a timed increase in leverage, and the margin limit lets only part of it through: on boosted days the stock book is a median 24.8% larger (mean 25.6%). Book size here is the stock positions' value divided by equity at each boosted session's close, compared session by session with the same run without the boost; only one of the 2,301 boosted sessions has a book below half of equity without the boost, and leaving it out changes neither figure. Moving the same boost 20 or 40 sessions before each release gives 178.8% and 176.4% (measured with the boost dates fixed in advance, where the true timing gives 206.8%), so most of the gain comes from the timing (60 sessions before, about one quarter earlier and close to the previous release, gives 193.7%). It raises concentration: the largest single position reaches 2.68x equity (MU, 2026-03-30) against 1.87x without it. Under the stress margin model it adds forced sales (368 against 30 in the full run; 193.6% against 164.9% a year). It depends on a daily feed of earnings-release filings: if the feed is missing or stale, the day's signal is not traded. The release dates are checked against a second source, the SEC's own filing record, for all 62 companies that were the strategy's largest stock holding on any session of the full run, from 2010-02-11 to 2026-10-16 (where the copy of the SEC record used ends): of 3,685 release dates, 3,684 match an SEC filing within one day (3,629 on the same day), the SEC record has no release the dates miss, and one date (MU, 2017-09-26) has no SEC release filing within three days. The way the next release date is estimated matters little: the two other estimates tested give 214.0% and 210.5%; the second has one margin breach under Robinhood-style margin in the full run and at least one in 47 of the 151 fresh four-year starts (none with the estimate used).

Sensitivity tests. Current rules, each change applied separately, research basis (Robinhood-style margin), full run 2010-02-11 to 2026-10-28. Base: CAGR 214.0%, maximum drawdown -59.8%. An excluded stock is removed from the universe for the whole run.

