Custom analytics on the data behind the score.
Credit models, book analyses, data pipelines and research, built on the same record that powers Cred's score: lending events on Aave, Compound and Morpho, and wallet behaviour across ten chains. Scoped engagements at a fixed fee, on your questions.
- engagement
- book analysis
- question
- which funded wallets to re-underwrite before the limit review
- inputs
- 412 funded wallets (CSV) · 18 months of repayment outcomes
- method
- score + report per wallet · band by outcome · protocol and chain concentration
- deliverable
- written review · per-wallet CSV · rerun monthly
- timeline
- 3 weeks · fixed fee
Four kinds of engagement.
Each one ends in something your team runs, reads or ships without us. Most engagements combine two: a book analysis that becomes a monthly pipeline, or a research question that turns into a model.
Custom credit models
A scoring or risk model fitted to your borrowers and your own loss definition, built from the lending record and validated the way the Andromeda model is: against borrower archetypes your analysts agree on, then in shadow on live traffic before it decides anything. Shipped as an API endpoint your systems call like the public score.
scoring endpoint · model card · validation notes
Portfolio and book analysis
Your funded wallets scored and read as one book: distribution by band, debt against collateral, liquidation exposure, concentration by protocol and chain, and the cohorts that behave differently from the rest. One-off before a policy change, or on a schedule.
written review · per-wallet CSV · rerun on request
Onchain data pipelines
Decoded wallet and protocol data delivered where your team already works: lending events priced in USD on the day they happened, balance history, cash flow and sanctions status, landing in your warehouse or a dashboard on a refresh schedule you set.
warehouse tables · dashboard · refresh schedule
Research reports
Studies of borrower and protocol behaviour, such as how repayment differs between Aave and Morpho or what a wallet does in the weeks before a liquidation, written for a risk committee, a board or a public audience, with the method and its limits stated.
report · figures · reproducible queries
Built on the record behind the score.
No engagement starts from a blank index. The data the public API reads is already decoded, priced and versioned, so the time goes into your question rather than into plumbing.
- Ethereumlending
- Baselending
- Arbitrum
- Optimism
- Polygon
- Avalanche
- BNB Chain
- Celo
- Linea
- Robinhood
- Lending eventsEvery supply, borrow, repay, withdraw and liquidation on Aave V2 and V3, Compound V2 and V3 and Morpho Blue, priced in USD at the day's rate.Ethereum · Base
- Wallet behaviourTransfers, holdings, contract interactions, first and last activity, and whether the address is a wallet or a contract.10 chains
- Balance history and cash flowPositions today and 30, 60 and 90 days ago; external, lending and conversion flows over 30, 90 and 365 days.per wallet
- Identity and trustENS, Gitcoin Passport stamps, identity attestations and sybil signals.per wallet
- SanctionsOFAC SDN and consolidated lists refreshed hourly, with one-hop exposure through funders and counterparties.EVM
- The score300 to 1000 with six factor ratings, from a deterministic model that names its version on every response.andromeda_1.x
How an engagement runs.
Four steps, each with something written at the end of it. The price is fixed at the first step, and nothing that will make decisions goes live without a shadow run beside what you do today.
- Step 1
Scope
A 30-minute call on the question you need answered, the data you hold and the decision it feeds. Within a few days you get a written scope and a fixed price.
written scope · fixed fee
- Step 2
Build
We work from Cred's indexed record plus whatever you share under NDA. Anything that will make decisions runs in shadow first, beside what you do today.
shadow run · weekly notes
- Step 3
Review
Results walked through with your risk or product team, with the method and its limits stated, and revised until the committee can defend them.
review session · revisions
- Step 4
Handoff
Delivered as an endpoint, warehouse tables, a dashboard or a report, with the queries behind it, so your team can rerun it without us.
code · queries · documentation
What holds across every engagement.
- Validated before it decidesA model has to rank borrower archetypes the way your analysts would, and run in shadow beside your current process, before it is allowed to set a term.
- Explainable at the level you can publishResults come with factor-level reasons a borrower can be given, never the weights a borrower could work.
- Gaps are reported as gapsWhen a chain or a feed cannot be read, the analysis says so and marks what depends on it, instead of treating missing data as a clean record.
- Yours to keepCode, queries and documentation are handed over at the end. Nothing in the deliverable needs us to run again.
- Your data stays yoursBorrower lists and loss history are shared under NDA, used for your engagement only, and deleted on request when it ends.
Questions
Who does the work?
The same people who build and maintain the Andromeda model and the Cred API. Whoever is on the scoping call is the person doing the analysis; nothing is handed to a separate delivery team.
How is it priced?
A fixed fee per scoped engagement, quoted after the call. Analyses that rerun on a schedule are priced per run. Once a model or pipeline is handed off, API usage is billed in Cred Units on the usual plans.
Do we need to share our data?
Not for research or pipelines built on public chain data. Custom models and book analyses are better with your borrower list and loss history, shared under NDA, but a list of wallet addresses is enough to start.
How long does an engagement take?
A book analysis typically takes two to three weeks. A custom model runs longer, usually six to ten weeks including its shadow period. Pipelines and reports depend on scope, and the written scope sets the dates before any work starts.
Can you work with data we already hold?
Yes. Exports from your own warehouse, Dune queries, a loan book in a spreadsheet or a list of addresses all work as inputs. We join them to the indexed record rather than asking you to rebuild anything.
Bring the question.
Thirty minutes on the decision you need to make and the data you hold. You leave with a written scope and a fixed price, or with a straight answer that it is not a fit.
Book a callPrefer email? [email protected]