TL;DR
Protected quality stocks returned 10.34% with 7.50% daily drawdown from January 2025 through September 2026. The simpler SPY trend control returned 25.91% with 5.80% drawdown. Longer retrospective selection improved growth without improving the worst drawdown. Cash helped in several bear windows, but the study did not establish a general rule that halves bear losses. The strongest protected screen selected business quality rather than cheap valuation.
The simple SPY trend control grew more and fell less recently
Portfolio value (USD)
- Protected quality 2026-09-30, 20:00:00 UTC
- $11,034.24
- Asset profitability 2026-09-30, 20:00:00 UTC
- $15,146.33
- Quality, no gate 2026-09-30, 20:00:00 UTC
- $11,887.62
- SPY, cash dividends 2026-09-30, 20:00:00 UTC
- $13,119.07
- SPY trend 2026-09-30, 20:00:00 UTC
- $12,590.78
Protected quality: 7.50% largest decline in plotted observations. 2026-01-16, 21:00:00 UTC ($10,979.23) to 2026-04-10, 20:00:00 UTC ($10,155.32).
Show the table · 7 rows, 5 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Protected quality | +10.34% | 7.50% | 82.42% | $56.46 |
| Asset profitability | +51.46% | 22.66% | 98.51% | $49.81 |
| Monthly loss guard | +46.58% | 21.59% | 98.52% | $57.16 |
| Quality, no gate | +18.88% | 16.90% | 98.45% | $22.78 |
| Ordinary value | +32.04% | 19.12% | 98.49% | $235.22 |
| SPY, cash dividends | +31.19% | 19.00% | 99.31% | $5.00 |
| SPY trend | +25.91% | 5.80% | 83.65% | $47.17 |
No matching rows. Clear the filter to see all records.
Each is a separate fresh-start native backtest. Stock books request 100% risk-on; actual investment is lower because of reserve, cash and rounding.
The fixed protected stock winner returned +10.34% from January 2025 through September 2026, with 7.50% daily drawdown. SPY using a native 200-Day trend filter returned +25.91% with 5.80% drawdown. The simpler market control grew more and fell less.
A changing selection procedure produced a different result. Across four retrospective tests from 2018 through September 2026, its reset-fold chain returned +255.81%, versus +220.54% for unchanged quality stocks and +211.98% for matched SPY. Its worst daily drawdown remained 35.73%, slightly greater than SPY's 34.10%.
Fourteen starting portfolios covered valuation, profitability, dividends, debt and momentum. Actual losing trades led to seven further rule variants. Cash produced useful protection in several bear windows, but neither the fixed winner nor the changing selection process established a general rule that halves bear losses.
Recent exposure and risk
Value (%)
- Protected quality 2026-09-30, 20:00:00 UTC
- -6.33%
- Asset profitability 2026-09-30, 20:00:00 UTC
- -7.51%
- Quality, no gate 2026-09-30, 20:00:00 UTC
- -5.62%
- SPY, cash dividends 2026-09-30, 20:00:00 UTC
- -1.93%
- SPY trend 2026-09-30, 20:00:00 UTC
- -1.96%
The protected training winner, a growth stock alternative and two actual SPY controls. Labels are shortened; the full table includes costs, drawdown and exposure.
The strongest protection idea became a quality basket
The protected winner screens for dividend growth, positive income, free cash flow and equity. It has no P/E or other cheapness requirement. Asset profitability also ranks a business-quality ratio without a valuation ceiling. Their results cannot show that cheap stock selection caused the protection.
The valuation families remain visible below. Cash exits change how much money is invested; they are a separate choice from which stocks qualify. Mean exposure and the unprotected alternatives help show that tradeoff.
In the final period, the protected winner gave up 20.85 percentage points of growth versus SPY hold. SPY trend averaged 83.65% invested, compared with 82.42% for the protected basket. Similar average exposure still permits different holdings and entry/exit dates.
