How it works
Five steps, and the limit of each one.
Every step below names the data it reads. Where a step cannot answer, the report says so rather than guessing, which is why the third block of the report exists at all.
- 01
The application arrives
Your ATS, email, or upload
A CV comes in through your ATS, by email, or by upload. The file is parsed from the request in memory. It is never written to disk or to object storage, and the extracted text is discarded once claims are built.
- 02
Claims are extracted
Extraction model hosted in the EU under a processing agreement
The CV becomes a list of separate, checkable claims: employer, role, start and end date, contact details, document metadata. Special categories under Article 9 of the GDPR, such as health, religion, political views, union membership or sexuality, are never extracted or stored, even when the CV contains them. The extraction prompt forbids it and a test checks it against a CV fixture that contains them.
- 03
Each claim is checked against a source
ARES and EU business registries, RDAP, web archive, DNS, numbering plans
Company registries confirm whether the employer existed and for how long. Domain registration records and web archives confirm whether the company had a presence during the claimed period. Mail infrastructure and phone numbering plans confirm whether contact details behave the way they claim to. Only company and domain data leaves Candora. The candidate's name is never sent to any of these sources.
- 04
The evidence report is written
Candora
Findings, verified claims and unverifiable claims go into three separate blocks. A claim appears in exactly one block. Every finding carries the source, the time of the query, the version of the rule that produced it and a sentence written for a person, not for a log file.
- 05
A person decides
Your team
The recruiter reads the report, agrees or overrides, and records what they did. The decision goes into an append only log together with the rule version in force at that moment, which is what makes the process reconstructable a year later.
What the process deliberately does not do
These are not gaps waiting to be filled. Each one is a decision with a reason, and each one has a test that fails if someone crosses it.
- It does not produce a score, a risk level or a probability at any point in the pipeline.
- It does not reject, and there is no code path that could.
- It does not send the candidate's name to any external source.
- It does not read anything on the candidate's device.
- It does not merge identities across customers without a human confirmation recorded in the database.
- It does not keep the CV file. Retention applies to the claims, and defaults to 30 days.
The fastest way to understand it is to run one.
Sample candidates, the same rules, ninety seconds, no signup.