TattvaDocs
Core concepts

Sources

The data behind every answer, how Tattva reads it, and what you can see and change on a source.

A source is data Tattva can read: a file you uploaded, a BigQuery table, or a written document. Every figure in an answer comes from a source.

Kinds of source

  • Files: XLSX, XLS, CSV, TSV, JSON, Parquet. Uploaded on Sources › Upload.
  • BigQuery tables: connected on Sources › BigQuery, usually by a data analyst. Tattva reads the table's structure; your questions run in BigQuery.
  • Documents: .md or .txt files such as metric rules, meeting notes or SOPs. Tattva reads definitions from metric documents, and decisions and conventions from the others. See Write a metric document.
  • Workspace sources: sources from the Workspace Library, marked Workspace. Every project sees them.

The Sources page

Sidebar → Sources. Tabs: All, Upload, Apps, BigQuery.

All sources

On All:

  • All Sources (N) and a Search files… box.
  • Filter by content chips narrow the list by what a source contributed, for example DATA or GLOSSARY_TERMS.
  • Each row shows the name, type, number of versions, status and health, and what the source contains.

Click a source to open it.

Processing

A new source or version goes through Parsing, Analyzing, Indexing and Ready. While it runs, the source shows the current step, for example "Mapping columns to concepts". A failure shows Pipeline failed — check logs.

During this, Tattva profiles every column, works out what each one means, links columns to business concepts, and proposes new definitions for your data analysts to review.

A source's page

The header shows the name, version and description. Tabs: Overview, Schema, Lineage, Queries.

Source overview

Overview

  • Summary and Key Findings: what Tattva thinks the source contains.
  • Inspector:
    • Granularity: one row per what, for example daily or per event. You can correct it.
    • Key columns: the identifiers and dates. You can add or remove them.
    • Evaluated by: which analysis ran, and when.
  • Gotchas: data issues worth knowing, such as many blank values in a column. Each has a severity: error, warn or info. You can acknowledge one (stop flagging it until the next upload) or suppress it (never flag it on this source again).
  • Versions: when a source has several versions, pick one to see it, and compare it with the previous one.

Schema: every column with Type, Role, Aggregation, Meaning, Nullable, Cardinality and Unique %. Click ✎ on a column to correct its meaning, role, unit, sensitivity or aggregation. Edits take effect after the next ontology build.

Schema tab

Lineage: the definitions and concepts in the ontology that came from this source.

Queries: example questions this source can answer. Click one to ask it in Reasoning.

Actions on a source

  • Edit changes the name and description for this project. The original stays and you can Reset to original.
  • ↑ Promote to workspace asks a data analyst to make a project source available to every project. You give a reason and click Submit request. The source shows "waiting for Data Analyst review" until someone decides.
  • ↻ Re-run analyze (BigQuery tables only) reads the table again and creates a new version.
  • Turn proposals off stops builds from proposing concepts and disambiguations from this source and from inferring relationships that touch it. Pending disambiguation cards that name the source are withdrawn from Review. Concepts, relationships and disambiguations that already exist stay. The button then reads Proposals off · turn on. Only the project that owns the source can change this; a workspace source switched off in the Workspace Library shows "proposals off in the Workspace Library" in every project. The project-wide settings are on Settings › Project › Ontology build (see Build the ontology).

Trust labels

  • Workspace-trusted: reviewed by a data analyst and shared across projects.
  • Project: used only in this project. Answers that rely on it carry a caveat.
  • pending review: a promotion request is waiting.

Versions

Uploading a file with the same name creates a new version of that source. Tattva keeps the earlier versions and shows what changed in the schema.

Tips

  • Upload raw rows, not summaries. One row per order gives Tattva more to work with than a pivot table.
  • Correct column meanings early. A wrong meaning leads to wrong joins and filters.
  • Use BigQuery for large data. Queries run in the warehouse; nothing is copied.
  • Read the gotchas before you trust a new source.
Sources — Tattva Docs