“Integration” in a crypto partnership announcement can describe anything from two logos on a landing page to a shared order book with pooled collateral. The word is used for all of it, and the difference between the cheapest and the most expensive version is roughly the difference between a press release and a counterparty relationship.

The announcement will rarely tell you which level it means. It says “integrated” and lets you supply the level you already had in mind. The way to find out is to look for the artefacts each level leaves behind, because every level produces a different kind of evidence, and some produce none at all.
- Four levels share one word. Co-marketing, technical connection, live with real funds, and shared liquidity are separate achievements.
- Each level leaves a different artefact. A logo, a documentation page, a deployed contract, or a shared balance sheet.
- Only one level is testable by a user. If you can move a small amount end to end today, the integration is real for you.
- No date means no level three. An announcement without an effective date describes an intention, not availability.
Four levels, one word
The levels are cumulative in effort but not in sequence. A company can reach level four with one partner and still be at level one with another, and the same word will appear in both announcements. Reading the word tells you nothing. Reading the artefacts tells you which level you are looking at.
| Level | What exists | Who did the work | What a user can do today | Evidence to look for |
|---|---|---|---|---|
| 1. Co-marketing | A shared announcement, logos, perhaps a joint event | Marketing teams on both sides | Nothing new; the products are unchanged | Two posts, no documentation, no date |
| 2. Technical connection | API keys issued, endpoints or SDK published, sandbox access | Engineering on both sides, often one side leading | Developers can build; users cannot necessarily transact | Versioned docs, changelog entry, status page |
| 3. Live with real funds | Production access, liquidity, compliance and a withdrawal path | Operations, risk and compliance, not just engineering | Move a minimum amount end to end and back | Contract address, effective date, both parties confirming |
| 4. Deep integration | Shared liquidity, shared accounts, or delegated custody | Treasury, legal and risk functions | Transact against a counterparty’s balance sheet | Published limits, collateral terms, failure handling |
The last column is the useful one. Every level except the first leaves something you can inspect without asking either company for permission.
Level one and level two: what the artefacts prove
The first two levels are the most common and the most over-read. Both are real work. Neither one means the product works.
The logo swap
A co-marketing agreement is a promise to mention each other. Both sides agree to display a mark, publish a post, and perhaps appear on a panel together. No code is written, no account is opened, and no funds move. This is a legitimate marketing activity, and it can be worth something: distribution is hard, and a partner with an audience can put a product in front of people who would never have found it.
What it does not imply is availability. A logo on a partner page is not a feature. If the announcement contains no documentation link, no contract address and no date, the correct reading is that two marketing departments agreed to say something nice, and the product you can use today is exactly the product you could use yesterday.
The technical connection
Level two means engineering has happened. The most reliable artefacts are a versioned documentation page, a changelog entry naming the partner, and published API endpoints or an SDK release. Documentation is difficult to fake at scale because it has to be specific: endpoint names, authentication method, rate limits, error codes and request examples.
This is also the level where permissions start to matter. An integration that issues an API key with withdrawal rights is a different risk from one that can only read balances, and announcements almost never mention which scopes were granted. Our note on API key permissions and leaks covers the distinction in detail; the short version is that a read-only connection and a trade-and-withdraw connection should not be described by the same sentence.
Sandbox is not production
Many level-two integrations exist only in a test environment. Sandbox access proves that two systems can exchange messages in a controlled setting with invented balances. It does not prove that production credentials have been issued, that limits have been configured, that compliance onboarding exists, or that anyone has ever settled a real transaction.
The difference is visible in the documentation. Production endpoints are usually a separate host, a separate key type and a separate onboarding step. If the only documented base URL is a sandbox host, the integration is a capability, not a service.
Level three: live, with real funds
Level three is the first one a user can test, and the only one that changes what you can do with your own money. It requires more than an API key: a deployed contract or a production endpoint, liquidity behind it, a compliance path for the jurisdictions served, a withdrawal route, and monitoring for when something breaks.
What “live” requires
On-chain integrations leave the clearest evidence. A deployed contract has an address, and the address has a history. You can check whether the contract is verified, whether it has been called by the partner’s known addresses, and whether the calls are recent or all clustered in the week of the announcement. A contract that has processed a handful of test transactions and nothing since is a proof of concept with a press release attached.
Off-chain integrations are harder. The strongest evidence is a production status page with incident history, a support article describing the new route, and a minimum amount that actually arrives. Marketing pages are the weakest evidence, because they are written by the same team that wrote the announcement.
The effective date
A live integration has a date on which it became live. The date may be phrased as “available from”, “effective”, or “now supported”, but it exists, and it is in the past or the immediate present. An announcement that says “will be available”, “rolling out” or “in the coming weeks” is describing a plan.
