AI QA Regression Kit gives independent AI consultants and workflow builders a repeatable way to compare prompts, models and workflows before handing an update to a client.

Define the cases. Run the same suite on both versions. Record reviewed results. See which old passes became failures, then document the release decision.

The included fictional example improves from 92% to 94% case pass rate, but introduces three regressions, including one critical unauthorized tool action. The dashboard stays on HOLD and shows the exact cases to investigate.

Editable Excel template and a complete worked example with 300 illustrative run records.

Google Sheets version delivered as portable XLSX import files, with step-by-step instructions. Native Sheets conversion and recalculation checked.

50 editable starter cases covering incomplete and malformed inputs, contradictory context, prompt injection, long documents, unsupported claims, output schemas, tool selection and repeat consistency.

A 10-tab workbook: Dashboard, Setup, Comparison, Runs, Test Cases, Versions, Findings, Release, Taxonomy and Readme.

16 failure categories and Critical / High / Medium / Low severity guidance.

A 14-page PDF QA playbook, 2-page quick start, editable DOCX client report and Markdown testing instructions.

CSV case and run templates, JSON schema examples, controlled fixtures and structured worked results.

An optional offline Python helper that validates recorded JSON data and exports spreadsheet-safe run inputs.

Pass rates compare the same fully evaluated cases. A case passes only when every required trial passes. Missing evidence, duplicate run records and Blocked cases stay visible. Critical candidate failures hold the calculation gate. Approval also requires a completed evidence checklist and a named decision-maker.

Each workbook copy supports 100 cases, 600 run slots, two compared versions, 1–3 trials per version, 20 configuration records and 100 findings. Optional mean-cost and mean-latency gates require complete measured candidate data.

Use Excel 365 / Excel 2021, or import the included Google Sheets files into your own Drive and save them as native Sheets. Python 3.10+ is needed only for the optional helper. Run your AI workflow separately in your own environment, then review and record the outcomes. The kit does not execute models, provide API access or automatically judge outputs.

The $99 Solo / consultant licence covers one purchaser and unlimited comparisons in their own freelance or sole-proprietor consulting practice. Share finished reports and relevant findings with clients. Team use, resale, redistribution and client handover of the master toolkit require separate permission/licensing.

Product-use and download support are available through Contra. Custom setup, integration, running your models and audit services are not included. For damaged files or a material mismatch with this listing, contact the seller; a fix, replacement or refund will be reviewed case by case. Applicable statutory rights are unaffected.

All example outputs and scores are fictional. The buyer defines the real requirements, reviews the evidence and makes the release decision.