FACT LANTERN is an offline evaluation workbench for teams who want to review customer-support answers rather than collect another generic prompt pack. Pair a standard-library Python CLI with 60 editable practice scenarios, a fictional store policy, a five-dimension human scoring rubric, sample answers and reviewer-entered scores, and a self-contained HTML summary report.

HOW IT WORKS

Create an answer-entry template, add manual or model-generated drafts, record human scores for accuracy to scenario, policy alignment, empathy, clarity, and privacy/escalation, then build a report you can open locally. Reports summarize reviewer-entered values only, exclude blank ratings from means, flag risk-marked answers without completed reviews, and show scenario context beside each answer. There is no API call, subscription, automatic truth judgment or pass/fail score.

WHAT IS INSIDE

Python source with seven automated tests; 60 scenarios across ten store-support categories; policy and rubric documents; editable sample CSVs; a generated fictional 12-case report; quick-start guide; three real-content PNG previews with editable HTML sources; and the buyer license. Cases cover shipping, address changes, returns, damage, warranty, privacy, payment claims, unknown product facts and order verification.

REQUIREMENTS

Python 3.10+ and a modern browser. Standard library only; no third-party runtime packages, API key, model account or network connection required. Files are editable on Windows, macOS and Linux with suitable Python installations.

IMPORTANT LIMITS

This is a reviewer aid, not a chatbot, factuality detector, compliance certification or legal/safety tool. The Northstar Supply merchant, policies, cases and sample answers are fictional. Replace them with organization-approved current policy before operational use. Use synthetic or de-identified cases, never customer secrets, and apply your own privacy and escalation requirements.

FREQUENT QUESTIONS

Does it score automatically? No. Human reviewers enter every score; the CLI validates formats and calculates descriptive summaries only.

Can I use my own cases? Yes. Keep the documented CSV columns and unique IDs, and adapt the policy and rubric to your approved standards.

Does it send data anywhere? No. The runtime makes no API or network calls. You remain responsible for local access controls.

LICENSE

Single-buyer personal and internal commercial use and adaptation allowed. No resale, sublicensing or redistribution of the source bundle as a standalone or competing product. No exclusivity or results guarantee. See included LICENSE.txt. Digital download only; no evaluation service or support subscription included.