Technology

From model answers to a traceable verdict.

Five components turn parallel model outputs into claims, evidence, verdicts and a traceable synthesis.

Follow one question through the architecture.

Click a component to see its input, role and output.

01Router
02Claim Engine
03Verification Router
04Recherche Judge
05Synthesizer

See how model answers become comparable claims.

Click an answer or a claim. The Claim Engine normalizes different wording before detecting agreements and contradictions.

Model answers
OpenAI

The DPA covers processing and appears to restrict region.

Claude

A DPA exists, but exclusive EEA processing is not established.

Gemini

SCCs may cover some transfers.

Mistral

Subprocessor location still needs verification.

Grok

Current configuration may contradict regional exclusivity.

Claim Engine
A DPA covers the processing

Comparable atomic statement · #1

Processing remains exclusively in the EEA

Comparable atomic statement · #2

SCCs cover some transfers

Comparable atomic statement · #3

Relevant subprocessors are located in the EEA

Comparable atomic statement · #4

Canonical claim
Normalized output

The graph preserves which models support, qualify or contradict each claim.

Parallel fan-out

One question. Multiple AIs in parallel. Then verification.

The same question is sent to multiple systems. Recherche.fr compares independent answers before verifying the claims that matter.

OpenAIOpenAIChatGPT
AnthropicAnthropicClaude
Google GeminiGoogleGemini
MistralMistralMistral
xAIxAIGrok
DeepSeekDeepSeekDeepSeek
QwenQwenQwen
MetaMetaLlama
03Independent answers

Each model answers separately.

04Extracted claims

Answers are decomposed into comparable statements.

05Agreement & disagreement

Model positions are compared at claim level.

06Independent evidence

Documents, registries, standards and authoritative sources.

07Recherche Judge

Our specialized model adjudicates against the evidence.

08Traceable synthesis

The conclusion remains linked to models, claims and evidence.

Illustrative configuration: the model set can vary by task.See how Recherche Judge adjudicates →

Five components, five distinct responsibilities.

The architecture deliberately separates model selection, claim extraction, verification, adjudication and synthesis. Some foundations are implemented today; the evidence-aware components remain active R&D and are being benchmarkinged and calibrated.

01Implemented / evolving

Recherche Router

Selects or recommends models and execution strategies according to task, domain, diversity, cost, latency and observed performance.

InputUser task + constraints
OutputModel / strategy selection
02Prototype

Claim & Contradiction Engine

Breaks long answers into comparable claims and separates material contradictions from differences in wording or style.

InputProvider responses
OutputClaims + disagreement graph
03Active R&D

Verification Router

Decides which claims need independent checking, what type of evidence is appropriate and how much verification is justified by risk, cost and latency.

InputClaims + risk signals
OutputVerification plan
04Active R&D

Recherche Judge

Adjudicates at claim level using evidence quality, source authority, freshness, contradictions, provenance and uncertainty — not model popularity.

InputClaims + independent evidence
OutputSupported / unresolved / contradicted
05Prototype

Recherche Synthesizer

Builds a final answer without hiding unresolved points and preserves the path back to the models, claims, evidence and verdicts that produced it.

InputJudged claims + provenance
OutputTraceable synthesis

What already exists

The current platform is not a concept deck. Work already covers authentication, workspaces, provider configuration, encrypted BYOK, research sessions, response persistence, multi-provider adapters, comparison, contradictions, synthesis, queued execution, staging and CI/E2E.

01
Multi-provider orchestration
02
Persistent research sessions and responses
03
Comparison and contradiction workflows
04
Secure provider settings and BYOK foundations
05
Staging-first delivery, CI and E2E foundations

More models are not the research contribution.

The difficult part is deciding when agreement is misleading, which claims justify the cost of verification, what counts as authoritative evidence, how Judge decisions should be calibrated and how reliability changes as models, providers and sources evolve.

Explore the research →
Interactive product model

Trace what each layer actually changes.

Select a stage and scenario. The module shows its input, transformation, output, and what fails if it is removed.

Illustrative · local state only
Active stage
Scenario
Input
Transformation
Output
Failure if absent
Claim Graph
Technical depth

The hard part is not calling eight models.

Fan-out and orchestration are standard engineering. The difficult work sits in the transformations that must remain inspectable and testable between model output and final answer.

01 · Claim canonicalisationTurn different wording into comparable atomic claims

Resolve paraphrases, negation, scope and coreference without erasing meaningful disagreement.

02 · Evidence independenceSeparate support from repetition

Track authority, freshness, provenance and conflicting evidence so multiple copies of the same assertion do not become independent proof.

03 · Judge de-biasingPrevent style, position and provider identity from becoming evidence

Evaluation design must test blinded provider identity, ordering effects, persuasive wording and uncertainty handling.

04 · Synthesis faithfulnessThe final answer must not invent a stronger conclusion

The Synthesizer should remain constrained by adjudicated claims, unresolved states and provenance.

Research boundary: these are active engineering and evaluation problems. The site does not present them as solved properties.
Compounding research asset

Verified Claim–Evidence Graph.

The architecture itself is reproducible. The potentially harder-to-replicate asset is the accumulated history of claims, contradictions, evidence, adjudications, corrections and later outcomes — if it is built with rigorous provenance and governance.

Claim identityCanonical claim + versions + scope

Track paraphrases and revisions without collapsing materially different statements into one node.

Evidence historySource · authority · freshness · provenance · contradiction

Evidence is attached with time and source lineage so historical decisions can be re-evaluated.

Adjudication historyVerdict · uncertainty · policy version · reviewer corrections

Each decision records the evaluation context rather than storing only a final label.

Learning loopLater outcomes can expose calibration and routing errors

Corrections can become evaluation material for Judge, Router and verification-policy research.

Important: this is a strategic data-flywheel design objective. Recherche.fr does not claim that a large proprietary graph already exists.