How this screener works

Three separate things happen, in order, and they answer three different questions. The criteria decide which companies are good enough to own. The score decides which of those to look at first. The signals decide when to act. A company can be excellent and still show no buy signal for a year — that is the system working, not failing.

The short versionHow this list is built

Eligibility — the seven criteria. A company reaches the shortlist by genuinely failing at most two of these seven: revenue growing at least 10% a year for five years; core EPS (net income with investment gains and losses taken out, per share) positive and growing at least 10% a year for five; net margin (on the same core earnings, investment gains and losses excluded) above 10% in each of five years; shareholders' equity growing at least 10% a year for five; ROC and ROIC both above 10% in each of five years; free cash flow growing at least 10% a year for five; and a market capitalisation of at least $1 billion. Three genuine failures removes the company, however strong it looks elsewhere — but a criterion the data simply can't evaluate (a bank missing a field, a young listing without enough history) is never counted as one of those failures, however many of the seven land there. One test overrides the budget: if the latest revenue, or the latest core net income (confirmed by operating income), is more than 10% below its peak of the previous three years, the company is out, whatever its five-year growth rate says.
Ranking — the score decides order, not membership. Everything that clears the gate is ranked on the 0 to 100 Compounder v2 score: growth, profitability, free cash flow and balance sheet (50 points of arithmetic), durability (10), and 15 for Moat, 15 for Management and 10 for the 15-year structural industry score (Ind 15Y). Every year of the record counts the same, and the score contains no price of any kind — it measures business quality, never value. A company that is not yet rated is left out of the ranking until it is.
Adjustment — cyclical businesses are demoted. The arithmetic cannot distinguish a compounder from a commodity cycle near its peak, since both show rising revenue, widening margins and climbing returns. Each company is therefore reviewed on its ten-year record alone and, where the pattern looks cyclical, moved to the bottom of the list. Semiconductor companies are treated as cyclical unless their revenue has not fallen in any of the last ten years. It is never removed, and the reasoning is shown on hover.
Buying — one ladder, three groups: 2/4, 3/4 and 4/4. Two conditions gate it: the price-to-earnings ratio must sit in the bottom 20% of the company's own five-year P/E range, or below that range altogether; and the price must sit in the bottom 20% of its 52-week range, or at least 35% below its 52-week high. Cheap, and out of favour. Those two alone are 2/4. Volume — 90-day volume at least 1.2× its pre-peak average — lifts a name to 3/4, the buy signal. The monthly MACD cross window on top of that is confirmation and makes 4/4; on its own it does not lift a rung. A company missing either gating condition is in no group at all, and a row reading “waiting” names what is missing.
A second path: dip in an uptrend. Some of the best compounders rise steadily and never fall to the bottom of their 52-week range. They are caught separately when they are cheap against their own five-year P/E range, 10–30% below their 52-week closing high, with the 200-day average still rising and the monthly MACD above its signal line — a pullback inside an uptrend rather than a selloff. Shown in its own Dip in uptrend view.
Selling — holdings only, and either of two raises the flag. Fundamental breakdown, meaning at least two of five ten-year trend tests show the business getting worse (shrinking revenue, earnings or free cash flow, or falling margin or ROIC); or valuation extreme, meaning the P/E stands in the top fifth of its own trailing five-year range, or above it. Technical exhaustion — top 20% of the 52-week range with a bearish monthly MACD — is computed and displayed but deliberately does not fire, because selling on price weakness is the opposite of holding a compounder through a drawdown. The badge is a prompt to look, not an instruction to act.
Your holdings — every account, with a concentration check. Each signed-in account keeps its own portfolio and sees only its own positions. A concentration panel under the table shows how the portfolio's market value splits by stock, sector, industry and country, and warns when one stock is over 20%, one sector over 30%, one industry over 25%, one country over 50%, or semiconductors over 30% in total.
Data and freshness. Fundamentals, analyst estimates and end-of-day prices come from EODHD, refreshed every night (Toronto time); a delayed quote updates the newest price bar about once an hour in between. The page header dates the price separately from the screen run. Figures are in millions of each company's reporting currency unless marked otherwise. Historical P/E is approximated using today's share count against each past year's net income, which ignores buybacks and issuance.

