Components
Current rules: the slower re-rank schedule, without the post-release cut or the inventory-versus-revenue screen, adopted 2026-10-31. Fixed-January walk-forward 231.31% on the research basis, with a maximum drawdown of -59.8%, at the first schedule anchor on data through 2026-10-28. The post-release cut on the top name was dropped on 2026-10-29 by decision (see Dropped). The inventory-versus-revenue screen was removed on 2026-10-31 by decision: its comparison with the sector median ran backwards where that median was negative, and most of its measured gain came from one stock flagged in March 2020 through that reversal. Its row (Oct 31) is in the table.
Every rule added to or removed from the strategy, oldest to newest, measured against the rules just before it, on the same data (through 2026-10-28). The measure is the fixed-January walk-forward: the average yearly return of twelve one-year tests, each starting in January, 2014 to 2025. The stock sleeves rebalance on a fixed calendar, so the result depends a little on which day that calendar starts. Each rule is therefore measured from 20 different start days (called anchors here), and the table shows the average over the 20. A gain of +1pp (one percentage point) means the average yearly return went up by one point, for example from 200% to 201%. A rule that was removed later stays in the table, because the rules after it were measured on top of it, and its removal is its own row.
The rows add up to today's rules. The earliest rules in the table average 95.61% a year over the 20 start days; adding every row gives 223.89%, which is today's rules averaged over the same 20 start days (231.31% from the first start day, the headline figure).
Tests
Each component is judged by the test for its type:
- Return add: the gain must survive without its single best year. With that year removed, the gain over the prior rules must still be at least +2pp, and positive at more than half of the anchors.
- Simplification: it must remove a rule or input and cost under 1pp.
- Drawdown device: judged here by a proposed drawdown test. It was pre-registered on 2026-10-29 as a research test. It has not been adopted as a gate for the live rules, and the rules of record are unchanged by it; it is applied here for information. It has three parts: depth (the worst fall from a peak is at least 2 points smaller on average, and smaller from at least 15 of the 20 start days), not one episode (with the year that improved most taken out, another year still improves by at least 1 point, and the other years are not worse on average), and cost (the return given up is at most 1 point a year for each point of drawdown saved, and never more than 3 points). The return test is shown alongside for reference.
- Removal by decision: a rule taken out by Karthik's decision, not by a test. Its row shows what removing it changed.
Result
In today's rules, 9 of 11 return adds pass the test without their best year. 2 of 2 simplifications pass the simplification test. The post-release cut on the top name (drawdown device) is no longer live: it was dropped on 2026-10-29 by decision. It is not counted here.
| Date (2026) | Rule | Type | Gain (average of 20 start days) | Without its best year (start days positive) | Result |
|---|---|---|---|---|---|
| Oct 13 | Daily event rotation with a rank buffer (on the Oct 13 starting rules) | Return add | +18.73pp | +9.57pp without 2022 (20 of 20) | Pass |
| Oct 13 | Leader-emergence admission | Return add | +12.87pp | +6.71pp without 2018 (20 of 20) | Pass |
| Oct 14 | Pre-release boost on the top name | Return add | +3.92pp | +2.89pp without 2025 (20 of 20) | Pass |
| Oct 14 | Leveraged-ETF sleeve halved while the pre-release window is on | Return add | +9.11pp | +6.79pp without 2025 (20 of 20) | Pass |
| Oct 14 | Regime refresh of the stock sleeves | Return add | +2.95pp | +1.75pp without 2020 (20 of 20) | Fail |
| Oct 15 | New stock volatility target paired with a resized leveraged-ETF sleeve | Return add | +11.86pp | +7.99pp without 2025 (20 of 20) | Pass |
| Oct 15 | Release-session hold (the pre-release window stays on through the estimated release session) | Return add | +2.25pp | +1.06pp without 2024 (20 of 20) | Fail |
| Oct 15 | Release estimate without the fallback rule | Simplification | -0.62pp | Does not apply (a simplification is judged on its cost) | Pass |
| Oct 16 | Post-release cut on the top name | Drawdown device | +2.57pp | -0.40pp without 2021 (0 of 20) | Dropped Oct 29 (had passed the proposed drawdown test) |
| Oct 16 | Fundamentals carry cap removed | Simplification | +0.30pp | Does not apply (a simplification is judged on its cost) | Pass |
| Oct 22 | Dip rule on the top name, with the leveraged-ETF sleeve halved | Return add | +10.91pp | +6.06pp without 2022 (20 of 20) | Pass |
| Oct 28 | Quality strip: hard quality strip plus a held-immune soft strip | Return add | +40.15pp | +34.20pp without 2021 (20 of 20) | Pass |
| Oct 28 | Gross-margin and inventory-versus-revenue screens in the quality strip: the soft screen also flags a fall in gross margin, plus an inventory-versus-revenue comparison with the sector that was removed on Oct 31 (see that row) | Return add | +11.12pp | +5.12pp without 2020 (20 of 20) | Pass |
| Oct 29 | Slower re-rank schedule: each stock sleeve re-ranks less often than it re-weights | Return add | +14.81pp | +10.51pp without 2024 (20 of 20) | Pass |
| Oct 29 | Post-release cut on the top name removed | Removal by decision | +3.10pp | -0.07pp without 2019 (11 of 20) | Removed by decision |
| Oct 31 | Inventory-versus-revenue screen removed from the quality strip | Removal by decision | -15.78pp | -17.41pp without 2019 (0 of 20) | Removed by decision |
Notes
Each component that fails its test, is not judged by the return gate, or was dropped carries these notes.
