Top 10 / Method

How the Top 10 score is built

A list is a published formula and its output, not a recommendation.

The score

How a pillar becomes a number

Every input is ranked across the companies it is compared with — the whole covered universe for a market-wide list, the sector alone for a sector list — and turned into a percentile from 0 to 100. Where a pillar has several inputs they are averaged before ranking. A score is the weighted sum of those percentiles, so it says where a company sits among its peers on that night, never what it is worth.

Each bar is a percentile: how this company ranked against every company the score ranked, 0–100.

Who is eligible

A company is ranked only if it is covered, carries a sector, clears the size and liquidity floors, has a price fresh enough to use, and has at least one live input for every required pillar. A company missing a whole required pillar is left out rather than scored with a gap filled in — an omission is visible, an invented input is not.

One pillar is the exception. Capital return can be missing entirely — a company may genuinely return no capital to shareholders, and a company we have not measured yet looks identical to one that returns none. Rather than drop the company, the score is taken over the pillars that could be measured and re-scaled, and the row says in its own sentence how many of them there were.

How the ten are chosen

At each rebalance the top 10 by score are taken, with two rules on top. A sitting member is held while it stays inside the top 15, so a list of ten does not churn on the difference between rank 10 and rank 11. And no more than 4 members of a market-wide list may come from one sector; when the cap binds, the company that was skipped is recorded by name, so the question “why is it not in the list when it outscores one that is?” has a written answer rather than a re-derivation.

Two different numbers follow from those rules, and they are easy to confuse. The cutoff is simply the 10th-best score on the night a list is rebalanced: a company below it is not selected, and the company ranked 11 is the one closest to it. The exit score is the 15th-best score, and it is the one a sitting member is measured against — which is why a member can score below the cutoff and still be in the list, and why the two numbers are never the same figure. Both are read off the same nightly ranking, so both move every night while membership does not.

Between rebalances the membership does not change, except that a company that stops being eligible leaves immediately. Its slot is not refilled until the next scheduled rebalance — filling it early would be a selection decision taken outside the calendar this page publishes — so a list can legitimately carry fewer than ten for a while, and each page shows the count it actually has.

How performance is tracked

Each list is valued equal-weight from the day it went live, rebased to 100, and shown beside the same benchmark rebased on the same day. Members are held from one rebalance to the next; a company that leaves is included up to the day it left. There is no backfilled record: every list is tracked from its own first live day forward, and that date is printed on its page.

Both curves are computed on a total-return basis. A cash distribution buys more of the company that paid it, at that company's own closing price on the day the share starts trading without the dividend; if that day has no usable close the purchase is made at the next one. No tax is modelled and no timing judgement is applied. The same arithmetic is used on the benchmark, from its own distributions, because our price series keeps dividends in the price on both sides — crediting one side and not the other would flatter whichever side was credited, and the one we publish is the list.

If the benchmark's own distribution record is unavailable or has stopped being updated, both sides fall back to price alone rather than running on different bases, and each stored row records which of the two produced it. A distribution larger than the share it was paid on is not treated as cash at all: that is a reorganisation recorded as a dividend, and the holding is left out of that period and counted rather than valued on arithmetic that cannot be right.

Every published curve is on that basis, as computed by the pass of 9 Oct 2026 — all 24 of them.

Before this method, both curves were price-only, which understated every list that holds dividend payers and understated the benchmark too. Each list's whole series is recomputed on every nightly pass, from that list's own first live day, so a curve is on one basis over its full length rather than switching part-way.

What these lists are not

A list is a published formula and its output, not a recommendation. The formula is fixed and published; when its weights change that is a new method version, and past scores are never recomputed under new weights, so a score is only comparable with others from the same version. The current version is 4.

Version 4 made five changes to how a company is judged: a short-term company is now ranked on how it has done against the market over a full year rather than over three months, so a single strong quarter no longer stands in for a year of strength; when two short-term companies end up with the same score, the weaker three-month performer comes first and the larger company after it, so the three-month figure settles ties instead of driving the ranking; a company priced in the most expensive tenth of the field on valuation is left out of the list entirely rather than being marked down inside it; and a candidate that moves too closely with the companies a list already holds is passed over for the next one down, so a list of ten is ten positions rather than one position held ten times; and a company whose daily swings are larger than nine tenths of the field enters only when its score is a clear five points above the tenth-place bar, so a list does not take on extra movement for a marginal gain in score. It took effect with the session of 4 Sep 2026; every score written before that carries its own version and is left exactly as it was.

How this score has actually ranked

The formula above is published so it can be argued with. This is the other half: what it has actually done, measured on the stored scores and the sessions that followed them, and re-read every week. Nothing here changes a score.

Read in the week of 5 Oct 2026. A figure is shown only once it rests on at least 20 sessions; below that the cell shows the depth it has instead, because an average over two sessions is a number and not yet an answer. A horizon no session has lived through yet says so, and says how many sessions it still owes.

Of the members in place at a rebalance, the share replaced ranges from 20% to 95% across the lists, measured over 2 rebalances. A list's own page shows its figure.

Back to the lists

The list that was retired

A second list, scored for a shorter horizon and rebalanced weekly, ran from 18 Aug 2026 to 16 Sep 2026. It was retired because its score did not separate winners from losers inside a sector: the inputs it leaned on — trend, momentum and setup — ranked companies in close to the order they did not go on to move. The list above is unchanged by its retirement; no weight, pillar, eligibility rule or rebalance date was altered. Its pages now lead here.

The terms on this page

What is the top 10 score? → · What is percentile rank? → · What is rebalance? → · What is hysteresis (why a list does not churn)? → · What is equal weight? →

Informational only — NOT financial advice.