The pipeline, in order: universe (every company of $1B or more on 22 exchanges across North America, Europe and Asia-Pacific — about 12,400 screened) → gate (at most two genuine failures of the seven criteria — up to two criteria with no data are free, each one beyond that counts as a failure — no shrinking-now veto, plus four housekeeping checks) → floors (Ind 15Y of at least 6, a moat rating of at least 6, and a management floor) → rank (the 0–100 score) → demote (cyclicals, semiconductors with a down year, and illiquid names, to the bottom) → sector cap (no sector above 25% of the list) → shortlist → buy timing → sell watch and the concentration check on holdings.

Step 1The gate — the 7 criteria

These are pass/fail. A company reaches the shortlist by genuinely failing at most two of the seven — three removes it. A criterion the data can't evaluate at all (see the note below) is never counted as one of those failures, however many of the seven fall in that bucket: only a criterion that was actually measured and fell short counts against the two. They test whether a business has actually compounded — not whether it is cheap, and not whether it is trending.

The shrinking-now veto (2 Oct 2026). A five-year growth rate only compares the two ends of the window, so a company that tripled early in the window and has been shrinking since can still read +20%/yr. Imeik (300896.SHE) was the case: revenue down 22% and core net income down 44% from their 2024 peak, yet both five-year rates cleared 10% because 2020 was a small base. So the revenue and earnings criteria also check the latest figure against its peak of the previous three years. More than 10% below that peak fails the criterion and removes the company outright — the two-failure budget is for a slow year, not a business in reverse. For earnings, operating income has to have fallen too, so a one-off tax gain in the peak year can't trigger it. A strong business removed this way (moat and management both 8+) shows up under Fallen quality instead.

The management floor (2 Oct 2026). A company whose management team has been rated below 5 of 10 is left off the Shortlist. Two cases are treated differently. When the model can't assess a team — it answers "cannot tell" on alignment, candor and governance, which is common for less-covered names and even for some large ones — the score is mostly the capital-allocation figure from the statements, so it is shown with a ? (for example 5?) and is not held to the floor: not knowing a team is not evidence that it is bad. And any company whose red flags the model itself scored 0 — restatements, fraud allegations, regulatory action — is removed whatever its total, marked ⚠. The red-flag area's computed half (cash conversion and dilution) still lowers the score but does not remove a company, because it fires on almost every recently listed growth company while its share count settles. A company with no rating at all is not ranked until it is rated (see the score), so it cannot reach the list unrated.

Moat and Ind 15Y follow the same "cannot tell" rule (October 2026). The model is given each company's own business description, not just its sector name, and says when it still cannot judge. A moat it could not judge is shown as 5?, and an industry score built on a revenue split it had to guess is shown with a ?; neither is held to its floor. The numbers still count in the score. Ratings are re-made whenever the instructions change, starting with every company the seven criteria admit, so a wrong rating cannot keep a company off the list that would get it corrected.

#	What it asks	The test
C1	Is the top line compounding?	Revenue grew ≥10%/yr for 5 years
C2	Does growth reach the bottom line?	Core EPS positive and growing ≥10%/yr for 5 years. Core = net income minus non-operating items (pretax − operating income − net interest) after tax, ÷ shares, so mark-to-market swings on investments neither pass nor fail a company; interest on debt still counts
C3	Is the business genuinely profitable, not just big?	Net margin above 10% in each of 5 years, on core net income (same definition as C2: investment gains and losses excluded, interest kept)
C4	Is that growth owned, not borrowed?	Shareholders' equity grew ≥10%/yr for 5 years
C5	Does capital earn a high return?	ROC and ROIC both above 10% in each of 5 years
C6	Do the profits arrive as cash?	Free cash flow grew ≥10%/yr for 5 years — or operating cash flow did, in which case the FCF shortfall is spending on growth (new plants, an acquisition), not a weaker business, and the test passes with a note saying so
C7	Is it big enough to be tradable?	Market cap ≥ $1B

Every growth test now uses five years (revised 30 Sep 2026). Revenue and equity were judged over ten years, and total-asset growth was a separate criterion; asset growth has been removed — a capital-light compounder grows assets slowly by design — and revenue and equity now use the same five-year window as EPS and FCF. The ≥10%/yr bar is unchanged.

