Claims#

Allowed claim:

NeuroTwin proposes and evaluates a neural translation benchmark and model for heterogeneous neural signals under strict leakage controls. The current experimental architecture is NeuroTwin NFC, a Neural Field Compiler that treats recordings as partial observations of a latent neural field.

Additional allowed current claims:

  • The repo implements leakage-audited neural translation benchmark infrastructure.

  • The repo implements an experimental NFC path.

  • The repo supports synthetic NFC smoke/falsification checks.

  • The repo supports MOABB Track A leakage/reproducibility evidence.

  • The repo can package A100 handoffs from clean committed HEAD.

Do not claim:

  • NeuroTwin beats baselines

  • NFC is proven

  • NeurIPS-quality result

  • first brain foundation model

  • first multimodal brain model

  • first stimulus-to-brain model

  • clinical digital twin

  • diagnostic model

  • treatment predictor

  • human brain clone

  • proven neurology breakthrough

  • depression/MDD classifier

  • full fNIRS support

  • exact TRIBE v2 reproduction

  • exact BrainVista reproduction

  • TurboQuant as the model contribution

  • A100 success if no evidence bundle exists

Synthetic smoke tests validate plumbing only. MOABB smoke tests validate the first real-data path only; they are not full NeuroTwin validation.

NFC synthetic tests validate plumbing only. Pair-Operator is an ablation/baseline inside NFC, not the main architecture.

Old short-step NFC synthetic signals produced before the corrected falsification gate are invalidated as evidence. They may be referenced only as debug history.

tribe_style is a NeuroTwin-native toy clean-room approximation for the TRIBE v2 stimulus-to-fMRI baseline lane. Do not describe it as exact TRIBE v2, as using TRIBE v2 weights, or as using real video/audio/text encoders unless explicit pretrained stimulus features are loaded.

Real stimulus claims require a verified source artifact hash. transcript_hash and synthetic feature hashes are plumbing-only and do not make stimulus-to-fMRI claims eligible.

Scientific claims require real prepared data, leakage audit pass, validation-selected checkpoints, final held-out test metrics, and an explicit scientific_claim_allowed=true decision in summary.json. A passed colocated paper_mode_gate.json means the artifact contract is satisfied; reports must show that gate status separately and must not promote claim allowance from the gate alone.

Conditional future claims:

  • NFC synthetic claim only after the strict synthetic gate passes.

  • Algonauts/fMRI claim only after real stimulus hashes are verified and held-out splits pass.

  • Model superiority only if NFC beats baselines and ablations under an evidence gate.

  • Semantic leakage audit claim only after exact or validated retrieval audit runs.