Skip to content
CDSBench
Private data

How private documents are handled

Everything on this page describes what the current code does — not a policy we intend to have. If a line here is not true of the deployment you are using, that is a bug we want reported.

Where documents go
Uploaded files are converted to text on the CDSBench server (PDFs are parsed there; nothing is sent to a third-party parser) and the text is sent over HTTPS to the CDSBench application and stored, together with the result, in the CDSBench database (Cloud SQL, europe-west1) against the email address you entered. Nothing is written to a third-party document store.
Public benchmark
Private customer cases are not automatically added to the public CDSBench dataset. Private analyses are a separate table from benchmark cases, are never listed publicly, and are never used for precedent retrieval.
Model providers
This deployment runs in mock mode: no document text is sent to any external model provider. When live providers are enabled, this page will state which vendors receive prompts.
Retention
Private analyses are retained until you ask for deletion. There is no automatic expiry yet.
Deletion
Self-service deletion is not built yet. Email the address you used with the subject “delete my analyses” and the records are removed from the database by hand; there are no backups that retain them beyond the provider's standard backup window.
Access
Results are returned to the browser session that ran the analysis. There is no login and no per-user page yet, so a result is not retrievable by email alone — keep the JSON export if you need it.
Analytics
Product analytics record that an analysis started and completed, its final status, and timing. Document text, entity names and instrument names are not sent to analytics. Google Analytics is not configured on this deployment.
Free limit
2 completed analyses per email address. Failed runs do not count. No payment details are collected anywhere on the site.

Public benchmark cases come from legally usable public material; their provenance is recorded per source. Methodology →