Young companies are judged on the years they have (revised 6 Oct 2026). A recent listing with fewer than five years of figures on file is measured over the longest window it does have, down to three years — the company card shows it as “(3 yrs)” or “(4 yrs)” — so it passes or fails on real numbers instead of reading “no data”. This only applies when the full window was never on file: a company that has five years is always judged on five, so a real five-year failure is never rescued by a shorter one. And “no data” is no longer unlimited: up to two criteria may be not evaluable (a bank without a classified balance sheet, say) at no cost; each one beyond two counts as a failure against the two-failure budget. Before this change a company could reach the Shortlist passing just one of the seven tests.

Four further checks keep the list honest, and they reject far more names than you would expect:

Funds, trusts, preferred shares, BDCs and CDRs are filtered out by name pattern, with a manual exclusion list on the server for the ones that read like operating companies.

Step 2The score — Compounder v2

The score does not decide who gets in. It decides the order of the companies that already got in. There is one score, Compounder v2, and it runs from 0 to 100 (since 13 Oct 2026 — the earlier 70 / 10 / 10 / 10 weighted score was removed, so every Score column, tooltip and detail card shows this number). It is quality only: the score contains no price. Eight parts:

#	Part	Points	Made of
1	Growth	15	Revenue growth 7.5 (full marks at 25%/yr) and per-share core earnings growth 7.5 (full marks at 30%/yr)
2	Profitability	15	Operating margin 5 (full at 30%), return on invested capital 6 (full at 25%, five-year average), core net margin 4 (full at 30%)
3	Free cash flow	10	FCF margin 6 (full at 33%) and five-year cash conversion 4 (FCF as a share of core earnings: 0 points at 40%, full at 100%)
4	Balance sheet	10	Net debt over operating profit 5 (full marks with no net debt, zero at 4×) and interest cover 5 (zero at 2×, full at 10×). Banks, insurers and real estate get a neutral 5 — debt ratios mean something else for them
5	Moat	15	The 0–10 moat rating × 1.5
6	Management	15	The 0–10 management score × 1.5 (capital allocation, alignment, candor, governance, red flags)
7	Ind 15Y	10	The 15-year structural industry score: 0 points at 4 or below, full 10 at 9 (6 earns 4, 7 earns 6, 8 earns 8)
8	Durability	10	Share of years with growth 3.5, worst revenue year 2.5 (0 points at a 20% fall, full if it never fell), analyst forecasts 4 (revenue and EPS, 2 each)

Forty of the 100 points are judgement (moat, management, Ind 15Y) and sixty are arithmetic from the statements. Revenue, earnings, margins, cash flow and returns average the whole record with equal weight per year, and growth is scored on the lower of that long average and the rate over the last three years: a recent boom earns no more than the long average, a recent decline pulls the score down. Earnings are core net income — investment gains and losses removed, interest kept — the same earnings the gate judges.

Safeguards built into the score

Bands

Bands are set from the real distribution of Shortlist scores (13 Oct 2026), not from round numbers: exceptional at 80 or more (roughly the top 6% of the Shortlist), strong at 72 or more (the top quarter), acceptable at 66 or more (the median and above). Below 66 a company has a score but no band. The numbers are fixed, so they will be re-checked as unrated companies get rated and the distribution shifts.

Emerging and Venture companies also carry a separate early-stage class score on the Early-stage classes page and the detail card. That is a different model, built for companies without ten years of profits, and its number is not comparable with the Compounder v2 score.

Step 3The cyclical check

The arithmetic cannot tell a compounder from a commodity cycle near its peak — both show rising revenue, widening margins and climbing ROE. So each gated company is shown a ten-year table of revenue, net income, margin, ROE and free cash flow and classified as durable, cyclical, one-off or unclear.

The classifier is deliberately not told the score, the criteria verdict or anything about the share price — telling it the screen already likes a company only invites agreement. Companies judged cyclical are moved to the bottom of the list, never removed, with the one-line reasoning on hover. You can disagree with any call and the company is still there.

