Nothing reaches a result in ecosphere until it has been checked against a published data contract and reviewed by a second person. Every figure carries its source and version, stand-in data is always labelled, and your organisation’s data stays your own.
Hazard data from ecosphere and a loan book from you follow the same path. A delivery is only usable once it has passed every stage, and once published it is never silently changed: a correction is a new version.
Ours or yours: hazard indicators, terrain, scenarios, or your counterparties and sites.
The published contract says what a valid delivery looks like. Every failed check is listed in plain language.
Someone other than the person who prepared it looks at it before anything is published.
The delivery gets a version number. Earlier versions stay readable, so past results can be reproduced.
Stress tests, dossiers and the assistant read published versions only, and record which ones they read.
The data contract is published: the fields each delivery must carry, the units, the ranges and the rules that tie them together. It is versioned like the data itself, so you always know which rules a figure was checked against.
When your own upload fails a check, the inbox says which rows and why, so the fix is a quick edit rather than a support ticket.
Every figure in ecosphere is shown the same way: a value, a range where there is uncertainty, a source and a version. Synthetic or stand-in data is always labelled, and the label stays with the figure into reports, exports and the assistant’s answers.
Each organisation, and each company inside it, is isolated. Your book, your runs and your assistant conversations are not visible to anyone else, and data crosses a boundary only when someone with the authority to share it decides to.
A group with several entities keeps each company’s data apart inside one organisation, and the organisation apart from every other client.
Sharing is an explicit act, not a setting someone forgot. Nothing moves between companies or organisations on its own.
Each organisation has a reporting currency, such as ZAR, KES or NGN, and exposures, expected credit loss and reports are shown in it.
The assistant answers plain-language questions from your organisation’s published data and results, and shows the run and versions behind each figure. You choose the model provider, so the conversation stays inside an agreement you already have.
The same published, versioned data reaches you three ways. A figure in the console, a row in a Parquet file and a value from the API are the same figure, with the same provenance.
Explore hazards, run stress tests, open the site atlas, manage the data inbox and talk to the assistant. Nothing to install; scenario reports export from here.
Columnar files for your data lake, warehouse or notebooks, with version and source columns alongside every value, so lineage survives the move into your own models.
Programmatic access to the same published data and results, to upload your counterparties and sites and to feed your own risk systems.
# stress_results/ST-0147/v1/counterparty.parquet counterparty_id string CP-0417 shock string extreme_rain_1in100 transition string delayed_transition horizon int16 2030 pd_baseline double 0.018 pd_stressed double 0.031 ecl_uplift decimal 6200000.00 currency string ZAR source_deliveries list<string> [hazard_v12, flood_v4, ngfs_v8, book_q3_v3]
In a working session we run a sample of your counterparties through a physical shock and a transition scenario, and walk you through every step of the chain from hazard to capital.