Fourteen starting ideas, six stock-selection families
Inspect the fourteen starting ideas
Scroll sideways to compare all columns. Select a heading to sort.
| Completed-year dividend increases | +484.62% / 36.30% | +183.92% / 16.62% |
| Quarterly FCF / market cap | +460.36% / 48.12% | +125.80% / 38.64% |
| Quarterly ROE proxy / TTM P/E | +322.00% / 42.78% | +94.71% / 29.23% |
| Quarterly income / assets | +457.82% / 29.70% | +137.60% / 28.97% |
| Earnings yield + positive momentum | +198.44% / 40.49% | +72.38% / 27.73% |
| Low-debt profitability | +275.67% / 33.78% | +71.26% / 34.43% |
| Dividend quality: monthly 50% risk-off | +287.11% / 22.69% | — |
| Asset profitability: monthly 50% risk-off | +227.07% / 25.22% | — |
No matching rows. Clear the filter to see all records.
2011-01-01 through 2021-12-31. Each cell is cumulative return / daily drawdown, after fees and cash dividends. All up to 20 equal-weight stocks. The last two rows are additional monthly half-exposure ideas.
The growth objective changes which portfolio looks best
Portfolio value (USD)
- Protected quality 2021-12-31, 21:00:00 UTC
- $29,958.86
- Asset profitability 2021-12-31, 21:00:00 UTC
- $55,781.64
- Quality, no gate 2021-12-31, 21:00:00 UTC
- $58,462.49
- SPY, cash dividends 2021-12-31, 21:00:00 UTC
- $41,165.16
- SPY trend 2021-12-31, 21:00:00 UTC
- $22,934.23
Protected quality: 17.10% largest decline in plotted observations. 2014-09-05, 21:18:02 UTC ($14,465.05) to 2016-03-10, 22:21:44 UTC ($11,991.16).
Unprotected quality produced the largest starting-grid growth: +484.62% with 36.30% drawdown. Asset profitability returned +457.82% with a smaller 29.70% drawdown. Full protection of dividend quality lowered growth to +183.92% and drawdown to 16.62%.
Selection used annualized growth divided by daily maximum drawdown, with no additional return floor. The highest ratio chose the protected basket. We retained growth alternatives rather than treating the smallest loss as the only useful result.
2011 to 2021: loss from peak
Value (%)
- Protected quality 2021-12-31, 21:00:00 UTC
- 0%
- Asset profitability 2021-12-31, 21:00:00 UTC
- -0.8%
- Quality, no gate 2021-12-31, 21:00:00 UTC
- 0%
- SPY, cash dividends 2021-12-31, 21:00:00 UTC
- -0.48%
- SPY trend 2021-12-31, 21:00:00 UTC
- -0.52%
Inspect earlier-period drawdowns
The cash cooldown cut trading but worsened one exit
The original protected tape contained 5,072 fills and 76 invested-to-flat segments; 49 lasted fewer than 10 days. The seven-day cooldown on repeated cash-rule firing reduced those counts to 4,432 fills, 60 segments and 32 short segments. Training fees fell from $1,216.10 to $960.29, while return rose from 183.92% to 199.59%. Drawdown increased from 16.62% to 17.10%.
The October 8, 2014 reentry then exited October 14 rather than October 9. That segment lost 3.23% instead of 1.99%. The cooldown addressed repeated exits while permitting a later loss to become larger. Its scope matters: it limits the cash rule, not entry.
Putting a seven-day cooldown on buys instead produced only 109.98% growth and 23.41% drawdown. Actual BMY lots still lost roughly 32%. Lower fees alone did not produce a better portfolio.
Show the table · 7 rows, 5 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Cash exit cooldown 7 days | +199.59% | 17.10% | $960.29 | 83.43% |
| Monthly stock trend: asset basket | +324.46% | 29.60% | $908.91 | 98.44% |
| Monthly stock trend: protected basket | +163.05% | 15.83% | $1,456.56 | 82.27% |
| Daily 50% risk-off refresh | +257.18% | 23.01% | $511.20 | 85.25% |
| Seven-day delay on buys | +109.98% | 23.41% | $756.93 | 78.39% |
| Monthly held-name loss guard | +448.23% | 29.58% | $664.73 | 98.45% |
| Monthly 50% + buy delay | +293.97% | 23.01% | $447.78 | 88.51% |
No matching rows. Clear the filter to see all records.