Semiconductors are cyclical by rule (added 1 Oct 2026). Chip demand runs in boom-and-bust cycles, so a company whose industry is semiconductors or semiconductor equipment is treated as cyclical — whatever the classifier said — unless its revenue has not fallen by more than 2% in any of the last ten years, i.e. it has already grown through a downturn. Fewer than five years of history counts as not proven. The hover note names the year and size of the drop.

Verdicts are cached and re-asked only when the fiscal year moves or the call is over 30 days old. If no key is configured the step is skipped and ranking behaves exactly as it did before.

AlongsideThe 15-year structural industry score

One question only: how likely is the economic activity this company takes part in to consume a larger share of global GDP around 2041? It ignores valuation, the company's own financials, moat and management — a great industry can still hold a bad investment. It is shown beside Moat and Mgmt as Ind 15Y. It does two jobs: it is worth 10 of the 100 score points — 0 points at 4 or below rising to the full 10 at 9, so 6 earns 4, 7 earns 6 and 8 earns 8 — and it sets a soft floor for the shortlist (see the note below).

Each economic activity (AI compute, grid equipment, pharma, banks, oil & gas and about fifty more) is scored 0–10 on ten factors. Each factor score is multiplied by its weight and the total is divided by 10, so the result stays on a 0–10 scale.

#	Factor	Weight	Main question
1	Structural demand inevitability	1.5	Will the economy almost be forced to consume more of it?
2	GDP-share expansion potential	1.5	Can spending compound faster than global GDP for 15 years?
3	Penetration runway	1.0	Is adoption still far below plausible saturation?
4	Technology tailwind	1.0	Does better technology increase demand, or make it obsolete?
5	Demographic tailwind	1.0	Do aging, urbanization and labor shortages raise demand?
6	Bottleneck / scarcity value	1.0	Do other megatrends need it to operate?
7	Replacement / recurring necessity	0.75	Does the spending keep repeating?
8	Emerging-market runway	0.75	Will billions of people getting richer consume much more of it?
9	Substitution resilience	0.75	Could the activity itself — not one company — be replaced?
10	15-year visibility	0.75	How confident can we be the thesis still holds in 2041?

From industry to company. A company is rarely one industry, so its revenue is split across up to four activities, and the cell shows two numbers:

Reading it: 9–10 exceptional megatrend · 8–8.9 very strong tailwind · 7–7.9 strong structural growth · 6–6.9 positive with material uncertainty · 5–5.9 roughly GDP-like · 4–4.9 weak · below 4 structural headwinds or decline. Green from 7.5. Hover any cell for the revenue split and the reasoning. The revenue split comes from the same classifier as the moat rating, is judged from the company's own description of what it sells, and refreshes on the same cycle; the industry scores are a curated judgement table, and a change to it re-scores every company at the next publish.

Step 4The buy signal

Timing, not selection. It runs on shortlist candidates and on your holdings, and reads identically in both tables. Four conditions are read. The first two gate it and both must be true at once; the other two rank what is already through the gate — they are marked ranks below. The count of the four is the buy group, and the sidebar carries one view per group.

Group	Means	Read it as
BUY 2/4	Cheap and near the low; volume not yet above 1.2×	The entry rung. The buy case is valuation — the market has not moved yet.
BUY 3/4	The gate plus volume	The buy signal. Valuation with participation behind the move.
BUY 4/4	3/4 plus the monthly MACD turn	Confirmation, and the rarest rung: near-low and the MACD turn pull against each other, so this list is often short or empty. Rare, not required.
DIP IN UPTREND	Cheap, 10–30% off the 52-week high, 200-day average rising, monthly MACD above signal	The second path, for compounders that rarely reach their 52-week low. Not a rung of the ladder.
A pale count, or “waiting…”	A gating condition is missing	No group. A stock near its low on heavy volume but still expensive is not a buy here.

