Company

Independent verification for a multi-model world.

Recherche.fr is a European R&D initiative developed in France. We are building a provider-independent layer for evidence-aware, traceable AI verification.

Capability status

Engineering exists. Research maturity is reported separately.

Implemented

  • Authentication and workspaces
  • Provider settings and real provider adapters
  • Encrypted BYOK foundations
  • Research sessions and response persistence
  • Comparison and queued execution foundations

Prototype / active R&D

  • Claim decomposition and contradiction handling
  • Evidence-aware verification routing
  • Recherche Judge calibration
  • Benchmark methodology
  • Industrial validation

No single percentage is used because engineering completion, evaluation quality and scientific maturity are different dimensions.

System boundary

What Recherche.fr is — and is not.

The product boundary matters: verification should remain independent from both the models being evaluated and the authoritative sources used as evidence.

01Claim extraction
02Verification routing
03Evidence-aware adjudication
04Provenance & traceability
Inside the verification layer

Independent, claim-centered verification.

Recherche.fr turns heterogeneous model outputs into comparable claims, decides what deserves independent checking and preserves the path from evidence to verdict.

AI outputs→Claims→Evidence→Judge→Traceable synthesis
Research frontier

Where the research frontier begins.

Engineering implementation and research maturity are tracked separately. Select a frontier to inspect the unresolved research question.

R&D 01

Judge calibration

Determine when claim-level adjudication is reliable, how confidence should be calibrated and how Judge bias is measured.