Explore an evidence problem with us
The free Cejel core is live today: deterministic, offline, evidence-bound, and it abstains when it can't be sure. We are exploring where the same approach could help with domain-specific evidence. Quant research and ML integrity are illustrative examples, not announced products or roadmap commitments. A design partnership begins as a bounded study, and no resulting pack is represented as available or validated unless it later passes preregistered calibration on repositories it has never seen.
Example: quant research evidence
Illustrative candidateA candidate study could examine whether a reported backtest is accompanied by inspectable evidence for dataset boundaries, timing assumptions, costs and fills, and derivation of the reported result. The exact rules, detectable evidence, and abstention conditions have not been set.
Potential fit: teams willing to examine failure modes and false-positive costs using non-sensitive or locally retained artifacts. Any resulting engineering evidence would not validate a strategy or its expected returns.
Example: ML integrity evidence
Illustrative candidateA candidate study could examine evidence for reproducible model builds, evaluation-split provenance, training and evaluation lineage, change control, and comparison between a reviewed model artifact and a deployed one. The exact rules, detectable evidence, and abstention conditions have not been set.
Potential fit: teams with a concrete model traceability or reproducibility problem, including higher-assurance environments. Any resulting engineering evidence would not constitute regulatory approval, certification, or compliance advice.
Propose a scoped design study
You bring a concrete evidence problem, the decision that depends on it, and a domain owner who can examine representative artifacts. Before work begins, we agree the sessions, inputs, deliverables, and stop conditions. The initial output is a candidate evidence model with known limitations, abstention conditions, and false-positive costs; a pilot implementation follows only if that study supports one.
You retain code and artifacts in your environment unless we explicitly agree otherwise. Discovery email and working-session notes are still communications with Barg Labs; before any repository access, we establish confidentiality, data handling, and IP terms. Nothing produced during a study should be relied on as a production assurance, validation, or compliance determination.