The three sidebar views nest. Buy 2/4 is the whole buy list — every name through the gate, with the 3/4 and 4/4 names in it too; Buy 3/4 narrows that to the ones with volume behind them — the buy signal; Buy 4/4 to those with the MACD turn as well. So the counts read as “55 through the gate, 20 of them with volume”, not as three separate piles. The badge on a row is always that row’s own rung, whichever view it is being read in, and the Buy signal column sorts by it.

One row per company, US listing first. A company listed in several places gets one row. Since October 2026 that row shows its US listing whenever the US line really trades — at least $5M a day over the last 60 sessions — so Novo Nordisk appears as NVO.US, AstraZeneca as AZN.US. The price, 52-week range, volume and MACD legs of the buy signal are then read on the US line, the one you would actually buy, while the gate, P/E, P/E band and every fundamental come from the home listing (NOVO-B.CO), so the ADR share ratio cannot distort them; the row notes “home listing NOVO-B.CO”. A thin over-the-counter line (PMCUF for Pro Medicus trades about $12k a day) never takes over: there the row stays on the home listing. If you hold a different listing, the buy views show “you hold …” under the ticker; searching for any listing says where the company sits on the Shortlist.

Fallen quality is a separate view, not a rung of the ladder. It lists companies the gate turns away because recent growth has stalled, but which have the record and the business the gate exists to find: revenue up at least 10%/yr over ten years, moat and management both rated 8 or more, margin and ROC/ROIC still passing — and which are cheap against their own five-year P/E range today. Each row names the tests it fails. A low price on a business like that is the market betting the slowdown lasts; the view keeps those names in sight so you can see if growth comes back, and it is labelled turnaround risk because that is what it is.

Reading the badge

Hovering the cell spells out all four readings with their actual numbers — where in the 52-week range the price sits, the volume ratio, where today's P/E sits in the five-year range, and how many months ago the monthly MACD crossed up.

Condition	Passes when	The idea
Cheap miss reads "not cheap"	Today's P/E sits in the bottom 20% of the company's own five-year P/E range — or below that range entirely, cheaper than it has been in five years	Cheap against its own history, not against other companies. The range is built from each fiscal year's highest and lowest traded price against that year's earnings, so it sees the lows that a year-end snapshot misses.
Near the low miss reads "off low"	Price sits in the bottom 20% of its 52-week range, or at least 35% below its 52-week high	Buying into weakness, not strength. Two readings of "out of favour" that disagree more than you would expect: position-in-range says on the floor, drawdown says far from the peak. On 2026-09-07, 20 cheap names sat 30%+ below their high while above the bottom fifth of their range — excluded only for having bounced. Either branch passes.
MACD turn ranks miss reads "no MACD turn"	The monthly MACD is inside its bearish-to-bullish cross window: it crossed up within the last 2 months. A real cross only (2 Oct 2026): a MACD that is still below its signal line but closing the gap shows as approaching on the row, but no longer passes the leg	Don't catch a falling knife. The month is slow enough not to be noise and fast enough that the cross still has the move ahead of it. Daily is too twitchy to time on; weekly and quarterly were both tried as extra conditions and dropped.
Volume ranks miss reads "no volume"	90-day average volume ≥ 1.2× the average of the year before the 52-week high	Other people are showing up too — sellers giving up and buyers absorbing them. Measured against what was normal before the selloff: the year leading up to the 52-week high. Changed 2026-10-02 from the trailing twelve months, because a long selloff inflated its own yardstick — after a year of heavy selling, the past year's average is itself heavy, and the leg went quiet exactly when volume was several times its old normal. (A five-year base, used before 2026-09-07, had the opposite problem: liquidity grows over five years, so everything cleared 1.2×.) Backtested 2016–2025 on near-low days: one year later the median name was up 20% with this leg on, 17.5% with the old one, 13% with no volume test (shortlist names, 2020 excluded); the out-of-sample Trend names agreed. A one-day volume spike on its own was tested too and did no better than no volume test at all.

Two more valuation bands are shown wherever P/E is, on the same five-year basis: P/B (market capitalisation over book value of equity) and D/E (total short- and long-term debt over equity). Both are display only — they feed no score and no buy or sell condition. They are there so a cheap P/E can be read against what the balance sheet is doing.

