What they found
The authors built SynthCorp, a platform that simulates how business events create and change company data over time. Each simulated data estate mixes relational databases, a document store and a data catalog, with test tasks and runnable answer programs. They made seven data estates (up to 1K operational tables) and 256 tasks across five domains. The abstract says the data passes automatic checks, such as foreign-key violations and database execution failures, with a near-zero error rate. Tests of three models in four data-context settings showed the benchmarks separate data contexts and model tiers. The authors also check answers against a log of the simulation, which verifies the answer programs without manual labeling, including on tasks that combine several events and expose reasoning failures in state-of-the-art agents.
Why it matters
Many current tests for SI data agents (systems that work with company data) are quickly maxed out and cannot probe specific skills. A tool that makes customizable, checkable tests could help people see where agents fail on messy business data. Skipping manual labeling could also make such tests cheaper to build. The abstract gives no detailed scores, and the benchmarks are not yet released.
How it changes the race
If the approach holds, teams building enterprise SI agents get a cheaper way to make harder, targeted tests without human annotators. It is an evaluation tool, not a capability gain, and its value depends on the promised benchmark release.
The abstract, by the authors
Current benchmarks for LLM-driven data agents saturate quickly and cannot target specific capabilities. We present SynthCorp, a platform that creates customizable data estates and benchmarks by simulating how business events produce and modify corporate data over time. Each data estate combines relational databases, a document store, and a data catalog, paired with evaluation tasks and executable gold programs. First, given a scenario and target complexity dimensions, SynthCorp synthesizes reusable event types and compiles each into a program over predefined simulation primitives. Second, SynthCorp schedules event instances on a virtual calendar, interleaving routine and disruption events so that tables, documents, and catalog entries remain consistent. Third, SynthCorp executes the calendar to materialize the data estate and records every operation in a simulation log. Fourth, SynthCorp generates analytical tasks and derives verified gold programs from the log. We use SynthCorp to generate initial evaluation benchmarks comprising seven data estates (up to 1K operational tables) and 256 tasks across five domains. The data estates pass deterministic pipeline checks (such as foreign-key violations, undeclared writes, recurrence violations, and database execution failures) with a near-zero error rate, introduce cross-asset complexities, and cover all requested dimensions. Evaluating three models across four data-context settings shows that the benchmarks separate data contexts and model tiers. Auditing candidate queries against causal provenance in the simulation log verifies gold programs without manual annotation, including on multi-event composite tasks that expose reasoning failures in state-of-the-art agents. We hope the community uses SynthCorp to generate more challenging and useful data estates and benchmarks that foster the development of enterprise data agents.
Quoted from arXiv:2611.14765. The research and its words belong to its authors.