Regime refresh of the stock sleeves (2026-10-14): Fail
- Why adopted. Adopted 2026-10-14, live 2026-10-15: when a QQQ rebound starts or the leveraged-ETF gate leaves off, every stock sleeve re-selects on the next session instead of waiting for its schedule. It passed the bar then in use (+2.08pp fixed-January walk-forward).
- How it fails the gate. Fails the return gate: +2.95pp, mostly from 2020 (+55.7pp in that year); +1.75pp without 2020, positive at 20 of 20 anchors but under +2pp.
Release-session hold (2026-10-15): Fail
- Why adopted. Adopted 2026-10-15, last in that day's changes, with its measured increment: the pre-release window on the top name stays on through the estimated release session itself.
- How it fails the gate. Fails the return gate: +2.25pp, mostly from 2024 (+36.7pp in that year); +1.06pp without 2024, positive at 20 of 20 anchors but under +2pp. Its adoption measurement already showed +0.92pp without 2024.
Release estimate without the fallback rule (2026-10-15): Pass
- Why adopted. A simplification decided 2026-10-15 from the same removal battery: the release-date estimate uses the year-over-year match only and drops the fallback rule, one fewer code path. Names with no year-over-year match get no estimate, so the pre-release window stays off for them.
- How it passes its test. Costs 0.62pp. No year improves; the loss is in 2015 and 2016. Passes the simplification test (cost under 1pp, removes a rule).
Post-release cut on the top name (2026-10-16): Dropped 2026-10-29
- Why adopted. Adopted 2026-10-15 as a drawdown device, live 2026-10-16: after a negative release-day reaction on the top name, the stock book is cut for 20 sessions.
- How it was judged. Passed the proposed drawdown test on the lineage: the maximum drawdown is 6.0pp shallower on average and shallower at all 20 anchors (+5.7 to +6.4pp). The gain is not one episode: the 2021 drawdown is 6.1pp shallower, 2018 is 3.9pp shallower and 2017 is 1.4pp shallower, and no year is deeper. Its return cost is within the limit: +2.57pp overall and -0.40pp without its best year (2021), against an allowed -3.0pp. On returns alone it would fail the return gate (2020 is 54.9pp worse and 2021 is 36.4pp better). In the rules with the slower re-rank schedule before the drop it failed the proposed drawdown test on cost (see Dropped).
- Why dropped. Dropped from the live rules on 2026-10-29 by decision, not by a test. Removing it from the slower-re-rank rules raised the fixed-January walk-forward from 243.61% to 247.87% (+4.3pp) and the CAGR from 222.3% to 225.2% (+2.9pp), and the maximum drawdown is 4.4pp deeper (-59.8% vs -55.4%), research basis, first schedule anchor, data through 2026-10-28.
Fundamentals carry cap removed (2026-10-16): Pass
- Why adopted. A simplification from the one-at-a-time removal battery run on 2026-10-15, decided 2026-10-16: each company's latest fundamentals report now carries until its next report with no age limit, one setting fewer on both screening layers. It was not adopted for return.
- How it passes its test. Costs nothing: +0.30pp, all of it from 2017 (every other year is unchanged), so 0.00pp without 2017. Passes the simplification test (cost under 1pp, removes a rule).
Post-release cut on the top name removed (2026-10-29): Removed by decision
- Why adopted. Removed by decision on 2026-10-29. It was a drawdown device: it cut the top name after its release, which made drawdowns shallower but cost return.
- How it passes its test. Not judged by a test. Removing it added 3.10pp on the average of the 20 start days, and the worst fall from a peak became 6.45pp deeper on average.
Inventory-versus-revenue screen removed from the quality strip (2026-10-31): Removed by decision
- Why adopted. Removed by decision on 2026-10-31. The screen compared a company's inventory growth with revenue growth against its sector median, and the comparison ran backwards where that median was negative. Most of its measured gain came from one stock flagged in March 2020 through that reversal, so the gain is not trusted.