MACD is computed on the daily, weekly, monthly and quarterly timeframes and all four are shown on the detail card. The monthly is the one that is read — for the buy-side leg that separates 3/4 and 4/4 from 2/4, and for the sell-side technical-exhaustion check. Daily, weekly and quarterly are shown and nothing more.

Why the monthly leg is a window rather than a state: a monthly MACD that turned bullish two years ago tells you the move already happened. Requiring the cross itself — the two months either side of it — keeps the signal pointed at the turn, which is the part of the move worth buying.

Step 5The sell signal

Runs on holdings only, and it is tracked independently of whether the company still passes the criteria — a stock you own is watched whether or not it would be bought today. Unlike the buy signal, any one condition raises a flag; the badge shows how many fired, e.g. SELL WATCH (1/3). Three are computed, but only two of them fire.

A flag is a prompt to look, not an instruction. One firing on a long-term holding is usually noise; both together, on a position you already had doubts about, is the case worth sitting down with.

Condition	Fires when	What it is telling you
Technical exhaustion does not fire	Price in the top 20% of the 52-week range and monthly MACD has turned bearish	Shown as context only. It is deliberately switched off as a trigger: selling on price weakness is the opposite of holding a compounder through a drawdown
Fundamental breakdown	Two or more of five trend tests, read off the ten-year series: revenue shrinking over 3 years; earnings shrinking, or a loss in the latest year; free cash flow shrinking while operating cash flow shrinks too (a dip from heavy capex is investment, not decay); net margin at least 3 points below the prior five years; ROIC (median) at least 3 points below the prior five years and now under 15%. One firing is context, not a watch.	The reason you bought it has changed. This is a trend, not a level, and that is the whole distinction: the seven criteria ask would I buy this today, which a merely slow compounder fails for years while a collapsing one looks identical. Deterioration asks whether the recent years are worse than the years before them. PayPal at +4.7% revenue and +16% FCF is slow, not decaying, and raises nothing; Intel, loss-making with ROIC at −2.1% against 10.4% before, raises three flags. Earlier versions of this check compared screen runs instead, and could not work: the run history is weeks deep and moves whenever a rule or a data source here changes, so it dated every "break" to whichever day we shipped something. It also fired on asset and equity growth, which a capital-light compounder fails by design — Qualys was flagged for growing assets ~8% while earning 60% ROIC, and buybacks shrink equity on purpose. Nothing about the current test touches run history: the verdict is the same on the first run as on the hundredth.
Valuation extreme	P/E in the top 20% of its own ~5-year range, or above that range entirely	You are being paid a price the business has rarely commanded. The mirror of the buy side's cheap at the bottom 20%. Firing only above the range made this unreachable in practice: measured across the 20 holdings, the richest sat at 37% of its own band and the median at 8%.

PortfolioYour holdings and the concentration check

Every account has its own portfolio. Positions are added, edited and removed in the editor on the Your holdings card, with an optional purchase history (lots) behind each ticker. Nobody sees anyone else's positions; the nightly run fetches data for every ticker any account holds, so a newly added ticker is priced within about fifteen minutes or, at the latest, after the next night's update.

The concentration panel (added 1 Oct 2026) sits under the holdings table and measures the portfolio by market value in US dollars. It shows the largest positions, and the split by sector, industry and country, and raises a warning when any of these limits is passed:

Check	Limit	Why it matters
One stock	20%	A single bad surprise should not set the whole portfolio back
One sector	30%	Sectors re-rate together on rates, regulation and the cycle
One industry	25%	Companies selling the same thing share the same demand cycle
One country	50%	Currency, policy and market risk move every company based there at once. For mainland China and Hong Kong the warning also names policy, sanctions and accounting-verification risk
Semiconductors in total	30%	Chip demand is cyclical, so chip companies tend to fall together in a downturn

Country means where the company is based, not where the share is listed — TSMC's US listing counts as Taiwan. Where a holding has no company profile yet, the exchange in its ticker is used instead (.SHG counts as China). Positions with no price yet are listed but not counted. The panel is a prompt to rebalance, not a rule the screen enforces: the shortlist itself is not capped by sector or country.