Methodology

The criteria are published first

The criteria by which every venue is categorised are published and versioned before any venue is measured against them, applied evenly to every row, and changed only through a changelog entry with a source. That order is the point: criteria written after the fact can be shaped to a conclusion, and criteria that change silently are a re-scoring nobody can audit.

Three pillars, no fourth

Price — what credit costs here, and whether the rate is discovered or set.
Quality — what stands behind a position, and what can change under the reader.
Composition — who and what makes up this market.

One figure per pillar, on every page

Every class, protocol family and venue page opens with the same three figures. The question is the same everywhere; the read that answers it differs by kind of lender, and each figure names its read and its kind — a posted ceiling, an estimate, a figure with no lien behind it — on the figure itself. None is a score or a sort key.

price: The rate a borrowed dollar pays, weighted by dollars. Pools: the 30-day trailing, debt-weighted USD borrow rate, continuously compounded — the same pools, gates and floor as the class curve, so a class's venues sum back to it exactly. Auction: the 30-day cleared rate. Desks: the current posted card, which is administered, not cleared. Corporate: the holder's price-implied yield — there is no borrow leg.
quality: Debt divided by the value of the bitcoin behind it. Observed where a read carries both sides. A desk's posted maximum LTV is a ceiling it sets, not an observation. Corporate debt has no lien on the issuer's bitcoin, so its figure is registered debt over bitcoin held — an analogue, not an LTV.
composition: The dollar debt tied to bitcoin, and how it is tied: exact (a market whose collateral is bitcoin), pool (lent from a reserve that holds bitcoin among other collateral — no dollar is assignable to it), estimate (a desk's book, often a bounded range), or filed (an issuer's registered debt, financed by bitcoin rather than backed by it).

The ten criteria, in full

These are the ten. Every venue in the registry is asked all of them, in the same words, whatever its class — and the n / 10 on a venue's page counts how many have been established from a public source. The values below are the ones actually observed across the 135 venues, with how many venues take each.

Pillar	Criterion	What it asks	Established
Price	Rate formation	How is the rate arrived at — discovered by bidding, set by a formula reading utilisation, or administered by whoever runs the venue? Observed: algorithmic utilisation (98 venues) · posted card (20 venues) · administered (10 venues) · bilateral unpublished (4 venues) · administered price feedback (1 venue) · borrower set (1 venue) · auction discovered (1 venue)	135 / 135
	Term structure	Is there a term at all? A fixed maturity agreed at origination is a different product from an open-ended position that can be called or repriced. Observed: open ended (83 venues) · fixed term (16 venues) · undisclosed (3 venues) · callable (2 venues) · not applicable (2 venues)	106 / 135
	Rate certainty	Once you are in, can the rate move under you before you are out? Observed: variable (81 venues) · fixed at origination (18 venues) · undisclosed (5 venues) · not applicable (2 venues)	106 / 135
Quality	Oracle dependency	What price feed decides whether you are liquidated, and how many independent sources stand behind it? Observed: multiple (52 venues) · single (25 venues) · undisclosed (21 venues) · not applicable (5 venues) · none (3 venues)	106 / 135
	Default & liquidation	What actually happens when collateral falls — an auction, a threshold seizure, a margin call, or a negotiation? Observed: threshold (91 venues) · not applicable (5 venues) · soft (4 venues) · undisclosed (3 venues) · none (2 venues) · discretionary (1 venue)	106 / 135
	Custody & backing	Who holds the bitcoin while the loan is outstanding, and what could they do with it? Observed: protocol custodial (76 venues) · third party custodial (14 venues) · custodial with reuse (6 venues) · not applicable (5 venues) · undisclosed (4 venues) · shared key (1 venue)	106 / 135
	Recourse	If the collateral does not cover the debt, does the lender have any claim on you beyond it? Observed: non recourse (76 venues) · undisclosed (18 venues) · recourse (10 venues) · not applicable (2 venues)	106 / 135
	Track record	How long has it been running, and what has gone wrong in that time — exploits, pauses, freezes? The months count how long this dataset has watched the venue, so they are a floor on its age rather than a claim about what came before. Observed: observed (127 venues)	127 / 135
Composition	Collateral accepted	What is actually accepted as collateral — which wrapper, and is it the bitcoin you think it is? Observed as a per-venue list of wrappers, not a term from a closed set — a wrapper is a distinct counterparty exposure, not a label.	106 / 135
	Book depth & concentration	How much is actually outstanding here, spread across how many distinct counterparties — and how much of it is the largest one, or the operator trading on its own account? Observed: disclosed (4 venues)	4 / 135

