LabLedger helps student researchers question a spreadsheet before using it. This edited walkthrough uses real browser captures and synthetic narration. All twenty demonstration records are fabricated. Analysis runs locally, without an account, API key or data upload.

Choose identifier, measurement, group and unit columns. North readings near twenty and South readings near seventy need separate baselines. The local Isolation Forest fits sixty four random partition trees per group and unit, with at least eight numeric readings.

This fixture flags five records: two duplicate identifiers, a missing reading, nonnumeric text and an injected reading of one hundred fifty. The anomaly score is a review suggestion, not proof of an error. The threshold is a prototype heuristic.

Each flag needs an explicit keep or exclude decision and a reason. In this synthetic example, the reviewer keeps the original duplicate identifier and excludes four deliberately injected problems. Export stays locked until review is complete.

The downloaded CSV contains sixteen kept records. The audit JSON retains all twenty records, including exclusions and reasons. Thirteen core tests and real download checks pass. Codex assisted with implementation and documentation. This one measurement prototype is a review aid, not a certification of scientific validity.