How it was tested

• Period: 2010-02-11 (TQQQ's first trading day) to 2026-10-28 only, so every result comes from the rate and market-structure regime described above.
• Universe: survivorship-bias-free, point-in-time S&P 500 and Nasdaq-100 membership, including later-delisted names. Prices are adjusted for splits and dividends.
• Filings: quarterly figures as originally reported, used only after the filing was public.
• Delistings: four held names were delisted, all acquisitions, cashed at their last close. Any other kind of delisting counts as a total loss.
• Costs and margin: 5 bps per side, margin interest, and maintenance checked at every close and at the daily lows under two broker margin models. Current rules: no breach in the full run and none in any of the 151 fresh four-year starts under Robinhood-style margin; 27 breaches in the full run under the headline model. The same rules without the pre-release boost have 5 breaches under the headline model and without the admission rule 31; without the daily check they have one breach under Robinhood-style margin in the full run and at least one in 47 of the 151 fresh four-year starts.
• Train, then check: the rules were chosen on the full history, so 2022 onward is not a clean test. On 2010-2021 the current rules compound at 172.8% a year, and a fresh account from 2022-01-03 at 338.5%. The same rules without the daily check give 91.8% and 221.8%, without the admission rule 129.4% and 291.7%, and without the pre-release boost 140.4% and 247.3%.
• Fresh starts: 151 four-year windows, each a new account, starting every month from 2010-03 to 2022-09. Current rules (Robinhood-style margin): median 166.2% a year, 10th percentile 109.7%, worst +86.5%. Without the pre-release boost: median 141.6%, 10th percentile 93.0%, worst +74.1%; without the admission rule 127.8%, 75.3% and +47.6%; without the daily check 103.2%, 43.9% and +7.2%. The current rules beat the rules without the daily check in 145 of the 151 starts.
• Walk-forward: re-choosing the rules each January from 2014 to 2025, using only returns from 2010-02-11 to that January, among the current rules and the same rules without the pre-release boost, without the admission rule and without the daily check, picks the current rules every year: 231.3% a year, the same as the current rules held fixed. Held fixed, the rules without the pre-release boost give 182.1%, without the admission rule 174.5% and without the daily check 112.7%; the current rules beat the rules without the daily check in 10 of the 12 years. These are the parts the rules were built from, so this is a check of the parts, not of every alternative tried. The rank buffer, the stock volatility target, the weighting power and the faster TQQQ re-entry were chosen on the full history.
• TQQQ gate: in four-year folds with a fresh account in each (current rules, Robinhood-style margin, data from 2010-02-11 only), the gated sleeve compounded at 131.7% a year in 2011-2014, 106.9% in 2015-2018, 337.3% in 2019-2022 and 396.3% in 2023 to 2026-10-28, against 124.1%, 110.3%, 319.6% and 382.2% with no sleeve and 131.7%, 103.3%, 329.9% and 399.2% with the sleeve always on. In 2015-2018 the sleeve cost return: 3.4 points with the gate and 7.0 always on. In each fold the stock book is the same and only the sleeve changes. Chosen on returns from 2010-02-11 to each fold's start, the gate was picked in 2 of the 4 folds and the always-on sleeve in the other 2; dropping the sleeve was never picked. Choosing this way gives 216.5% a year over the four folds, against 217.9% for the gate held fixed and 211.1% for no sleeve. No fold has a margin breach.
• Held-out halves: the current rules beat the same rules without the daily check in 10 of 10 splits of the years into two halves, with and without a 63-session gap between the halves (smallest gain 17.8 points). The pre-release boost and the admission rule each pass the same test (below).
• Pre-release boost: pre-registered before any run, with its pass conditions, nearby settings, placebo timings and a sign check fixed in advance. On the current rules the walk-forward is 231.3% with it against 182.1% without it; across 20 starts of the stock schedule (0 to 19 sessions after 2010-02-11) it adds +45.2 to +50.6 points (mean +48.0), positive in all 20. It beats the rules without it in all 151 fresh four-year starts (median gain 29.7 points, smallest 12.3), and in 10 of 10 held-out halves, with and without a 63-session gap between the halves (smallest gain 18.2 and 16.4 points). On 2010-2021 it gives 172.8% against 140.4%, and on a fresh account from 2022-01-03 338.5% against 247.3%. Nearby settings: a smaller boost gives a walk-forward of 223.4% and a larger one 232.5%; shorter windows give 221.0% and 229.0%, a longer one 237.0%; cutting instead of boosting over the same windows gives 183.1%, about the same as no boost. Its gain over the rules without it has a Sharpe ratio of 1.75 against a luck bar of 0.86 at the 2,579 variants counted now, a deflated Sharpe ratio above 0.99, so it passes the deflated test.
• Admission rule: pre-registered before any run, with its nearby settings fixed in advance. On the current rules the walk-forward is 231.3% with it against 174.5% without it; across 20 starts of the stock schedule (0 to 19 sessions after 2010-02-11) it adds +50.2 to +64.2 points (mean +56.9), positive in all 20. It beats the rules without it in 141 of the 151 fresh four-year starts (median gain 54.5 points, worst -5.2), and in 10 of 10 held-out halves, with and without a 63-session gap between the halves (smallest gain 7.6 points both ways). On 2010-2021 it gives 172.8% against 129.4%, and on a fresh account from 2022-01-03 338.5% against 291.7%. Nearby settings: a tighter rank limit gives a walk-forward of 177.2% and looser ones 199.5% and 167.2%; a smaller or larger market-cap margin gives 231.3% and 231.9%.
• Nearby settings: on the current rules (214.0% a year, walk-forward 231.3%), one step down and one step up for each of the replacement margin, the TQQQ gate threshold, the stock volatility target and the weighting power gives 209.1% to 215.4% (walk-forward 225.0% to 233.3%). The rank buffer is the least smooth: both of its nearby values give less, 204.6% and 201.9% (walk-forward 220.4% and 213.8%). With no replacement margin at all (any held name that has fallen in the ranking rotates out) the CAGR is 179.0%. Shifting every stock sleeve's schedule by 1 to 4 sessions gives 203.5% to 210.4% (walk-forward 216.5% to 225.9%). None of these has a breach under Robinhood-style margin, and the maximum drawdown stays between -59.8% and -60.5%.
• Faster re-entry: it fired in four of the gate's off periods (2011, 2020, 2022, 2025; 34 trading days in all), each time ending in the normal re-entry. Its gain rests on those four episodes and is small: on the current rules, nearby levels give 213.6% and 213.2% a year against 214.0%, and no faster re-entry gives 212.5% (walk-forward 230.2% against 231.3%), with the early re-rank when the gate turns back on kept.
• Multiple testing: 2,579 variants have been tested in the research rounds. Each part of the current rules was measured against the same rules without it, over the full run on Robinhood-style margin, with a bootstrap 90% interval for the gain (resampling blocks of about 20 sessions). Without the daily check, and the rules that rely on it (admission, the separate re-rank schedule and an early re-rank after a market rebound or when the TQQQ gate turns back on), the CAGR is 123.5% instead of 214.0%; the gain's 90% interval is +42.9 to +145.2 points. Without the admission rule: 168.7%, interval +13.3 to +83.5 points. Without the pre-release boost: 168.2%, interval +25.4 to +75.6 points. Deflated for the 2,579 variants, the daily check's Sharpe ratio test gives 0.25 and the admission rule's 0.11, so neither passes the deflated test; they rest on the walk-forward, the held-out halves and the nearby settings. The pre-release boost's gives above 0.99 and passes.
• Luck benchmark: with 2,579 variants tested, the best improvement over the rules it replaces that luck alone would be expected to produce has a Sharpe ratio of about 0.86 (1.63 using the spread of every stored variant), and this bar rises with every variant added. On the current rules, the daily check's improvement over the same rules without it has a Sharpe ratio of 0.70 and the admission rule's 0.58, both below the bar; the pre-release boost's is 1.75, above both bars. For whole strategies (return above cash) the luck bar is 0.43 to 0.86; Alphy's Sharpe ratio is 1.84 on the current rules, and holding QQQ gives 0.90 and SPY 0.79 over the same years (TQQQ buy and hold is 0.87).
• Concentration: without its five largest contributors (NVDA, PLTR, MU, TSLA, MRNA) the CAGR is 94.3% and the maximum drawdown -82.9% (current rules, research basis). The same five names supply 53% of the stock book's positive contribution (each day's weight times the next day's return, summed by name).

