Ten pillars, — checks. A pillar's reading is the share of its checks sitting in the risky band — nothing is weighted, nothing is summed into a score.
Every one of the — checks, grouped by pillar and ordered worst first. The fraction beside each row is exactly what the radar plots as that pillar's wedge.
Three rules, and they are the whole model.
Each check names one measured number — a volatility, a ratio, a filing count — and gives it one of three states. Never an opinion, and always with its source attached.
Rank asks where the number sits among — investable peers — names passing a tradeability floor of $5m a day, $5 a share and $300m of market value. Level asks whether it crosses a cut solved from realised forward drawdown inside that same cohort. The check takes the harsher answer and says which one fired.
Why not the whole market. Ranked against all — listed names, most of them micro-caps, NVIDIA's 47% annualised volatility sits in the 34th percentile and reads as calm. Among investable peers it is the 57th. The first number is true and useless; the second is the one a fund would act on. Both are shown on every check that differs.
Every pillar aims to hold five checks, so a shape on the radar means the same thing on every axis. The counts are never added together: we do not claim to know the exchange rate between a leverage flag and a tail flag, and inventing one would be the whole product's central lie.
Every pillar owns one sector, and every check owns one ring inside it. One cell per check, worst nearest the centre. The wheel is always completely filled, so the total area is identical for every company and carries no meaning at all — only the mix of colour does. Read it the way you would read the square grid below, because it is the same grid wrapped into a circle.
Why area was retired as the signal. An earlier version drew a filled wedge whose radius was the flag count, with a paler wedge behind it for the checks on watch. That made a pillar with nothing flagged draw nothing at all — the best state was the least visible — and it let a pillar with five checks on watch reach the rim while its label read 0/5. Radius meant two different things on the two layers. Constant area with a varying mix removes the ambiguity entirely.
The radar has one flaw worth knowing. Its silhouette depends on the ORDER of the axes, and that order is arbitrary — swap two pillars and identical data draws a different shape. It is kept because it is the fastest thing on the page to read, and it is paired with the square grid under "Pillar readings", which encodes the same checks with no such freedom. When the two disagree, trust the grid.
Why exactly five checks in every pillar. So that one axis of the radar means the same as another. Getting there took removing a genuine duplicate rather than padding: Omega equals 1 + gain-to-pain by identity — verified to the last decimal on every name here — so counting both would have registered one fact twice. The rule is that uniformity must be reachable with real evidence; if a pillar could not honestly reach five, it would show four and say so rather than invent a fifth.
All — checks on the percentile axis, grouped by pillar. The shaded band on the right is where a check is flagged on rank; checks decided by a level cut instead are drawn hollow.
Six years of daily closes. The underwater curve is the same series expressed as distance from its own running peak — exactly what the drawdown pillar measures.
Volatility is a single number standing in for the two pictures below: the level through time, and the distribution of daily moves it summarises.
The record every percentile on this page is computed from.
Five of these ratios are checks in the efficiency pillar. The rest are carried as context — informative, but too correlated with each other to count separately.
Each pillar states what it measures, why it is separate, and every check inside it with the number, where it sits, which test fired and what it came from. Click any pillar header to fold it away.
Whether the flags above belong to this company or to everything around it.
The two heaviest inputs, measured monthly since 2011, each against the median of the whole universe in the same month. These are the quantities themselves — not a score built from them — so the comparison stays checkable.
Width, not direction.
The model has no view on direction. A Monte-Carlo fan drawn from the current volatility regime with fat tails. The middle line is not a target — it is the absence of an opinion. Only the width carries information.
The limits are part of the product, not a disclaimer bolted to the end.
It does not answer whether to buy. Every figure measures how far a name can fall, not which way it will go. Flagged checks are not a sell and clean checks are not a buy.
It is not a timing signal. Tested directly: when a name's own risk reading peaked it went on to return +15% over the next six months; when it bottomed, −9.9%. The intuitive reading is backwards, so we do not offer it.
There is no forecast of the business. No analyst estimates, no price target, no view on demand for what the company sells. Everything here is measured from filings and prices that already exist.
The checks are not equally important, and we do not claim to know the ranking. That is why there is no combined score: any single number would be our weighting imposed on your decision.
Every source below is free and public, and every check on this page names the one it came from.
How current the data is is not filed away here — it is a check like any other, inside Disclosure & data trust, because staleness is a fact about the company's reporting rather than a footnote about our method.