2011 to 2021 comparisons against unchanged parents. The primary remained the round-two cash-exit cooldown. Daily half exposure differs from the starting monthly half rule.
A smaller targeted loss did not guarantee better total growth
Show the table · 3 rows, 4 columns
Scroll sideways to compare all columns. Select a heading to sort.
| FL: April 2017 entry | −52.62% | −23.69% | −23.69% |
| AMAT: June 2018 entry | −36.21% | Entry removed | −20.16% |
| XRX: March 2020 entry | −38.13% | Entry removed | −41.24% |
No matching rows. Clear the filter to see all records.
Worst dollar-loss FIFO fragment from each specified entry window, including allocated fill fees and excluding credited dividends. These are realized lot fragments, not independent full round trips. The loss guard is checked at monthly selection against average position cost; it is not an immediate 15% stop.
The stock-trend guard removed the AMAT and XRX entries and reduced the FL fragment loss. Full-period asset-basket growth fell from 457.82% to 324.46%, while drawdown barely changed, 29.70% to 29.60%. The monthly loss guard preserved more growth at 448.23% and 29.58% drawdown, but its XRX sale was worse. It sold that March 6, 2020 lot on April 6 instead of June 5, before the later recovery.
All seven captured event tapes reconciled fill fees and dividend cash to native totals and had no unmatched sold quantities in the FIFO replay. This covered captured fills; sampled condition and episode inspections do not establish every signal cause or certify every historical corporate action.
The training winner remained weak on 2022 to 2024 dates
Portfolio value (USD)
- Protected quality 2024-12-31, 21:00:00 UTC
- $11,062.42
- Asset profitability 2024-12-31, 21:00:00 UTC
- $14,122.93
- Quality, no gate 2024-12-31, 21:00:00 UTC
- $12,485.97
- SPY, cash dividends 2024-12-31, 21:00:00 UTC
- $12,677.56
- SPY trend 2024-12-31, 21:00:00 UTC
- $11,516.56
Protected quality: 18.58% largest decline in plotted observations. 2022-01-04, 21:00:00 UTC ($10,010.34) to 2023-03-17, 21:03:59 UTC ($8,149.99).
Show the table · 7 rows, 5 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Protected quality | +10.62% | 18.58% | 69.94% | $163.15 |
| Asset profitability | +41.23% | 28.36% | 98.43% | $74.48 |
| Monthly loss guard | +38.37% | 28.61% | 98.43% | $94.29 |
| Quality, no gate | +24.86% | 21.19% | 98.43% | $35.06 |
| Ordinary value | +3.06% | 23.36% | 98.49% | $268.29 |
| SPY, cash dividends | +26.78% | 24.74% | 98.33% | $5.00 |
| SPY trend | +15.17% | 20.61% | 69.46% | $174.06 |
No matching rows. Clear the filter to see all records.
Frozen rules; no changes after these later comparisons. Each portfolio starts flat with $10,000.
2022 to 2024: loss from peak
Value (%)
- Protected quality 2024-12-31, 21:00:00 UTC
- -4.97%
- Asset profitability 2024-12-31, 21:00:00 UTC
- -5.41%
- Quality, no gate 2024-12-31, 21:00:00 UTC
- -4.97%
- SPY, cash dividends 2024-12-31, 21:00:00 UTC
- -3.47%
- SPY trend 2024-12-31, 21:00:00 UTC
- -3.56%
Inspect 2022 to 2024 drawdowns
The frozen winner did not halve every bear loss
Show the table · 4 rows, 7 columns
Scroll sideways to compare all columns. Select a heading to sort.
| 2011-04-29 to 2011-10-03 | -4.35% | 5.58% | 14.69% | 18.89% | 70.45% | 39.22% |
| 2018-09-20 to 2018-12-24 | -5.18% | 5.50% | 12.93% | 17.97% | 69.38% | 27.51% |
| 2020-02-19 to 2020-03-23 | -11.32% | 11.40% | 36.11% | 30.73% | 62.91% | 30.28% |
| 2022-01-03 to 2022-10-12 | -12.46% | 13.13% | 21.18% | 24.74% | 46.93% | 17.25% |
No matching rows. Clear the filter to see all records.
