Every threshold, written down.

The composite

Each category scores 0 to 10. The composite is their weighted mean across the categories that scored, renormalised over that weight, and expressed 0 to 100.

A category with no available input scores nothing rather than zero. A gap in someone else's API is a fact about that API, and scoring it as zero would rank the well-instrumented above the good. Each row shows what share of the weight its composite rests on, and a composite resting on less than 60% is withheld — that is not a rating.

Categories are shown as words rather than numbers — Poor, Weak, Fair, Good, Excellent, each covering two points of the ten. Most of the precision in the underlying numbers is not real: several rest on a hand assessment, and the rest on interpolation between the breakpoints below. The number is on hover for anyone who wants it, and the composite stays numeric because the ranking and the week-over-week change need it.

Capital locked, and why it is not scored

Each row reports total value locked, because the published quantitative work makes it the strongest single predictor of an asset's value — measured against that one asset's own price over time.

It is not turned into a score, because that finding does not survive being applied across different assets. Tested over this week's field, TVL as a share of market cap ranks Bitcoin 23rd of 29 and Monero 27th, while ranking a lending protocol first — against a hand assessment that puts Bitcoin and Monero near the top for real use. The rank correlation is 0.33, which at this sample size is not significant. TVL measures capital parked in contracts: a real fact about a lending market, and a meaningless one about a payments network, a privacy coin or a compute marketplace.

So it is reported, and it is given to the weekly assessment as evidence to weigh where it applies. No arithmetic turns it into a rating.

Cash flow — 25%

Revenue over the last 30 days, annualised, as a share of market cap. Revenue rather than fees: fees include what passes straight through to liquidity providers, which is not the protocol earning anything.

Those breakpoints are spaced by powers of ten rather than evenly, because the field is. Yields run from about 7.5% down to nothing, with the middle of the pack around a hundredth of a percent — four orders of magnitude. Evenly spaced thresholds put two thirds of the assets within a point of zero, which told you almost nothing about the difference between them. The scale is still absolute rather than a ranking against the field: ranking would hand someone a ten in a week when nothing earned, and would move an asset's score because other assets changed.

Four fifths of the score is that level. The last fifth is direction — this 30 days against the 30 before it — kept small because one month of growth is noise and the level is the signal.

Source: DefiLlama, summed across a protocol's versions. A protocol with no DefiLlama adapter is unscored here, not scored zero.

Value accrual — 7%

What share of that revenue reaches token holders — through a burn, a buyback or a fee share — rather than stopping at the treasury or the company. A protocol can earn well while its token has no claim on any of it, and cash flow alone would reward that.

Measured, not assumed: DefiLlama reports holders' revenue and protocol revenue as separate series. A protocol earning nothing is unscored here rather than zero, because scoring it would double-count the cash flow result.

Tokenomics — 15%

Deliberately not “does it have a fixed supply”. What is measured is the net change in circulating supply over twelve months, which is issuance minus burn. A chain with no maximum supply that burns as fast as it mints scores like a fixed-supply coin; a capped coin still releasing 15% a year to insiders gets no credit for the cap.

That is the smallest of the three terms, at a quarter. What is still owed to somebody matters more than what was minted last year: the published quantitative work on this found scheduled unlocks the clearest negative signal on token value and burns the weakest term that reached significance. Supply already issued is history; supply still to come is the part that can still dilute you.

Forty per cent is therefore what is scheduled to unlock over the next twelve months, as a share of what circulates today.

And thirty-five per cent is the ceiling on all future dilution: everything still issuable under the cap, as a share of what circulates now. Without it the score cannot separate two assets issuing at the same rate when one is nearly finished and the other has most of its supply to come — an asset growing 1.4% a year with 4% left to issue ever is in a different position from one growing 2.2% with 67% still ahead of it, and the growth rate alone reports them as near-identical. An asset with no cap is not scored on this term rather than scored zero.

Unlock schedules are hand-researched and done for only a handful of assets so far, so most rows renormalise onto the other two terms. Each row's detail says which terms its score actually rests on.

