Synthetic proof run

See what happens to a messy client export.

This example uses a synthetic UK orders dataset, not client data. It shows the difference between a raw export, the cleaned output, the rejected rows, the flagged rows, and the signed delivery package.

Compare rows Download samples

Before and after

Same business record, different operational value.

The point is not just changing formats. The service separates usable rows, rejected rows, and rows that need operator/client judgement.

Raw export

Messy input row

order_id: ORD-100777
order_date: 31/02/2025
quantity: -1
phone: withheld

Reviewed output

Cleaned or retained with evidence

order_date: 2025-02-28
quantity: correct from client evidence
status: retained_after_review
audit trail: flag reason preserved

Clean

Ready rows

Rows that satisfy the agreed schema and policy continue into the cleaned dataset. Sensitive fields follow the configured keep, mask, hash, or remove policy.

Rejected

Rows stopped

Rows with blocking issues are separated into rejects with machine-readable and plain-English reasons, instead of silently polluting the clean output.

Flagged

Rows needing judgement

Ambiguous records can be retained only with an operator/client decision and audit trail, so repeat runs do not hide judgement calls.

What the report proves

Every output file has a reason to exist.

The synthetic run produced the same delivery structure used for a paid-client workflow: clean data, quality files, reports, manifests, and signature evidence.

Screenshot of the synthetic CleanLLM run report showing run details and executive summary — Synthetic report preview captured from the local validation run.

data/clean.parquet Validated retained rows for downstream use.

quality/rejects.csv Rejected records and reasons.

quality/flagged_rows.csv Rows retained after explicit review.

reports/report.html Human-readable summary for review.

integrity_manifest.json SHA-256 hashes for delivery contents.

integrity_manifest.sig Signature evidence for the package.

Open the samples

Small synthetic files you can inspect.

These files are intentionally tiny extracts for the website. They demonstrate the categories of output without exposing live client data.

Messy input CSV Clean output CSV Rejects CSV Flagged rows CSV

Why this is different from manual spreadsheet cleanup

The value is repeatability: the next weekly or monthly export can be checked against the same approved rules instead of relying on someone remembering every manual edit.

Start with your own export shape

Send a short description of the recurring file, the manual fixes, and the outcome you need. If the file is a fit, Audit Ready Data will issue a signed upload link.

Start fit check Review handling