What was tried and decided against

Every candidate this program has decided against is listed here in full.

Candidates decided against in this research program. Each delta is in percentage points (pp) against the base named in its row, on the window named in its row. Deltas in different rows were measured against different bases and do not rank against each other.

Outcomes: a candidate can pass the bar and still not be adopted, be rejected on its measurements, or be stopped by a diagnostic before a backtest could bind.

Bases

Each delta was measured against the base named in its row:

• Oct 11 rules: the live rules from 2026-10-11 to 2026-10-13.
• Oct 13 starting rules: the rules adopted on 2026-10-13, before event rotation (replaced the same day).
• event-rotation finalist: the Oct 13 starting rules plus daily event rotation, before the admission rule (a research base on 2026-10-13).
• Oct 13 rules: event rotation plus the leader-emergence admission, live 2026-10-13 to 2026-10-14.
• Oct 14 rules: the Oct 13 rules plus the pre-release boost on the top name, from 2026-10-14; the variant with the smaller leveraged-ETF sleeve in the pre-release window followed the same day.
• Oct 15 rules: the Oct 14 rules after the 2026-10-15 changes (volatility target, release-session hold, release estimate without the fallback rule), live 2026-10-15 to 2026-10-16.
• Oct 16 rules: the Oct 15 rules after the 2026-10-16 changes (post-release cut added, carry cap removed), live 2026-10-16 to 2026-10-22.
• Oct 14-22 rules: the rules in use from 2026-10-14 to 2026-10-22 (the Oct 14, Oct 15 and Oct 16 rules above); rows that name only this range were measured on one of them.
• Oct 22 rules: the Oct 16 rules plus the dip rule on the top name with a smaller leveraged-ETF sleeve, live 2026-10-22 to 2026-10-28 (replaced by the quality strip on 2026-10-28).
• early-October combined base: a combined variant that served briefly as the research base in October 2026, before 2026-10-13.
• retired base: an earlier research base, retired before the October 2026 bases; those rows ran on a window ending 2026-09-18.
• published-factor test base: a one-off variant of the rules (a regime-gated, capped weighting) used as the base for the published-factor batch.
• matched 5-sleeve all-momentum control: a five-sleeve all-momentum portfolio built as the control for the event-sleeve study.

Passed the bar, not adopted (5)

Rejected (195)

Stopped by a diagnostic (14)