2011/2018/2020 windows are slices of 2011 to 2021 books;2022 is a slice of 2022 to 2024 books. Window drawdown starts from the first saved observation in that window. SPY cash exposure drifts as dividends accumulate; different fresh-start periods must not be spliced.
The frozen rule cut SPY’s recorded drawdown by more than half in 2011, 2018 and 2020. The reduction was 46.93% in the 2022 window, below the proposed halfway target. These windows are slices of existing books, rather than newly restarted portfolios.
Earlier experiment: keep the 65.64% and 56.49% results separate
Drawdown uses the last saved observation per UTC date within each specified window. Average stock exposure uses those observations, including event dates.
Inspect the separate earlier bear-window experiment
Scroll sideways to compare all columns. Select a heading to sort.
| 2020-02-19 to 2020-03-23 | 11.72% | 35.54% | 34.10% | 65.64% | 30.30% |
| 2022-01-03 to 2022-10-12 | 10.41% | 21.14% | 23.91% | 56.49% | 14.77% |
No matching rows. Clear the filter to see all records.
Drawdown uses the last saved observation per UTC date within each specified window. Average stock exposure uses those observations, including event dates.
The earlier immediate-cash quality portfolio used a separate 2020 to 2024 fresh-start comparison, without the new cash-exit cooldown. Its 2020 and 2022 drawdown reductions were 65.64% and 56.49% versus that comparison's SPY cash book. These historical figures belong to those original rules and starting dates; they are not the new frozen winner's results.
Four retrospective folds of the same selection procedure
Inspect each chronological selection
The first two earlier-period selections chose quality stocks without the cash gate. The third chose protected quality, and the last chose the monthly loss guard. Each next-period test started with cash. The eight-candidate shortlist used already known history, so these selections remain retrospective.
Scroll sideways to compare all columns. Select a heading to sort.
| 2011-01-01 to 2017-12-31 | 2018-01-01 to 2019-12-31 | Quality, no gate | +37.68% / 14.53% | +37.68% / 14.53% | +23.52% / 19.57% | -1.94% / 17.11% |
| 2011-01-01 to 2019-12-31 | 2020-01-01 to 2021-12-31 | Quality, no gate | +54.02% / 35.73% | +54.02% / 35.73% | +49.75% / 34.10% | +36.54% / 18.50% |
| 2011-01-01 to 2021-12-31 | 2022-01-01 to 2023-12-31 | Protected quality | -3.86% / 18.58% | +8.72% / 21.19% | +2.06% / 24.74% | -4.79% / 20.61% |
| 2011-01-01 to 2023-12-31 | 2024-01-01 to 2026-09-30 | Monthly loss guard | +74.54% / 22.79% | +39.04% / 18.97% | +65.26% / 18.78% | +54.13% / 9.83% |
No matching rows. Clear the filter to see all records.
Eight frozen candidates; select on earlier dates by the same CAGR/daily-drawdown criterion. Each next-date test starts flat. The shortlist was devised using known 2011 to 2021 history, including early test years, so these folds are retrospective.
Reselecting grew more overall but kept a 35.73% drawdown
Portfolio value (USD)
- Changing selection 2026-09-30, 20:00:00 UTC
- $35,581.13
- Unchanged quality 2026-09-30, 20:00:00 UTC
- $32,053.62
- SPY cash dividends 2026-09-30, 20:00:00 UTC
- $31,198.09
- SPY trend 2026-09-30, 20:00:00 UTC
- $19,648.50
Changing selection: 35.73% largest decline in plotted observations. 2020-02-14, 23:26:15 UTC ($14,410.48) to 2020-03-23, 22:58:34 UTC ($9,261.26).
Show the table · 5 rows, 4 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Changing selection | +255.81% | 35.73% | $35,581.13 |
| Unchanged quality | +220.54% | 35.73% | $32,053.62 |
| Ordinary value | +89.31% | 59.44% | $18,930.86 |
| SPY cash dividends | +211.98% | 34.10% | $31,198.09 |
| SPY trend | +96.49% | 21.88% | $19,648.50 |
No matching rows. Clear the filter to see all records.
