Inspect split metadata

Find exact IDs, text duplicates, crossing groups and missing segments locally.

DemoNothing is emailed or uploaded. Synthetic inputs are prefilled for exploration.How the rules work →

Your inputs

Open a sanitised local manifestProcessed in memory. Saving a report requires a separate action.CSV or JSON split manifest *

[
  {
    "id": "doc-001",
    "split": "train",
    "group": "acme-template",
    "date": "2026-01-12",
    "segment": "invoice",
    "synthetic": false,
    "text": "Invoice total 125 dollars"
  },
  {
    "id": "doc-002",
    "split": "train",
    "group": "delta-template",
    "date": "2026-03-01",
    "segment": "invoice",
    "synthetic": false,
    "text": "Invoice total 240 dollars"
  },
  {
    "id": "doc-001",
    "split": "holdout",
    "group": "acme-template",
    "date": "2026-02-01",
    "segment": "invoice",
    "synthetic": false,
    "text": "Invoice total 125 dollars"
  },
  {
    "id": "doc-003",
    "split": "validation",
    "group": "beta-template",
    "date": "2026-01-21",
    "segment": "receipt",
    "synthetic": true,
    "text": "Receipt total 57 dollars"
  },
  {
    "id": "doc-004",
    "split": "holdout",
    "group": "gamma-template",
    "date": "2026-04-01",
    "segment": "invoice",
    "synthetic": false,
    "text": "Invoice total 860 dollars"
  }
]

JSON array or CSV: id, split (train/validation/holdout), group, date, segment, synthetic; optional sanitised text. Maximum 2 MB / 10,000 rows.Required production segments *Require holdout after training *

Inputs stay in memory until you explicitly save. Only enter sanitised material.

YOUR DECISION, WITH THE WORKING

See what the
evidence supports.

Use the example inputs or replace them with your own. Inspect the complete result, then download it with its assumptions. No email required.

inspected Scope
findings
coverage
limitations
split Plan

Inspect split metadata

Your inputs

See what theevidence supports.

See what the
evidence supports.