Results, types, and limits

The workbench and conversation previews have different result sizes. Do not use a conversation preview limit as a workbench transfer limit.

Workbench results

The workbench transfers source results in full rather than silently using the first 10,000 rows. For example, BigQuery pages are fetched in batches until the reported result is complete. Other SQL connectors are asked for uncapped results, subject to their own limits.

When a source reports a total row count that does not match the returned data, FastBI fails the transfer rather than silently joining an incomplete prefix. Result columns must have unique names; alias duplicate names before running.

The result viewer is paginated. It displays up to 500 rows per page; this is a display-page size, not a total query limit. Large queries can still fail because of source capacity, query size, or transfer limits.

Analysis previews

Lake analysis retains up to 5,000 rows and marks larger results as truncated. Conversation previews have additional analysis-depth and byte limits.

For complete aggregate results, aggregate the intended data in SQL. A small preview does not mean the query scanned a small amount of data, and increasing LIMIT does not increase the conversation’s preview budget.

Types across sources

FastBI preserves supported column types when combining results. Integers and supported explicit decimals can be represented exactly rather than being forced through floating-point storage. Decimal precision above the supported bound, or an unspecified scale, may remain text to avoid guessing or rounding.

Arrays, JSON, binary, spatial, interval, and other non-portable source types can be transferred as text representations. That is different from a native Lake table that already contains structs, arrays, or maps. Use the complex-type reference only when the actual column type supports it.

Inspect result column types before joining or aggregating. Cast compatible values explicitly, preserve identifier formatting, and do not assume numeric-looking IDs are interchangeable with integers.

Temporary result lifetime

Workbench result relations have a 24-hour lifetime . If a result has expired, rerun the query before downloading or saving it. Saving the SQL text does not pin the prior result forever.

Save to Lake

Use Save to Lake on an available completed result. Dataset and table names must start with a lowercase letter, contain only lowercase letters, digits, or underscores, and be at most 32 characters each.

Saving creates a private dataset and retains permissions for every contributing source. Existing destination tables are not overwritten; choose another name if there is a collision. A saved result is a snapshot of that execution, not a scheduled refresh.

Cancellation and retries

Use Cancel to stop active work. Results are only published after the query completes successfully. A source query already submitted may still have consumed provider resources, so cancellation is not a guarantee that all costs are reversed.

Failed stages report an error or reference ID. Check the cause before rerunning: retries can perform new work and create new usage. Source access is checked again during execution and when derived results are read.