- How it passes its test. Not judged by a test. Removing it cost 15.78pp on the average of the 20 start days (16.56pp from the first start day, 247.87% to 231.31%). The worst fall from a peak is unchanged.
In today's rules
Today's rules are the slower re-rank schedule, without the post-release cut or the inventory-versus-revenue screen: fixed-January walk-forward 231.31% at the first anchor (223.89% on the mean over the 20 anchors). Each of these components was taken out of them alone, and the result compared with them on the same 20 anchors. Taking a rule out is a simplification, so it passes if it costs under 1pp. If it costs more, the rule earns its place in today's rules. A margin breach is a session on which the account's value, at the close or at the day's low, fell below the broker's maintenance requirement, counted over the full backtest under the headline margin model (maintenance as measured with a broker's own margin calculator, plus a stress requirement on concentrated positions). The backtest then sells pro rata at the next open and continues. The breach columns give that count with the rule in and with it taken out, averaged over the 20 anchors. They are shown for information and are not part of the test.
| Component | Taken out: walk-forward change, average (range over the 20 anchors) | Maximum drawdown change | Margin breaches, with the rule / without it | Result |
|---|---|---|---|---|
| Regime refresh of the stock sleeves | -5.79pp (-9.45pp to -3.37pp) | +0.02pp | 26.4 / 26.6 | Earns its place (costs 5.79pp) |
| Release-session hold | -2.02pp (-2.42pp to -1.55pp) | +0.00pp | 26.4 / 25.3 | Earns its place (costs 2.02pp) |
| Release-session hold and regime refresh together | -7.86pp (-11.46pp to -5.47pp) | +0.02pp | 26.4 / 25.5 | Earn their place (cost 7.86pp) |
- Regime refresh of the stock sleeves: in today's rules its gain without its best year (2016) is +3.38pp, at least +2pp, and taking it out costs 5.79pp, so taking it out fails the simplification test.
- Release-session hold: in today's rules its gain without its best year (2024) is +1.31pp, below +2pp, but taking it out costs 2.02pp, so taking it out fails the simplification test.
Dropped
Components taken out of the live rules by decision. They are kept here with their measurements.
Post-release cut on the top name: dropped 2026-10-29
Adopted 2026-10-16 as a drawdown device. Dropped from the live rules on 2026-10-29 by decision; the drop was a choice, not the result of a test. Removing it from the slower-re-rank rules moved the fixed-January walk-forward from 243.61% to 247.87% (+4.3pp) and the CAGR from 222.3% to 225.2% (+2.9pp). The maximum drawdown is 4.4pp deeper (-59.8% vs -55.4%). Research basis, first schedule anchor, data through 2026-10-28. Over the 20 anchors the mean maximum drawdown is -61.9% without it vs -55.5% with it.
Before the drop it was measured in the rules with the slower re-rank schedule, which still had it (fixed-January walk-forward 243.61% at the first anchor), by the proposed drawdown test:
| Component | Taken out: walk-forward change (anchor least favourable to the rule) | Maximum drawdown change | Margin breaches, with the rule / without it | Result |
|---|---|---|---|---|
| Post-release cut on the top name | +3.10pp (+5.14pp) | 6.45pp deeper | 19.1 / 26.4 | Fail (proposed drawdown test, failed: the cost check) |
- Depth check: maximum drawdown 6.45pp shallower with it (at least 3.85pp, at 20 of 20 anchors), passes. Not-one-episode check: best year 2021 (+6.80pp), next 2025 (+6.28pp), the other years average +0.95pp, passes. Cost check: it cost 3.10pp of walk-forward against a limit of 3.00pp, and 9.41pp without its best return year (2021), fails.
- On the lineage it had passed the proposed drawdown test (see Notes).
All the rules together
Measured end to end, from an earlier set of rules to the current one (the slower re-rank schedule, without the post-release cut or the inventory-versus-revenue screen):
- Oct 22 rules to today's rules: +53.41pp; without its best year (2019) +48.55pp, positive from 20 of 20 start days.
- Oct 16 rules to today's rules: +64.32pp; without its best year (2022) +55.40pp, positive from 20 of 20 start days.
- Event-rotation rules before the admission rule (2026-10-13) to today's rules: +109.54pp; without its best year (2025) +98.34pp, positive from 20 of 20 start days.
- Starting rules (2026-10-13, before event rotation) to today's rules: +128.27pp; without its best year (2022) +113.44pp, positive from 20 of 20 start days.
A component that fails its test is shown here, not hidden. It stays live unless it is removed by decision. The tests were introduced on 2026-10-29, and most components were adopted before that. The drawdown test is a proposed research test, not a gate on the live rules.