Plans are not worthless. They tell you what the companies intend, and they can be a useful signal about strategy. They are simply not evidence of availability, and the distinction should survive the headline. The practical rule: if you cannot find a date, assume level two.
Level four: deep integration
Level four is where the two companies stop being vendors to each other and start carrying each other’s risk. Shared liquidity, shared accounts, delegated custody or a common collateral pool all mean that one party’s failure becomes the other party’s problem.
Shared liquidity is a balance-sheet relationship
When a venue routes orders into a partner’s book, or holds inventory on a partner’s venue to quote prices, the arrangement depends on the partner’s solvency and operational reliability. The user sees one interface; behind it, two balance sheets are connected. That is a materially different claim from “we integrated”, and it should come with published limits, settlement terms and a description of what happens if one side halts.
The liquidity question is worth separating from the integration question. Depth is a property of the market, not of the announcement, and our piece on exchange liquidity and why depth matters more than volume explains how to measure it. A level-four partnership may improve depth, worsen it, or leave it unchanged. Only the book tells you which.
Shared accounts and who owns the claim
Some integrations blur the boundary of whose account you are actually using. If a product opens a sub-account at a partner venue on your behalf, your balance may be a claim on the partner rather than on the brand you signed up with. That is not automatically bad, but it changes who you chase when something fails, and it changes which entity’s risk controls apply.
Our note on sub-accounts and separating risk sets out how those structures are usually organised. The point for reading an announcement is narrower: any arrangement that moves your funds into another entity’s ledger should say so, in the text, without requiring you to infer it from a support article published six months later.
How to verify an integration yourself
The whole check takes about fifteen minutes and requires no contact with either company. Do it before you rely on the integration for anything that matters.
- Read the announcement for a verb and a date. “Has integrated” plus a past date is testable; “will integrate” is not.
- Open the documentation and look for a changelog entry naming the partner, not just a landing page.
- Find the contract address, endpoint or product page, and confirm it exists on the partner’s own domain rather than only on the announcer’s site.
- Check the contract on a block explorer: is it verified, who has called it, and are the calls recent.
- Look for a status page and read the incident history for the period after the announcement.
- Move the minimum amount end to end, and then move it back out. The return leg is the part that fails.
- Compare what each company says. If only one side has announced it, treat the claim as one-sided.
An integration is proven by a completed round trip at the smallest size that the system allows, not by a logo, a landing page or a co-signed quote. Everything before the round trip is a claim about capability.
The round trip matters because integrations usually fail at the edges. Deposits may work while withdrawals are still manual. The asset may be supported for trading but not for transfers. One jurisdiction may be enabled and another queued. These are exactly the details that a marketing page omits and that a small test exposes.
Announcements that are only intentions
A large share of partnership announcements describe something that does not exist yet. That is normal in any industry. The problem is that the grammar of intention and the grammar of completion are close enough that a headline can cross from one to the other without anyone lying.
Reading the verb
“Signed a memorandum of understanding” means the parties agreed to keep talking. “Entered a strategic alliance” usually means the same thing with a larger budget for the announcement. “Is exploring” means a conversation. “Plans to integrate” means a roadmap item. None of these is a commitment to deliver, and none has an effective date that binds anyone.
The stronger verbs are duller. “Now supports”, “is live”, “has deployed” and “is available to customers in” are claims that can be checked and, if wrong, contradicted. When an announcement uses exciting verbs, look for a dull sentence somewhere in the body that gives the real status. It is usually there, in paragraph seven, phrased as a caveat.
The two-company test
The simplest test is whether both parties are willing to say it. A real integration normally produces two announcements, two documentation updates and two sets of support articles, because both sides have customers who need to know. When only the smaller company announces, and the larger one has nothing on its own channels, the asymmetry is the signal.
Asymmetry has innocent explanations: a partner may not want to publicise a small reseller, or may be holding its announcement for an event. But an unexplained asymmetry should lower your confidence until the second side appears. Companies are usually eager to claim distribution. Silence from the larger party is a choice.
What to record, and why
Keep a short note for integrations you actually depend on: the date you verified it, the level you observed, the smallest amount you moved, the fee you paid, and how long the return leg took. This takes a minute and gives you something that no announcement can: a baseline.
Baselines matter because integrations degrade. A partner changes an API version, a compliance team tightens a limit, a liquidity provider withdraws, a chain upgrade breaks a contract. When something stops working, the useful question is not “did they announce this” but “when did it last work for me, and what changed”. Your own record answers that; the press release never will.
The broader habit is to treat every partnership claim as a question about artefacts rather than about intentions. Ask what exists, who can use it, and what happens when it fails. The answers sort the announcement into one of four levels quickly, and the level is the only part of the claim that affects you. Everything else is positioning, and positioning is what the announcement was paid to produce.

0 comments
No comments yet — ask a question and we will answer it here.