Under each score the table marks supply. ∞ means no maximum is recorded and more can be minted. A percentage means supply is capped, and is the share of that cap already in circulation — 96% is nearly finished issuing, 60% has most of its supply still to come.

Uncapped assets carry no percentage on purpose. There would be nothing fixed to measure against, so the figure would have to use whatever exists today, and 100% would then mean “nothing is held back for now” rather than “issuance is nearly done” — two different facts printed identically.

Supply change is derived from CoinGecko: market cap divided by price gives circulating supply, and the ratio of that series a year apart gives the change. An asset marked ∞ has no maximum supply recorded, which is not quite the same claim as unlimited issuance. Where no unlock schedule has been researched the category scores on supply change alone and the row says so — an unresearched schedule is unknown, not zero.

Distribution — 10%

Whether the supply was handed to a small group. This is scored from the allocation disclosed at launch — team, insiders and private investors as a share of total supply — and not from a snapshot of the largest wallets. Holder lists are topped by exchange, bridge and staking contracts, and telling those apart from a person is exactly what the paid data services sell. Launch allocation answers the same question from the project's own documents.

Weighted 70% against that and 30% against float — circulating supply over total supply.

Launch allocation is hand-researched and not yet done for every asset. Where it is missing the category falls back to float alone, which is a weak measure by itself, and the row is labelled accordingly.

Builder activity — 22%

Unique commit authors across the project's main repositories over the last 30 days. Authors rather than commits, because commit count is inflated by merges and bots and author count is not.

This is the one category scored against the field rather than against a fixed threshold: the score is the asset's percentile among everything rated that week, times ten. Ties share the midpoint of their block. That percentile is then adjusted by direction — the same window against the equally long one before it.

Both windows are read 5 pages deep at most. A repository busier than that is marked as capped in its row; the cap applies equally to both windows, so the direction comparison stays fair. Projects whose development is not on public GitHub are unscored here.

Not every project publishes its source. Where there is no public repository there are no commits to count, and the asset is marked n/a here rather than scored zero — its rating is spread across the remaining categories instead. Nothing is dropped from the list for being closed source. Where development can be evidenced another way — releases shipped, protocol upgrades, changelogs, engineering roles open — that is assessed instead and the score carries an asterisk, with the evidence in the row. Social mentions are deliberately not used for this: development chatter is trivial to manufacture, and a project with nothing to show produces the most of it.

Only the default branch is counted, so this measures work that has landed rather than work in progress. A project that develops on long-lived branches and merges rarely reads lower than one that merges continuously, even at identical effort. Which repositories represent a project is also a judgement: a deployed immutable contract receives no commits by design, so counting it would report zero for a project under active development elsewhere.

Real usage — 9%

This is the one category that resists a formula. It is assessed against three questions:

• Are fees still paid by people who are not being paid to pay them — that is, does revenue survive the end of an incentive programme?
• Is there non-speculative throughput: stablecoin settlement, tokenised assets, storage or compute consumed, goods actually moved?
• Are there named external counterparties in production, rather than a pilot in a press release?

Nothing in that excludes an industry. A game with real players and real in-game spending answers the first two; a game with four hundred wallets and a token answers none. The measure does not care what the sector is, only whether people pay for it.

Assessed weekly with the evidence recorded in each row. This is judgement, and it is labelled as judgement.

Sentiment — 12%

Social volume and the direction it is moving. It is never weighted above the smallest share on purpose: it is the noisiest input, the easiest to manufacture, and the one where a high reading is as often a warning as a recommendation. It is reported so you can see it, not so it can decide the ranking.

What this does not measure

• Price, momentum, or anything about what an asset might do next.
• Liquidity and market depth, so a high score says nothing about whether a position can be exited.
• Security history, audit coverage, or outstanding exploits.
• Regulatory exposure.
• Whether the revenue being measured is durable, as opposed to one good month.

Not financial advice. These are ratings of measurable properties and not recommendations to buy or sell anything.