Independent 2018 to 2019, 2020 to 2021, 2022 to 2023 and 2024 to September 2026 test curves normalized to previous ending capital. Entry fees and first-session movement retained. This is a hypothetical reset procedure, not one continuously held account.
2018 to September 2026: retrospective selection loss from peak
Value (%)
- Changing selection 2026-09-30, 20:00:00 UTC
- -7.51%
- Unchanged quality 2026-09-30, 20:00:00 UTC
- -5.62%
- SPY cash dividends 2026-09-30, 20:00:00 UTC
- -1.92%
- SPY trend 2026-09-30, 20:00:00 UTC
- -1.96%
Exact rules and execution conventions
Read exact stock, cash, timing and comparison rules
Protected quality stock selection
Use the native S&P 500 source with a minimum market cap of $500 million. Exclude missing market cap and select one share class per company. Exclude Brown-Forman aliases before ranking for the known 2016 split defect. Require positive indicated dividend yield, quarterly income, quarterly free cash flow and equity. Require at least one completed-year increase in regular per-share dividends; rank the consecutive completed-year increase count, capped at five. Select up to 20 stocks with equal target weights.
Entry, refresh and cash rule
When SPY price exceeds its native 200-Day SMA, enter if the book is flat; otherwise refresh after 30 calendar days since the selection strategy last fired. Request 100% deployment. At or below the average, request 0% stock deployment only when at least 7 days have elapsed since the cash rule last fired. This exit cooldown can delay liquidation after reentry.
Growth alternatives
Asset profitability ranks latest-quarter net income / total assets among positive income, FCF and equity names. It has no market gate or valuation ceiling. The monthly loss repair additionally excludes held candidates with position return at or below -15% versus average cost during scheduled selection; unheld candidates can qualify again. Unprotected quality keeps the same dividend screen without the market gate.
Matched SPY controls
The hold book buys 100% of available buying power when flat, once, and accumulates cash dividends. The trend book buys when flat above the native 200-Day average and sells all below it; its strict-below exit differs from the stock basket's at-or-below condition. Exposure is measured, not assumed constant. Native reinvested comparisonValue is reported separately in private receipts and never substituted for these cash-policy books.
Costs, data and timing
Every book starts at $10,000, uses Day backtests and OpenClose evaluation, fractional positions and 0.05% notional fees on each filled stock order. Native execution reserve and rounding remain. Cash dividends may be redeployed by stock rebalances. The backtests do not separately model slippage, spreads, market impact or taxes. Native SMA uses a calendar-duration window of captured observations, not 200 trading-session closes. Fundamentals use the observations supplied to the engine at selection.
Universe and evidence limits
Historical S&P 500 membership maps apply when present; native behavior passes through a missing map. There is no explicit fail-closed membership filter. Ratios with latest quarterly income or FCF are quarterly proxies. Native daily charts take the last saved observation per reported UTC date, including some corporate-action dates. No live portfolio was activated.
Aurora inspected the trades; native backtests produced the numbers
Read the agent and native-engine workflow
Codex declared 14 starting portfolios and executed the saved native specifications. Aurora read the actual parent portfolios, inspected native losing-event histories and proposed both repair rounds. Codex checked the rule semantics, compiled supported changes, selected by the declared earlier-period criterion and calculated the exported histories.
MCP supplied the interface for portfolio authoring and backtests. Complete owned histories and tapes were then exported read-only through the same native models because paginated history queries repeatedly loaded full position histories. Five completed MCP curves matched those exports within $0.00005 NAV rounding. No backtest was replaced by an independent synthetic price replay.
Inspect the full evidence
The downloads contain all 14 initial variants, both repair rounds, eight exact frozen rule graphs, every fold's training rankings and test results, 69 actual daily curves and seven fill-audit summaries. Private account, portfolio and backtest identifiers are excluded.
All dates in this campaign were already known. The shortlist used 2011 to 2021, including dates later reused in early folds. These comparisons help examine tradeoffs; a new prospective claim would require future observations.
No comments yet.