How to read an empty cell. Every filled cell carries the contract, document or filing that establishes it; a value without a source is left empty. A cell without an answer is one of four different things, and none is shown as another:

not applicable — the question does not apply to this kind of venue (a perpetual funding rate has no loan, so it has no liquidation mechanism);
undisclosed — the question applies and the venue does not publish the answer: a fact about the venue;
unknown — the answer could not be established from public sources: a fact about this dataset, not about the venue;
empty — nobody has established it yet.

The count on a venue's page is of cells with a value and a source. Today "unknown" can be recorded only for track record and book depth; for the other criteria, an answer that could not be established shows as empty.

A higher count is not a better venue. n / 10 measures how much of this dataset's homework is done on a venue, not how good the venue is. A venue with 10 of 10 may be a worse place to borrow than one with half that — it is simply one whose answers are all established.

No score, ever

There is no composite score, no stars, no ranking and no reliability index. A reader weighs the criteria; this site does not weigh them for the reader. Tables are not sorted by rate by default and no venue is floated. A single number would also be the thing that makes this a benchmark rather than a description, which it is not and will not become.

Unknown is a value

A criterion that cannot be established from public sources reads unknown with its reason — never guessed, never quietly dropped. A criterion that does not apply to a venue's class reads as the class-specific answer instead, which is informative in itself: a CeFi desk's upgradeability is "terms changeable by the operator at notice", which is a different statement from "unknown", and the two are never conflated. How a fact is known travels with the fact.

Every venue, the same row

Every venue appears as one row among the others, produced by the same pipeline from the same criteria — including any venue at which the operator holds a position. Where a venue is weak — book depth, external counterparty count, months live — its row says so with the same plainness as anyone else's. The rows where a venue looks good are only believable if the rows where it does not are visible.

The research floor

Not every venue is researched in full. A venue is listed, not researched where both of two things hold: this dataset reads it only through an aggregator, and its 30-day book has not reached $5m on any day of the last 90. Every other venue is researched in full — including any venue this dataset reads first-hand from its own contracts or events, at whatever size, because there it can stand behind the detail. As of 2026-10-13, 95 venues are researched in full and 40 are listed.

The floor is drawn on how a venue is observed and how large it is, never on which venue it is: no venue is named in the rule. A listed venue stays in the registry and its rate and book stay in every series; what it does not get is ten sourced criteria. An empty cell on a listed venue reads "below the research floor" rather than "not researched yet", and a criterion already researched on one is shown as before. It is a statement about where this dataset spends its research, not about the venue.

Two kinds of collateral: bitcoin and ether

The stack began with credit secured by bitcoin. It also measures credit secured by ether, and the reader chooses which (“Secured by”). Both are measured the same way — the same six figures, the same kinds and absences — and they are never added together: a dollar lent against bitcoin and a dollar lent against ether are different exposures. The currency a loan is made in stays a separate choice. 106 venues take ether as collateral today, as of 2026-10-13.

What counts as ether

Three families: ether itself (WETH, ETH), staking tokens (WSTETH, OSETH, CBETH, RETH, OETH and 9 more) and restaking tokens (WEETH, TETH, RSETH, WEETHS, EZETH and 3 more). Each figure states its split. Synthetic ether — a debt token, not a claim on ether — does not count.

Staking tokens at their own rate, the market beside it

A staking token is converted to ether at the rate its own contract states (for wstETH, wstETH.stEthPerToken(), 1.2462 ETH on 2026-10-12). That is the redemption rate, not the market price; a token can trade below it. The market price is published beside it as its own series, never folded in (6 of 24 tokens have one; the others say none is observed). A token whose rate is not read yet is left out of the figure and named in its limitation.

Where the ether is held

Ether on another chain is held by whoever runs its bridge. It counts only where that custody is established: on Ethereum, on rollups secured by Ethereum's own bridge, and on tokens whose bridge escrow has been read and reconciled. Ether bridged under custody that is not established is shown beside the figure and never added — $1.47bn on 2026-10-13.

