Splitline by SEAI

03 / Splitline field manual

Inspect row datasets

JSONL, CSV, and TSV share one row-shaped workflow. Use the file rail for dataset context, then move among Rows, Schema, Issues, and Raw without changing the source.

Updated

Organize loaded files

The file rail shows every loaded file, its inferred split, row count, size, parse-issue count, and unsaved-edit state. Choose All, train, valid, test, or other to filter the rail; select a file to make it active.

The masthead summarizes all loaded row files with a health score, total rows, fields, splits, and flags. The score is a pointer, not a verdict: use Issues to read the evidence behind every deduction.

Active file and loaded dataset are different scopes.

Rows, Schema, search, the row inspector, and Raw describe the active file. Issues checks every loaded row file and compares their schemas.

Browse and search rows

  1. Choose a file

    Select it in the rail. Rows displays 25 records at a time, with column widths negotiated from the visible data.

  2. Search every field

    Type in Search every field or press ⌘ K. Search scans top-level and nested values in the active file, then paginates the matching rows.

  3. Open the evidence

    Select a row to open the inspector. Expand fields to read long text, nested objects, and arrays, or copy the complete row as formatted JSON.

Splitline Rows view showing train and validation files, field-profile table headers, and a selected record in the row inspector Inspect full resolution
Select a record to keep its complete shape beside the searchable table.

Read Schema and Issues

Schema profiles fields across the active file. For each field it shows the inferred type, coverage, unique-value count, numeric range or string-length range, and most common value. Switch files to compare their profiles directly.

Splitline Schema view showing inferred field types, coverage bars, unique counts, ranges, and common values for the active file Inspect full resolution
Schema is deliberately file-scoped: the heading names how many fields and rows belong to the active file.

Issues checks the loaded row dataset as a group. It keeps valid records available while reporting:

  • malformed or unsupported records with their file and source location;
  • fields missing, added, or changed in type relative to the first loaded row file;
  • mixed value types within a field; and
  • files with no valid rows.
Splitline Issues view listing a malformed record, schema drift, mixed field types, and the loaded files involved Inspect full resolution
Issues spans the loaded row files so cross-split drift stays visible beside per-file parse problems.

Inspect Raw source

Open Raw when you need the exact retained JSONL, CSV, or TSV text. For a complete file at or below 1 MiB, this view preserves untouched whitespace and malformed lines. Pending edits are applied with the same patch operation that Save will use, so the preview matches the eventual output while untouched lines remain byte-identical.

Raw is unavailable when:

  • a Finder load stopped at a size, record, issue, or field-value budget, because a partial file retains no complete source; or
  • the complete retained source is larger than 1 MiB, because mounting it would no longer be an inspection-sized operation.

Stay in Rows, Schema, and Issues for those files. Splitline does not reconstruct parsed records and call the result “raw.”

Check split inference

Splitline infers the split from file and path names. A split word in the filename wins over a split-like parent folder.

Split Recognized examples
Train train.jsonl, training-records.csv
Validation validation.jsonl, valid.tsv, val.csv, dev.jsonl
Test test.jsonl, test-records.tsv
Other Any supported row file without a recognized split name

These rules are identical for JSONL, CSV, and TSV. The format tag changes; profiling, issue checks, Rows, editing, and Analyze do not branch into a different workflow.

Guide screenshot · full resolution