A read on the provenance layer under Agentic Risk Intelligence. And where the March announcement meets the spring ship.
The March 11 announcement put traceability, citations, and Agent-to-Agent interoperability at the center of the story. The spring rollout has to carry those claims in front of analysts and federal reviewers who will test them line by line. This page is a short read on the layer the product sits on, and the first conversation we would like to have.
Most agentic rollouts over-index on the model and under-index on the layer above it.
We work on a simple idea. About 60% of what an analyst workflow needs is traditional code and database work. About 30% is rule-based logic. Around 10% is a genuine AI problem. The job is sorting those three into the right layers before anything is shipped, and documenting the sort so a federal reviewer can follow it.
Applied to the Babel Street stack, the map resolves quickly. Entity resolution and multilingual analytics (Rosette heritage) are a code-and-model layer. Source ingestion across hundreds of PAI and CAI feeds is a database-and-pipeline layer. Analyst tradecraft rules (what to elevate, when to flag, how to cite) are rule-based logic. The genuinely agentic surface is narrower than the announcement reads and more defensible when the boundary is drawn on purpose.
The provenance evidence a federal customer needs during ATO review is produced by that boundary being drawn cleanly, not by a post-hoc annotation pass over model outputs. That is the sort of work we would propose to do with the product and AI team.
A multi-agent platform for DRC mining compliance with the same structural problem as an OSINT investigative workflow: high-stakes output, long source chains, regulator-grade citation requirements, and a human in the final loop. We drew the boundary between deterministic documentation and agentic reasoning at the start of the build. Compliance work that had consumed more than 160 hours per cycle collapsed to under five minutes with provenance intact.
- Compliance narrative generation Analyst investigative brief with cited sources
- Regulator review trail ATO / IC review evidence for agentic outputs
- Multi-agent orchestration layer Agent-to-Agent handoffs inside a single investigation
The methodology is published.
Interpretable Context Methodology: Folder Structure as Agent Architecture, submitted to ACM TiiS. The core claim is that agent context can be organized as a layered filesystem with measurable interpretability and reproducibility gains. Relevant whenever an agentic product has to stand up to outside review. Open source under MIT license.
Repo: github.com/RinDig/Interpretable-Context-Methodology-ICM-
Who is on the other side of the table.
- 01Jake Van Clief, founder. Marine Corps veteran, eight years, cryptographic systems and F-35/F-18 avionics. MSc Future Governance, University of Edinburgh.
- 02Published in ACM TiiS (ICM) and arXiv (Ethics Engine, a psychometric assessment tool for ideological and moral patterns in LLMs).
- 031,500 plus enterprise learners trained since May 2025 across Correlation One (Pacific Life, Colgate-Palmolive) and KPMG UK, one of the Big Four.
- 04Eduba partners with NLP Logix for work that sits below the orchestration layer. NLP Logix has been in machine learning since 2011 and runs over 150 data scientists.
Thirty minutes with Matt Van Clief.
Bring the one agentic workflow from the spring rollout that you most need to defend in front of a federal reviewer. We will walk through the 60 / 30 / 10 map live, identify where the provenance evidence has to originate, and scope the first engagement from there.
If the conversation lands somewhere else, we will say so on the call.