The criterion that buys a company is now the criterion that sells it. Every other exit rule tested — a percentage cut, a rank cutoff, a trailing stop, a moving average — was an arbitrary number bolted onto the screen, and all of them were inert or harmful. This one improved both halves of the sample and halved turnover. Bounding the growth gate above, rather than only below, improved them again.
What was removed: the 8% annual cut, the 2% quarterly cut, the “must be in the new top 25” rank test, and every stop-loss variant. Three rules became one.
| Rebalance | Previous rules | Gate exit + growth band | S&P 500 |
|---|---|---|---|
| Annual | 12.17%/yr | 15.54%/yr | 13.42%/yr |
| Quarterly | 14.76%/yr | 20.01%/yr | 13.42%/yr |
| Turnover (quarterly) | 546% | 288% | — |
Both cadences beat the index under the new rule; neither did reliably before. Turnover halving matters more than it looks — trading costs were the binding constraint on whether any of this was worth running at a retail account size.
This is the test that mattered. Nearly every other improvement tried in this project looked good over twelve years and turned out to be a strong second half masking a weak first. This one does not.
| Era | Previous rules | Gate exit | Turnover |
|---|---|---|---|
| 2014–2019 | 4.21%/yr | 8.71%/yr | 578% → 287% |
| 2020–2025 | 26.38%/yr | 30.06%/yr | 535% → 302% |
Removing each gate in turn, holding the rest constant:
| Gate removed | Effect on return | Verdict |
|---|---|---|
| revenue growth band 20–80% | −3.18pp | load-bearing |
| ROIC ≥ 10% | −2.46pp | load-bearing |
| FCF conversion ≥ 50% | −1.63pp | load-bearing |
| net debt ≤ 3× EBITDA | −0.99pp | load-bearing |
| net income > 0 | 0.00pp | never binds |
| free cash flow > 0 | 0.00pp | never binds |
Two gates have never excluded a single company that the others did not already exclude — positive earnings and positive cash flow follow from requiring 50% cash conversion and a 10% return on capital. They are kept for safety on future data, not because they do anything.
The growth gate only works with a ceiling on it. As a floor alone it contributed +0.15pp over twelve years and swung +5.35 / −4.21 by era — it was the strategy’s single largest source of regime dependence. Bounded above at 80% it becomes the most load-bearing gate of the six, worth 3.18pp, and it improves both halves of the sample.
The gate had a floor and no upper bound, so a company reporting 41,644% revenue growth cleared it. Of 3,366 gate-passing company-quarters, 280 report growth above 80% — 95 of them above 200% and 52 above 500%.
Those are almost never businesses tripling in size. Healthcare is over-represented among them by 2.6× and energy by 4.2×: pre-commercial biotech growing from a near-zero revenue base, milestone and licensing lumpiness, merger accounting, and 2020 base effects. Alaska Air printed 3,663% because its revenue had collapsed the year before. Expand Energy’s 168% is the Chesapeake–Southwestern combination, not organic growth.
| Period | No ceiling | Ceiling at 80% | Random exclusion, same count | p |
|---|---|---|---|---|
| full 2014–25 | 16.69% | 20.01% | 16.44% | 0.000 |
| 2014–2019 | 5.48% | 8.71% | 6.03% | 0.010 |
| 2020–2025 | 28.01% | 30.06% | 27.13% | 0.070 |
The comparison that matters is the third column. Excluding the same number of gate-passers at random does not help — its median is 16.44% against 16.69% for excluding nothing. So the gain is in which names are dropped, not how many. Over the full period the ceiling beat all 300 random draws.
This is a data-quality filter rather than a return prediction, which is why it holds across regimes where nothing else did. The composite score has capped growth at 50% since it was first written, for exactly this reason. The gate simply never did.
The direction is well supported; the threshold is not. Every ceiling between 60% and 120% improved both halves, and 80% was chosen from that range rather than established by it.
Each threshold swept while the others were held at default, and measured on the floors alone before the growth ceiling above was added. None sits on a peak, which is the reassurance being sought: ROIC scores better at 20% than at 10%, leverage better at 1× than 3×, and the growth floor better lower than higher — the last of which is what led to testing a ceiling in the first place. Had the chosen values topped their own curves, that would have been evidence they were reverse-engineered from the data.
