What’s in a version
A version is built from per-component artifacts. Each component (column stats, EBM model, EBM graphs, UMAP, etc.) writes its own outputs and registers them in a singlemanifest.json. The pieces that the UI most directly consumes:
Surprise finding writes its top-N output directly to the
summand-task-outputs DynamoDB table — not to S3 — so the Insights tab reads from there. Predictions for single rows are produced on demand by loading the model. Batch scoring a whole dataset writes the results as a new scored dataset.
All S3 artifacts are stored under a versioned path (s3://summand-artifacts/semantic-layers/{datasetId}/versions/{version}/) so older versions remain readable.
Why a computed-context layer
Every part of the UI — and the Summand assistant — needs to answer questions like “What’s the mean ofrevenue by segment?” or “Which features matter most?” without re-querying the source. The computed-context layer is the cached, structured answer to those questions.
Concretely:
- Summand gets grounding facts from the computed-context layer instead of hallucinating numbers.
- Charts suggested by Summand reference real columns and aggregations from the computed-context layer.
- Refresh comparisons become straightforward — diff version N against version N–1.
Versioning model
A version is(datasetId, version_number). Versions are append-only: a new component run creates a new version and does not change the previous one. Refreshing or re-uploading data doesn’t create a version by itself; the next component run does.
Time travel
The Overview tab shows the latest version by default. Older versions stay readable, and in chat Summand can read a specific older version or compare versions, which helps you track how a model’s understanding of the data has shifted over time.Storage and retention
Computed-context versions don’t count toward tier-based dataset caps. Summand keeps every version while the dataset exists, on every tier.Relationship to data warehousing
The computed-context layer is not a metrics store like Cube or LookML. It’s narrower:- It describes a single dataset, not a curated cross-source business model.
- It’s generated, not authored — you don’t write
.ymlto maintain it. - It exists to power Summand’s analysis surface; it doesn’t expose itself as a SQL endpoint.