`datasignature`¶

datasignature computes a deterministic, TabDat-native SHA-256 fingerprint of the active dataset. It is useful for recording a reproducibility baseline before a scripted analysis or checking whether a data preparation step changed the values or schema.

Syntax¶

datasignature

The command accepts no varlist, options, if clause, assignment, or by: prefix.

What is covered¶

The versioned signature includes:

public column names and canonicalized logical types, in schema order;
active row order; and
every cell value, including explicit null and non-finite-value encodings.

It excludes session-local labels, panel metadata, source paths, backend choice, and execution mode. This is a TabDat-native data fingerprint, not a byte-for-byte Parquet checksum or a compatibility promise for Stata, SAS, or SPSS signatures.

Execution behavior¶

The command is read-only. Eager and DuckDB-lazy data are scanned in bounded row batches. Polars-lazy uses bounded collect_batches scans and keeps the existing lazy plan active. The active dataset, labels, panel metadata, named tables, and materialization state are unchanged.

Output¶

Human output reports the algorithm, scanned row count, public-column count, and a 64-character lowercase hexadecimal signature:

Data signature
Algorithm: sha256
Rows: 3
Columns: 4
Signature: 3b7e...

JSON output uses DatasignatureResult:

{
  "schema_version": 1,
  "result_type": "DatasignatureResult",
  "data": {
    "algorithm": "sha256",
    "signature": "3b7e...",
    "row_count": 3,
    "column_count": 4
  }
}

An empty dataset is valid and returns zero rows with a schema-dependent signature. Running the command without an active dataset uses the standard no-active-dataset error.

Examples¶

use data/analytic.parquet, lazy engine=polars
datasignature

Save the JSON signature in a run manifest and compare it before rerunning an analysis. A changed value, null, schema, or row order produces a different signature.

datasignature¶