Nothing artificial about our intelligence.
A family of deterministic tools for regulated work. Same input in, byte-identical output out, every time. Provable, reproducible, outside the model-risk regime. Built on the Retained Learning thesis.
Each tool here is deterministic by design. Cheapest handler wins: cache, rules, embedding match, and only then AI as a controlled escalation, with a downward trajectory on AI calls per unit of work.
What You See is What You Download. Deterministic PDF table extraction with embedding-based template matching and a retained-learning tuning loop. 100% reproducible across 90 extractions, 0 errors on 2,332 cells across wild PDFs. Free tier: 10 pages per month.
Open WYSIWYD →A reconciliation and audit framework for service businesses. Engagement plane plus firm plane plus ratchet plane. Cheapest handler wins across SQL rules, LIKE aliases, firm memory, then LLM tail. In production on the CallRevu engagement.
Coming soonThe institutional memory layer underneath. Decision traces compound across engagements. Every fix becomes a future-proof rule. Quarterly audit, threshold-gated promotions, conflict resolution, rollback.
Coming soonOn April 17, 2026, the Federal Reserve, OCC, and FDIC issued SR 26-2 (guidance PDF), replacing SR 11-7 after 15 years. The new framework explicitly excludes deterministic rule-based processes and software from the definition of a "model" and therefore from the full model-validation burden.
Translation for CFOs and compliance teams: a deterministic extraction or reconciliation pipeline does not need annual model validation. The doloop family is built to qualify for that exclusion by design.
Per-token LLM costs are still falling, but per-call reproducibility is not. As enterprise auditability, SOC 2 continuous-evidence requirements, and regulator clarity all increase through 2026, the wedge for deterministic alternatives gets sharper, not duller.
