Batch Output Regression Lab
Compare up to 100 cases across ten output versions using sixteen explicit validators and forty worked examples. Inspect missing outputs, introduced failures and resolved failures.
Loading your tools…
Projects stay in this browser. Save a backup before changing devices. Use the export format offered by each tool.
From your inputs to a usable result
01
Retain cases and output versions
Add up to one hundred cases and ten versions, paste exact outputs or import a local JSON batch. A missing output stays distinct from an intentionally empty one.
02
Define explicit requirements
Choose sixteen deterministic validators, set global or case-specific rules, and compare every version with a selected baseline. Forty fictional worked examples show expected checks.
03
Inspect changes and keep the evidence
Review the complete matrix, introduced and resolved rule failures, and exact-text changes. Export CSV findings, TXT reviews, JSON audit reports or an editable project backup.
Before you start
Where is my project saved?
Projects are saved in this browser for this site account. Download a JSON backup before changing devices or clearing browser storage.
Is an AI account required?
No. The text transformations and inspections run in your browser. The app does not send text to an AI service.
Will my original text be replaced?
The tool keeps the source in the project and shows a separate preview. Export the preview only after checking it.
Does passing prove an answer is correct?
No. The app checks explicit mechanical requirements such as exact text, counts, placeholders, JSON structure and decimal bounds. It does not score factual truth or semantic quality.
Can a version pass when an output is missing?
Missing outputs and outputs with no enabled rules are separate statuses. They are never counted as passes or as comparable improvements.
What changes are compared?
All versions use the same current rules. The report shows overall regressions/improvements and individual introduced/resolved failures, including changes hidden behind an unchanged failing status.
Are my original outputs changed?
No. Outputs remain exact saved text. Reports describe results separately. Any change to the batch invalidates the previous report until you run it again.
What are the limits?
One hundred cases, ten versions, one thousand output pairs, sixteen global rules plus sixteen rules per case, twenty thousand characters per output and eight MB of combined UTF-8 output text.
Does the app upload my outputs?
Evaluation and projects stay in this browser. Owner Save as Template deliberately uploads the supplied inputs and outputs as site content, so use fictional or cleared examples.
Can I resume later?
Reopen the saved project on this browser and account, or restore its JSON backup. Keep backups in a place you trust because they include your supplied material.