ReadyPack

A dataset without a datasheet is a folder of CSV. ReadyPack takes your dataset plus the findings your audits already produced and hands back a publishable pack: a data-health score, a datasheet, Croissant metadata, a changelog, and a plain report.

Why this exists

You can run six honesty checks over a dataset and still ship something no one can reproduce, because the results live in six terminals and the dataset lives in a seventh. Packaging is the unglamorous step everyone skips. ReadyPack does that step and refuses to do it silently.

What it does

• Aggregates findings reports. Point it at the JSON envelopes from the other tools (EnrichAudit, LinguaCheck, SourceLock, WhyChanged, NuanceBench, FilterGap, DecayWatch) and it folds them into one report, tagging each finding with the audit it came from.
• Scores data health. A single 0–100 score with a letter grade. Penalties are log-scaled (high=6, medium=3, low=1), so the first problems cost the most and a long tail of minor notes can't sink an otherwise good dataset.
• Writes the datasheet. Composition, fields, per-audit findings, and a limitations section built from the warnings your audits emitted.
• Emits Croissant metadata. The mlcommons schema, so the pack is discoverable by dataset tooling.
• Keeps a changelog. Every packaged row and every reviewed removal is an event with a timestamp.
• Packages it up. cleaned.csv, datasheet.md, croissant.json, changelog.json, report.md, zipped into package.zip.
• Never silently deletes. Rows only leave the export when you hand over ids you already reviewed. The tool's default is to change nothing.

What it is not

It is not a checker. It does not look for leaks, gaps, PII, or bots — the other tools in the family do that. ReadyPack's job starts once you have findings and ends with a documented, packaged dataset.

Install

Python 3.10+, no runtime dependencies.

git clone https://github.com/hexsyro/ReadyPack.git
cd ReadyPack
python -m pip install .

For development:

python -m pip install -e ".[dev]"
python -m pytest

Usage

readypack dataset.csv \
  --findings filtergap-report.json \
  --findings linguacheck-report.json \
  --name "product-mentions-2026" \
  --description "Mentions collected for the launch study." \
  --drop-id r003 \
  --out ./pack

Without --findings, ReadyPack still documents and packages the dataset — it just has no audits to fold in. Without --out, it prints the report and writes nothing.

Options

Flag	What it does
--findings FILE	A findings report (JSON envelope) from another tool; repeatable
--out DIR	Write the package (cleaned.csv + docs + package.zip) to DIR
--name NAME	Dataset name for the datasheet and Croissant metadata
--description TEXT	Dataset description
--drop-id ID	Drop a reviewed row id from the packaged export; repeatable
--config FILE	JSON config; explicit flags win
--format {table,json,csv,md}	Output format (default table)
--output FILE	Write the report to FILE instead of stdout
--fail-on {none,low,medium,high}	Exit 1 if any finding is at or above this severity
--version	Print version and exit

Output (json)

{
  "tool": "readypack",
  "version": "0.1.0",
  "status": "success",
  "findings": [
    {
      "severity": "high",
      "field": "bot_confidence",
      "message": "Mention scored 0.95 bot/spam confidence (threshold 0.50).",
      "record_id": "r003",
      "metadata": {"check": "bot_confidence", "confidence": 0.95, "audit": "filtergap"}
    }
  ],
  "summary": {
    "checked": 5,
    "name": "product-mentions-2026",
    "dataset": {"fields": ["id", "text", "language", "split"], "row_count": 4, "original_row_count": 5},
    "sources": {"filtergap": {"findings": 1, "by_severity": {"high": 1, "medium": 0, "low": 0}, "score": 78.9, "grade": "B"}},
    "findings_total": 1,
    "health_score": {"score": 78.9, "grade": "B", "weights": {"high": 6.0, "medium": 3.0, "low": 1.0}},
    "drop_count": 1,
    "datasheet": "# Datasheet: product-mentions-2026\n...",
    "croissant": {"@type": "sc:Dataset", "name": "product-mentions-2026"},
    "report_md": "# ReadyPack report: product-mentions-2026\n...",
    "changelog": [{"at": "2026-10-19T00:00:00+00:00", "actor": "readypack", "event": "Packaged 4 row(s)..."}]
  },
  "warnings": ["Review every finding before publishing — ReadyPack never silently deletes rows."]
}

The pack on disk (--out) contains cleaned.csv, datasheet.md, croissant.json, changelog.json, report.md, and package.zip.

Using it in CI

readypack dataset.csv --findings audit.json --fail-on high || exit 1

Exit codes: 0 ran fine, 1 gate tripped, 2 input broken.

Library use

from readypack import run, health_score

result = run(rows, {"name": "demo", "reports": [filtergap_report]})
result.summary["health_score"]["score"]

health_score([{"severity": "high"}])  # (78.9, "B")

Limitations, honestly

• It trusts your reports. ReadyPack folds in the findings it is given. If an audit was wrong, the score is wrong. It scores findings, not truth.
• The score is a summary, not a judgment. A 92 with one high finding is still a dataset with a high finding. Read the findings, not just the grade.
• Dropping rows is manual on purpose. You pass the ids. ReadyPack will not infer them, because the whole point is that a human decided.
• Croissant output is minimal. Enough to be valid and discoverable, not the full vocabulary.

License

MIT. See LICENSE.