Paste your DDL. Beam audits it. Alex & Morgan narrate.

reflection/Beam-1.0-501B-A23B turns your schema definition into a structured dataset (stage 1), the deterministic DataBard rubric scores it (stage 2), and Beam writes a two-host analysis using the exact production prompts (stage 3). Beam is a thinking model — its raw reasoning streams into the open panel throughout. Runs on Reflection's free daily allowance; when it's exhausted, we say so.

Dataset

Optional research questionOutput format

Free tier: 280 generations/day shared, 8 per visitor, resets 00:00 UTC. Featured runs cache — replays are instant and spend nothing.

· 1 Parse schema→· 2 Score health→· 3 Narrate findings

The fine print

Featured datasets are DataBard's labelled demo schemas (a Uniswap V3 subgraph, an e-commerce warehouse); their audio briefings are pre-recorded production episodes.
Health scores come from the production rubric — the LLM cannot flatter them.
This same workload, graded over 24 runs, is published at /benchmarks/beam, with the measured dataset on Hugging Face.