4. Risks

• Drawdowns: -57% in the backtest of the current rules (2021-01-26 to 2021-03-08, headline model; -60% with Robinhood-style margin). Expect deeper ones.
• Leverage: up to 2x at rebalances, more after drift. A gap down past the margin buffer forces sales at the open.
• Concentration: one stock can exceed 100% of equity (under the current rules MU reached 268% of equity on 2026-03-30). Each sleeve holds almost only its largest pick, and the daily check refills every sleeve from the same ranking: under the current rules the four sleeves hold the same names on 71% of sessions and the same largest pick on 93%. A single-name crash hits hard.
• TQQQ: daily reset. A choppy decline while the gate is on can lose most of the sleeve. The faster re-entry buys back while volatility is still elevated; a second leg down after it would hit the sleeve.
• Regime: the backtest covers one rate and market-structure regime (2010+). It may not repeat.
• Overfitting: the rules were selected on the same data. Live returns are likely lower.
• Daily check: the between-rebalance sales are decided on the prior close and filled at the open, so a stale feed or a missed daily run degrades the rule (section 3).
• Pre-release boost: it holds a larger stock book into the largest holding's earnings release window; a release that comes earlier than estimated, or a sharp move before it, hits the larger book. It needs a daily feed of earnings-release filings; if that feed is missing or stale the day is not traded.
• Quality strip: the hard strip can empty a sleeve slot when a held name fails the screen; the soft strip does not force held names out, so a soft-flagged holding can stay until another rule exits it.
• Delisting: a non-acquisition delisting is a total loss of that position.

5. Trade log

Daily trade log: every backtest trading day (date, ticker, buy or sell, weight before and after as % of equity, reason) and the live log, updated after each daily run. Weights only; no prices, share counts or account amounts.

