How the risk ratings work
Every rating is built the same way: a fact is sourced, a fixed rule turns that fact into a 1–5 score, and scores roll up into weighted categories. No score is ever set by hand or by feel — within an asset class the same rule applies to every asset, so two assets with the same facts always get the same score. Volatility thresholds are the one place the classes differ: they are calibrated per class (crypto / equity / commodity), because 50% annualised volatility is unremarkable for a token and near the top of the S&P 500. The table is chosen by class, never per asset.
What the bands mean
- Low Risk (under 1.75) — rated factors look benign: steady pricing, deep liquidity, no obvious structural red flags.
- Medium-Low Risk (1.75–2.25) — mostly sound, with one or two areas worth watching.
- Medium-High Risk (2.25–2.75) — real but ordinary risk, with some factors scoring poorly.
- High Risk (2.75–3.25) — several rated factors score badly; materially riskier than a typical holding.
- Extreme Risk (3.25 and above) — poor across the board; treat any position as speculative.
These boundaries are deliberately not spread evenly across the 1–5 factor scale, because composite scores never occupy that range. A composite averages many factors, so reaching 4+ would need almost every factor to be terrible — which essentially never happens for a real listed company. Splitting 1–5 evenly would leave the top two bands permanently empty and dump most assets into one bucket. The boundaries sit where the distribution actually lies — but they are still absolute, not percentiles, so adding worse assets later fills the upper bands rather than relabelling everyone else. The dashed lines below are those boundaries against the live distribution:
The current rated universe: 615 banded assets spanning 1.18 to 3.72, median 2.29 — computed from the same risk files every page reads, so it moves with each refresh.
Why some assets show “Partial” instead of a band
A factor with no researched fact behind it is left unrated and dropped from the calculation entirely, rather than filled in with a guess or a neutral placeholder. The remaining weights are renormalised so the score reflects only what is actually known.
An overall band is only assigned once at least 60% of the weighting is fact-backed and the Market / Volatility category is rated. That second condition matters for newly-listed assets: a token with only a few months of history can clear the coverage threshold on structural factors alone and be labelled low risk without its price risk ever having been measured.
This matters more than it sounds. Rate a stablecoin on volatility and liquidity alone and it looks flawless — both are excellent by construction — while everything that actually carries its risk (what backs it, who holds the reserves, the regulatory position) sits unrated. Showing a confident “Low Risk” there would be asserting the unknown half resembles the known half. It usually doesn’t.
Crypto factors
Market / Volatility
30% of scoreAnnualised volatility over three years, worst peak-to-trough drawdown, and beta against Bitcoin.
How violently the price moves, and how much of that is the whole market moving versus this asset specifically.
Automated — derived from monthly price history.
Liquidity
10% of scoreDaily volume against market cap, share of volume on the single largest exchange, and number of meaningful listings.
Whether you can actually get out at size. Thin volume concentrated on one exchange is the risk that bites when you most need to sell.
Automated — CoinGecko market and exchange data.
Holder Concentration
15% of scoreShare of supply held by the top 10 wallets, and tokens unlocking in the next six months.
A few holders able to exit at once, or a large scheduled unlock, is supply overhang the price hasn't absorbed yet.
Manual research.
Smart Contract / Technical
12% of scoreAudit status and recency, whether an admin key can change the contract, and for proof-of-work chains the share of its mining algorithm that secures it.
Code and consensus risk. An unaudited contract, or one a single key can rewrite, can lose funds regardless of price. The hashrate share matters because a minority user of an algorithm a larger chain also uses can be attacked with rented hashpower — Ethereum Classic was attacked repeatedly this way, and is safe now only because the larger chain left the algorithm.
Manual research.
Custody / Bridge
10% of scoreWhether the asset depends on a bridge or a centralised issuer honouring redemption.
Bridges are among the most exploited things in crypto, and a wrapped asset is only as good as its backing.
Manual research.
Governance / Team
10% of scoreWhether the team is identifiable and has a track record, and how centralised governance is.
Anonymous teams with unilateral control are the precondition for most rug pulls.
Manual research.
Regulatory
10% of scorePre-mine and ICO history, centralised issuance, and any enforcement action or delisting.
Regulatory reclassification can remove access to exchanges regardless of how the technology performs.
Manual research.
Systemic / Correlation
3% of scoreCorrelation to Bitcoin.
Tells you how much diversification you're actually getting. Highly correlated holdings fall together.
Automated — derived from price history.
Equity factors
Market / Volatility
20% of scoreAnnualised volatility over three years, maximum drawdown, and beta against the S&P 500.
How much the share price swings, and how much of that is simply the market moving.
Automated — derived from monthly price history.
Liquidity
15% of scoreAverage daily traded value and the free float.
Whether a position can be exited without moving the price against you.
Automated — volume from price history; float is manual.
Leverage / Financial
20% of scoreBorrowings against shareholders' equity, and how many times operating profit covers interest — scored on sector-aware thresholds.
Debt is what turns a bad year into an existential one. Thresholds vary by sector because leverage means different things in different businesses: a regulated utility funding long-lived assets against rate-set cash flows is not the same as a software company at the same ratio, and for a bank borrowing is the product.
Automated — SEC EDGAR XBRL filings (US-listed only).
Earnings Quality
15% of scoreVolatility of year-on-year earnings per share.
Erratic or negative earnings make valuation guesswork and tend to precede unpleasant surprises.
Automated — SEC EDGAR XBRL filings (US-listed only).
Concentration
10% of scoreRevenue dependence on the largest customer, and geographic concentration.
A business leaning on one customer or one volatile region carries risk that the share price often doesn't reflect until it lands.
Manual research (10-K filings).
Governance
10% of scoreBoard independence, share-class structure, and litigation or restatement history.
Weak oversight and dual-class control mean minority holders have little recourse when management errs.
Manual research (proxy statements).
Sector / Regulatory
10% of scoreBaseline regulatory exposure of the company's sector.
Utilities and financials face policy risk that tech generally doesn't, independent of how well the business is run.
Automated — fixed lookup from the company's GICS sector.
Known limitations
- Crypto’s manual factors — audits, admin keys, team, bridges, regulatory — are researched by hand, so most coins sit around 40% coverage and show a partial score rather than a band.
- UK and EU companies aren’t SEC filers, so their leverage and earnings figures come from a different source than the US set’s (EDGAR). Interest coverage is deliberately left unrated for banks and insurers in every region — their interest expense is the cost of funding deposits, the business itself, not a burden earnings have to clear.
- Research coverage is uneven between regions, and that tilts scores. Governance and concentration facts have been researched for most UK and EU names and almost no US ones. Because an unrated factor is dropped rather than guessed, checking a factor a company scores well on lowers its composite — so a researched company can read as safer than an identical unresearched one. Rankings within a region are sound; comparisons across regions are provisional until the US set is researched to the same depth.
- Some real risks still have no factor and are noted in each asset’s fact file instead — LEO’s dependence on its issuer funding a buyback out of Tether profits, and discretionary (rather than scheduled) token releases at Chainlink and Stellar, which the unlock factor cannot see.
- Thresholds in the rule tables are considered judgement calls, not settled science. They’re visible in the repo and open to revision.
- Prices are shown in each asset’s native currency and never converted, so returns across currencies aren’t directly comparable.
Ratings summarise sourced facts against fixed rules. They are not investment advice, and no substitute for your own diligence — particularly on anything you intend to size meaningfully.