Most of the surfaces are noise — growth swings five points between adjacent settings — and should not be chased. One is clean: leverage runs 1× 18.34%, 2× 17.07%, 3× 16.69%, 10× 16.55%, monotonically. Balance-sheet strength is doing real work.
| Exit rule | Return | vs no rule |
|---|---|---|
| none — boundary exits only | 15.23% | — |
| trailing stop, 30% from peak | 15.05% | −0.18 |
| trailing stop, 20% | 12.94% | −2.29 |
| trailing stop, 10% | 8.49% | −6.74 |
| underperforms the index by 5% | 6.82% | −8.41 |
| close below the 200-day moving average | 6.75% | −8.48 |
The gradient is monotone: the tighter the stop, the worse the outcome. Individual holdings run roughly 24% quarterly volatility, so a 10% trailing stop fires on ordinary noise rather than on anything informative — it crystallises the downside of a distribution whose upside it then misses.
Every formation from December 2013 to July 2026. Weight is the position’s share of the book at the close of the period; return is over that quarter alone. The final formation is still open.
| Ticker | Sector | Score | Held since | Weight | Return to date | Value | Outcome |
|---|---|---|---|---|---|---|---|
| STX | Technology | 47.6 | 2025-04-01 | 17.40% | -7.2% | £224,768 | open |
| FIX | Industrials | 53.9 | 2024-04-01 | 8.05% | -13.6% | £104,067 | open |
| LRCX | Technology | 50.0 | 2025-07-01 | 5.32% | -21.4% | £68,766 | open |
| SEZL | Technology | 65.1 | 2026-04-01 | 4.85% | -31.2% | £62,681 | open |
| AVGO | Technology | 47.2 | 2024-07-01 | 4.39% | -3.1% | £56,689 | open |
| APH | Technology | 73.9 | 2025-04-01 | 4.32% | -3.9% | £55,798 | open |
| FIVE | Consumer Disc | 65.9 | 2026-07-01 | 4.07% | +38.1% | £52,570 | open |
| HL | Materials | 77.6 | 2026-07-01 | 3.91% | +32.7% | £50,510 | open |
| CF | Materials | 79.6 | 2026-07-01 | 3.65% | +23.9% | £47,176 | open |
| BELFA | Technology | 72.1 | 2025-12-31 | 3.51% | -28.1% | £45,300 | open |
| SCCO | Materials | 65.8 | 2026-07-01 | 3.49% | +18.4% | £45,081 | open |
| PTC | Technology | 69.6 | 2026-07-01 | 3.47% | +17.7% | £44,822 | open |
| PLX | Healthcare | 85.4 | 2026-07-01 | 3.42% | +16.1% | £44,195 | open |
| AMCR | Industrials | 66.6 | 2026-07-01 | 3.05% | +3.5% | £39,388 | open |
| GMED | Healthcare | 68.9 | 2026-07-01 | 3.04% | +3.1% | £39,239 | open |
| NBIX | Healthcare | 70.4 | 2026-04-01 | 3.01% | -7.2% | £38,878 | open |
| NUTX | Technology | 94.2 | 2026-07-01 | 2.99% | +1.3% | £38,576 | open |
| TPC | Industrials | 77.8 | 2026-04-01 | 2.88% | +11.1% | £37,176 | open |
| CODA | Technology | 73.6 | 2025-12-31 | 2.87% | +5.6% | £37,117 | open |
| FSLR | Technology | 76.1 | 2026-04-01 | 2.63% | -11.9% | £33,948 | open |
| NXT | Technology | 67.1 | 2025-12-31 | 2.55% | -27.2% | £32,925 | open |
| ANIP | Healthcare | 70.9 | 2026-07-01 | 2.53% | -14.2% | £32,652 | open |
| CHE | Healthcare | 68.0 | 2025-07-01 | 1.80% | +8.2% | £23,273 | open |
| LMB | Industrials | 60.8 | 2026-04-01 | 1.59% | -35.8% | £20,586 | open |
| NOW | Technology | 56.0 | 2023-12-29 | 1.23% | +33.5% | £15,874 | open |
Across the 50 closed formations: 756 positions carried into the next quarter, 458 sold on a gate failure, 6 removed by delisting. Under the previous rules almost nothing was ever carried — the book was rebuilt from scratch each period. Holdings now persist for as long as the business keeps clearing the bar.
Both start at £100,000 on 31 December 2013. The line runs past the backtest’s January 2026 exit to the present formation.