NewR.A.I.N. Lab Jev Judgment
Review curated research claims with Jev questions and a separate deterministic promotion gate.
Added to Jevfast
NewHow it uses Jev
In the explicit strict judgment cycle, R.A.I.N. Lab sends a curated evidence packet about one candidate claim to Jev. Jev evaluates five independent questions over the same state: a four-option claim-support Choice, a four-level evidence-quality Score, and three Noul questions about contradiction, scope, and need for human review. Deterministic validation and a local gate decide PASS, REVISE, HUMAN_REVIEW, or UNAVAILABLE; the provider cannot change the evidence, thresholds, or disposition rules.
What Jev decides
Illustrative research-claim reviewExample answers · not a recorded Jev response · Source ↗
Question 1 · claim support
YOUR APP
INSTRUCTION
Classify support for the claim as supported, mixed, unsupported, or insufficient evidence.
STATE
The repository's documented example packet claims a simulated resonance peak at 40 Hz. It records a fixed 30–50 Hz sweep, three observations at 40 Hz, tool evidence, simulation-only limitations, a formal result, numerical validation, synthesis, and peer critique. This is the source-defined request context, not a live Jev call.
JEV · CHOICE
- supported
- mixed
- unsupported
- insufficient_evidence
Question 2 · evidence quality
YOUR APP
INSTRUCTION
Rate evidence quality on the documented four-level rubric.
STATE
The repository's documented example packet claims a simulated resonance peak at 40 Hz. It records a fixed 30–50 Hz sweep, three observations at 40 Hz, tool evidence, simulation-only limitations, a formal result, numerical validation, synthesis, and peer critique. This is the source-defined request context, not a live Jev call.
JEV · SCORE
0
Question 3 · contradiction present
YOUR APP
INSTRUCTION
Estimate the probability that the packet contains a meaningful contradiction.
STATE
The repository's documented example packet claims a simulated resonance peak at 40 Hz. It records a fixed 30–50 Hz sweep, three observations at 40 Hz, tool evidence, simulation-only limitations, a formal result, numerical validation, synthesis, and peer critique. This is the source-defined request context, not a live Jev call.
JEV · NOUL
YesNo
Question 4 · scope violation
YOUR APP
INSTRUCTION
Estimate whether the claim exceeds what the evidence supports.
STATE
The repository's documented example packet claims a simulated resonance peak at 40 Hz. It records a fixed 30–50 Hz sweep, three observations at 40 Hz, tool evidence, simulation-only limitations, a formal result, numerical validation, synthesis, and peer critique. This is the source-defined request context, not a live Jev call.
JEV · NOUL
YesNo
Question 5 · human review
YOUR APP
INSTRUCTION
Estimate whether the claim requires explicit human review.
STATE
The repository's documented example packet claims a simulated resonance peak at 40 Hz. It records a fixed 30–50 Hz sweep, three observations at 40 Hz, tool evidence, simulation-only limitations, a formal result, numerical validation, synthesis, and peer critique. This is the source-defined request context, not a live Jev call.
JEV · NOUL
YesNo
App workflow
- Curate an evidence packet
The caller supplies the claim, method, observations, quantitative results, tool evidence, limitations, source identifiers, formal and numerical checks, synthesis, and peer critique.
- Apply host prechecks
The strict-cycle command rejects malformed or oversized packets and routes missing evidence or failed validation before calling the remote provider.
- Ask five atomic questions
The JudgmentProvider evaluates the immutable packet using one Choice, one Score, and three Noul questions. Every question sees the same canonical state.
- Run the deterministic gate
Local policy combines the typed results with validation status, confidence, and fixed thresholds to determine PASS, REVISE, HUMAN_REVIEW, or UNAVAILABLE.
- Record the judgment
The workflow stores the typed answers, probabilities, confidence, policy, provider/model, disposition, and reason codes in a replayable artifact before discovery promotion.
Why it is interesting
The judgment layer sits after peer critique and before discovery promotion, while retaining formal and numerical checks as separate evidence. Its bounded packet and deterministic gate make model uncertainty a routing signal rather than treating a Jev answer as scientific proof.