Ether lent against ether

Much of the debt secured by ether is ether borrowed against staked ether — leverage on staking yield rather than credit in the usual sense. It is its own book (ETH → ETH), labelled as such and kept out of the headline, which is the dollar book.

Pooled markets

In a pooled market an account can post ether together with other collateral, and its debt is not split between them. The ether figures of a pooled venue are read from the accounts that post ether only; the others are not in them, and each such figure says so.

Prices and history

Ether's dollar value is the daily on-chain basket fixing, not a third-party quote; every ether figure in dollars carries that limitation. The ether figures start on the first day they were read directly. Earlier history, from 2020-12-07, is rebuilt from each venue's own contracts at past blocks, family by family, and checked before it is written; each page states where its history starts and every date its coverage changed. Debt above a single venue is read once a month, on the month's last Monday, because pooled venues are read account by account: between those readings it is absent, never filled in.

Coverage

As of 2026-10-13, 135 venues are in the registry and 85 of 128 contributed to the current reading. A venue absent from a reading is a declared gap in observation, not an observation of zero.

Third-party sources

Most of what this site shows is measured by Aletheia from chains and venues directly. Where a figure rests on a third-party series, the payload carries its credit and the page shows it beside the figure. The sources in use:

Source: Board of Governors of the Federal Reserve System (H.15, H.4.1, H.6).

SOFR, EFFR and ON RRP data are subject to the Terms of Use posted at newyorkfed.org. The New York Fed is not responsible for publication of these data by Aletheia Analytics, does not sanction or endorse any particular republication, and has no liability for your use. Aletheia Analytics is not affiliated with the New York Fed. SOFR is calculated using data provided under a license granted to the New York Fed by DTCC Solutions LLC. Federal Reserve Bank of New York

Data provided for free by IEX. By accessing or using IEX Historical Data, you agree to the IEX Historical Data Terms of Use. IEX

Machine access

The same data is served to agents over MCP at https://mcp.bitcoincreditstack.com/mcp, read-only. Nothing on that server signs, submits, broadcasts or prepares a transaction.

Where the operator holds a position

Aletheia Analytics may hold positions at venues in this dataset. Where it does, that is disclosed on that venue's page, as one of its Composition facts — the criteria already provide for it, under own-account share. It is not stated as a property of the dataset, because it is not one: a position at one venue says nothing about how any other venue is measured, and every venue here is measured the same way.

Access routes — what may be listed, and what may not

Each venue row can carry access routes: MCP servers through which an agent can act on that venue. These criteria are published before any route is listed, for the same reason the venue criteria are — listing rules written after the first listing can be shaped to admit it.

Scope: MCP servers only. Not an SDK, not a REST API, not a web application. The field answers one question — can an agent connected to this dataset act on this venue, and through which MCP? Broadening it would turn a machine-relevant field into a links directory.

A route is listed only if all five hold:

It is an MCP server, speaking MCP over a documented transport.
It is publicly documented by its operator, in the operator's own words, at a citable URL.
It was reachable when listed — initialize answered, and the date is recorded.
Its operator is identified. Never an anonymous endpoint.
Its scope is recorded as observed — read and write tool counts read from tools/list, never from the operator's description of itself.

And is not listed if any of these hold:

It requires custody of keys to function. Naming a server that reads, or that prepares an unsigned transaction, is descriptive. Naming one that holds keys points a reader toward handing over control, which is a different act.
It was not reachable when listed.

Listing is not endorsement, and the order is not a ranking. Routes are recorded as dated facts in registry order. Aletheia integrates with none of them and prepares transaction calldata for no contract outside the Gavel Protocol.

Routes are re-checked, and carry a last-checked date. "Reachable when listed" goes stale exactly the way a configuration snippet does. A route that stops answering is marked unreachable with the date — never silently dropped, because a disappearing row and a broken row are different facts.

Three states, and the third is not the second. A venue row reads either route(s) listed, or surveyed — no route found, or not yet surveyed. We looked and found nothing and we have not looked are different claims about a venue, and collapsing them would state the first while meaning the second.

Parity is the same question asked of every row, not the same answer. A venue with a published MCP carries one; most do not. That is also what lets the Gavel row carry a route without being privileged: it is asked the same question as the other 161.