6. Changelog

• 2026-10-31: the charts, the two-margin-model table, the calendar-year returns and the 2022-onward figures are redrawn for the current rules, from the same engine run as the headline figures; a drawdown chart is added. The tables for retired rule sets are removed from this section.
• 2026-10-31: the results table's current-rules row is now measured in every column, with the same code and the same 151 four-year windows as the earlier rows: four-year CAGR median 167.0%, 10th percentile 110.3%, worst +86.8%; Sharpe 1.84.
• 2026-10-31: strategy change. The paper strategy changes. The inventory-versus-revenue screen is removed from the soft quality strip; the hard strip, the gross-margin screen and every other rule are unchanged. The screen's measured gain was not trusted: it compared each company with its sector's median, and where that median was negative the comparison ran backwards, and most of its gain came from one stock flagged in March 2020 through that reversal. Fixed-January walk-forward 231.31% on the research basis vs 247.87% for the prior control on the same data (data through 2026-10-28); CAGR 214.0% vs 225.2%; maximum drawdown -59.8% on the research basis and -57.2% on the headline model, both unchanged to the first decimal. Over 20 start-day shifts of the schedule the walk-forward is 15.8 points lower on average. Paper results recorded before 2026-10-31 were made with the screen on.
• 2026-10-29: paper strategy change. The post-release cut is removed: since 2026-10-16 the stock book had been held at a smaller size for a set number of sessions after the largest holding's own earnings release whenever the stock reacted worse than the Nasdaq-100 over the release. Nothing else changes. Fixed-January walk-forward 247.87% on the research basis vs 243.61% for the prior control on the same data (data through 2026-10-28); CAGR 225.2% vs 222.3%. The maximum drawdown is deeper: -59.8% vs -55.4% on the research basis and -57.2% vs -55.4% on the headline model. Over 20 start-day shifts of the schedule the mean CAGR is 218.2% vs 216.1% and the mean maximum drawdown -61.9% vs -55.5%. Exact cut values are not published.
• 2026-10-29: paper strategy change: slower re-rank schedule. Each stock sleeve now re-ranks less often than before; on the re-weight days in between it keeps its picks and only resets weights, volatility scaling and the quality strip. All sleeves re-weight on the same days. Fixed-January walk-forward 243.61% on the research basis vs 221.73% for the prior control on the same data (data through 2026-10-28); maximum drawdown -55.4% vs -55.3%. Over 60 start-day shifts of the schedule it beat the prior control at every one; at 3 of them the maximum drawdown was deeper than the accepted limit (up to 0.74 points deeper than the prior control at the same shift). Exact schedule values are not published.
• 2026-10-28: paper strategy change: two more soft screens. It replaces the soft screen of the first quality strip (it is not added on top): the soft screen now also flags inventory growing faster than revenue relative to the sector, and a year-over-year fall in gross margin. The hard strip is unchanged. Fixed-January walk-forward 223.13% on the research basis (prior control 211.01%). Sector labels are current, not point-in-time. Exact thresholds are not published.
• 2026-10-28: paper strategy change: quality strip added. After the stock sleeves form, a hard balance-sheet strip removes failing names; a held-immune soft balance-sheet and accrual strip can cut weight on other failing names without forcing a held name out. Fixed-January walk-forward 211.01% on the research basis (prior control 171.61%). Exact strip thresholds are not published.
• 2026-10-24: the current-rules table gains five comparator rows (QQQ with a 200-day trend rule, VGT, Bitcoin from September 2014, NVIDIA and Broadcom) and two time-underwater columns (share of sessions below the running peak, and the longest spell in months; median / worst per four-year window).
• 2026-10-24: the results table shows the current rules next to buy-and-hold TQQQ, QQQ and SPY. Research-basis CAGR 173.0% through 2026-10-16, MaxDD -55.3%, Sharpe 1.71 through 2026-10-02. The charts are still the previously published path.
• 2026-10-14: pre-release boost adopted. When the stock the strategy holds most of is in the run-up to its estimated next earnings release, the stock book is held at a larger size, from the next open until the open of the estimated release day. Backtest CAGR 143.7% -> 150.2%, MaxDD -57.6% -> -57.6%, margin breaches 2 -> 2 (the same dates), fresh four-year starts with a breach 60 -> 60 of 151 (Robinhood-style margin 136.8% -> 140.2%, breaches 1 -> 1, 47 -> 47 of 151). Research simulator: 138.0% -> 141.5%, walk-forward 128.7% -> 132.6%. Its limits are listed in section 3.
• 2026-10-14: the backtest now starts from a smaller notional account, so the higher borrowing spreads of a small loan apply for longer. The rules are unchanged; only the backtest's financing moves. Backtest CAGR 144.1% -> 143.7% (Robinhood-style margin 137.1% -> 136.8%); MaxDD, the margin breaches and the fresh four-year starts with a breach are unchanged. Research simulator: 138.3% -> 138.0%, walk-forward 128.9% -> 128.7%.
• 2026-10-13: admission adopted. On every session each sleeve also admits the best-ranked name it does not hold when that name sits at the top of the ranking and is larger by market cap than every name the sleeve holds; it replaces the worst-ranked held name and the sleeve re-weights. Backtest CAGR 129.5% -> 144.1%, MaxDD -66.7% -> -57.6%, margin breaches 2 -> 2 (2018-07-30 replaced by 2020-02-28), fresh four-year starts with a breach 79 -> 60 of 151 (Robinhood-style margin 123.0% -> 137.1%, breaches 1 -> 1, 47 -> 47 of 151). Research simulator: 124.1% -> 138.3%, walk-forward 114.9% -> 128.9%. Its limits are listed in section 3.
• 2026-10-13: the strategy is renamed Alphy (it had an internal code name). Name only: the rules and every number in this paper are unchanged.
• 2026-10-13: daily check adopted. Every stock sleeve is now checked on every session, not only at its scheduled rebalance: a held name that no longer passes the filters is sold at the next open, and a held name that has fallen in the ranking is sold when the best eligible name not held scores ahead of it by a wide statistical margin; the sleeve then refills and re-weights. The rank buffer returns to its earlier value, and eligibility again requires the long-term average to be rising (that rule drives the daily exits here; removing it from this configuration lowers the backtest by 1.3 points, while the same removal from the previous rules changed nothing). Backtest CAGR 110.9% -> 129.5%, MaxDD -59.8% -> -66.7%, margin breaches 1 -> 2 (2018-07-30 added), fresh four-year starts with a breach 47 -> 79 of 151 (Robinhood-style margin 106.1% -> 123.0%, breaches 0 -> 1, 1 -> 47 of 151). Research simulator: 107.2% -> 124.1%, walk-forward 93.1% -> 114.9%. Its known weaknesses are listed in section 3.
• 2026-10-13: eligibility slope conjunct removed: a name no longer needs its long-term average to be rising. The rule bound on 26,985 of 293,868 otherwise-eligible cells (9.2%) over 2010-02-11 to 2026-10-02 and changed no pick, so the backtest results are unchanged; the inertness is empirical, not logical, and the check must be re-run if the rank buffer changes (at a tighter buffer the rule would change 3 rebalances).
• 2026-10-13: weighting power and the stock volatility target raised, and a faster TQQQ re-entry adopted. Backtest CAGR 97.8% -> 110.9% (Robinhood-style margin 93.8% -> 106.1%).
• 2026-10-11: weighting power raised. Backtest CAGR 93.1% -> 97.8% (Robinhood-style margin 88.9% -> 93.8%).
• 2026-10-11: stock volatility target raised.
• 2026-10-10: rank buffer tightened (a held name stays only while it ranks near the top of the ranking).
• 2026-10-10: delisting rule cause-aware (acquisitions cashed at the last close, other delistings a total loss).
• 2026-10-10: cash and distress filters compare each stock only with filings available at the time.