Family
Sentence-level NLI grounding
productionMITtextinternal
HallucinationCheck.
Sentence-level entailment check that asserts every factual sentence in a model output is grounded against a retrieved source passage before delivery.
Foundation · DeBERTa-v3-base-NLI (MIT)
What it does
The honest description.
Sentence-level entailment check that asserts every factual sentence in a model output is grounded against a retrieved source passage before delivery.
Architecture
How it works.
Independent grounding signal feeding the Judge in DEEP trials and emitted as `hallucination.signal.v1` for the flywheel.
Training data & drift
What it learned from, and how we watch it.
Source
DeBERTa-v3-base pre-trained on MNLI + ANLI (MIT). No FinAi-specific fine-tune yet; per-domain adaptation is on the M5 backlog.
Drift baseline · current quarter
Drift baseline not yet documented for this model.
What it's NOT
- Reduces but does not eliminate hallucination on long-form output — frame as 'every claim grounded against source', never 'cannot hallucinate'.
- NLI confidence is on a 0–1 scale; a borderline contradiction can still pass at low thresholds.
- Operates per-sentence; multi-sentence reasoning chains are not currently scored as a unit.
Who calls this model
Across the house and the API.
Internal routes
No public routes documented yet.
Featured in
Sentence-level entailment score
Source-passage retrieval at runtime
Per-trial verdict cached with the Courtroom decision
Version history · immutable once published
Every release, traceable.
Data Mukodi holds about you is subject to DPDP §12 erasure (use /consent in the app), and any concern about this model can be raised via [email protected] under DPDP §13 grievance, or via /security.