The three are never averaged. A low-risk strategy can run on a protocol rated caution, and a protocol with no signal at all can be rated avoid.
Every spot in the monthly picks carries one of four levels. The level rates the strategy: its leverage, its liquidity and how new its mechanics are.
These levels rate the strategy. The protocol's own risk is separate: the Solid / Caution / Avoid verdict in our risk report.
Every protocol we track has a dated risk report. The report scores seven criteria, each green, yellow or red with a line of reasoning, and the verdict follows from the worst of them.
Verdict is a gate (worst flaw wins), not an average. Our read, not financial advice.
One red on a criterion where funds can be lost caps the verdict at caution or avoid, however green the rest is. An average would hide exactly the flaw that matters.
Each report names its own avoid triggers: the events that would move it to avoid.
Reports are dated and refreshed monthly. A fact we could not verify on the research date is marked in the report, and the next refresh checks it again.
Between refreshes a check runs hourly. A report gets the “review” chip in the risk and report tables when a hack lands after its research date, or when its TVL falls 15% in 24 hours or 30% in 7 days after it, for protocols over $1M. Read that verdict as possibly out of date until the next refresh.
A decentralised perpetual-futures exchange is rated no lower than caution. These exchanges run on a different custody and settlement model, and avoid would read as “never touch” for venues we trade on ourselves. avoid is kept for an exchange with an exploit or a user loss that was not repaid. When this floor lifts a verdict, the report says so and keeps every red that would have set it lower.
No numeric scores: a 3 against a 4 is false precision, and an average hides the one flaw that matters. No community ratings: a verdict is our own diligence, with a date on it.
An exception is a verdict we keep although one of the report's own avoid triggers has tripped. It is the curator's call, stated at the top of the report, and decided again at the next monthly refresh. This list is built from the reports themselves.
| Protocol | Verdict held | What tripped | Report dated |
|---|---|---|---|
| Mellow | caution | One of the three avoid triggers named in August is still tripped. In dollars earnETH's share of the protocol fell from 74.4% to 73.4%, but only because ETH fell: in ETH earnETH grew from 83,294 to 90,987, and its routing into strETH rose from about 99.4% to about 99.8%. The other two did not trip: no upgrade was executed and no exit was frozen between 2 October and 7 November. Caution is held as the curator's call of 7 November 2026 and is reviewed again at the December refresh. | Nov 7, 2026 |
The perp-exchange floor above is a standing rule, not an exception, so it is not listed here.
The red / yellow / green signal is an auto heuristic, not a rating. It is set from the same facts shown in the table below:
Venues (CEX / perp-DEX / chain) sit in the same list; their signal fires only on a recent incident, and scanner facts like TVL or audits don't apply to them (shown as dashes). Facts auto-refresh daily from DeFiLlama; oracle & venue notes are curated. Always verify at the venue / protocol before acting.
Whether a named party can freeze or seize an individual holder's balance - as opposed to pausing the protocol for everyone - is covered inside each protocol's report rather than flagged here. It is a real risk, but it does not feed this signal.
The signal never changes a verdict by itself. A hack or a sharp TVL drop after a report's date sets the review chip described above, and the verdict waits for the next refresh.
“Refresh” is how often a scanner asks its source. The time a table prints, “data as of …”, is when its data was actually measured, and that is the one to trust.
| Table | Source | Refresh | When a reading is late |
|---|---|---|---|
| Portfolio | Pacifica, Solana RPC, Pendle | every 5 min | marked stale after 30 min |
| Stables · peg | DeFiLlama stablecoins | every 5 min | marked stale after 60 min |
| Stables · yields | DeFiLlama yield pools | hourly | no value |
| Stables · 7-day average | DeFiLlama pool history | every 4 h | no value |
| Stables · market | Artemis | daily | no value |
| Yield pools | DeFiLlama yield pools | hourly | marked stale after 3 h |
| Pendle PT markets | Pendle API | hourly | no value |
| Lending · markets | DeFiLlama lending | hourly | no value |
| Lending · vaults | Morpho, Euler, Kamino, Spark | hourly | hidden after 3 h |
| Funding | Each exchange's API | every 30 min | leaves the table after 2 days |
| Funding · exit costs | Order books and swap quotes | hourly | no quote after 3 h |
| Vaults | Exchange APIs and DeFiLlama | every 15 min | left out of totals after 3 h |
| Risk · facts | DeFiLlama protocols | daily | marked stale on a report after 2 days |
| Risk · hacks | DeFiLlama hacks | hourly | no value |
| Review chip | Our reports, hacks and TVL history | hourly | no value |
| Picks and reports | Our desk | by hand, monthly | greyed after 30 days |