Live project report
Measure Data Quality
A factual view of pipeline reliability, validation results, field coverage, duplicate flags, and human correction activity. No composite score hides the gaps.
Changes over time
Weekly quality evidence
Run success means the source run completed. A clean report has zero recorded issues, zero rule-check failures, and zero record failures. The two rates use different samples.
Corrections count data-work prompts tagged as a bug fix or QA correction. Duplicate flags use the latest full-dataset report in each week. Historical report counts include reruns, and source mix can change the clean-report rate.
Current completeness
Evidence present in normalized records
Coverage answers whether a field or evidence link exists. It does not prove that the value is correct.
A source may legitimately omit a public price or listing image. Missing coverage is a review signal, not an automatic defect.
Online delivery should be represented explicitly, not treated as a broken street address.
A public source URL gives reviewers a place to verify dates, ages, prices, schedules, and ownership.
Available evidence
What the project already records
These signals can support trend reporting today without pretending they are one universal accuracy percentage.
Source runs
Status, timestamps, created, updated, expired, and record-failure counts.
Ingest reports
Required-field rules, issues, record failures, samples, review status, and manual-review requirements.
Duplicate reports
Within-source and cross-source candidate groups, affected records, severity, and approval scope.
Normalized records
Current source evidence, dates, locations, ages, prices, descriptions, registration paths, and images.
Prompt history
Human corrections tied to data work, affected area, verification performed, and outcome.
Approval gates
Pending review, local publication, Data Refresh staging, and production push remain separate states.
The important gap: field-level source agreement
KIDO can measure reliability, completeness, validation, duplicate flags, and correction activity now. A true accuracy rate needs a labeled audit sample that compares each normalized field with authoritative source evidence. That audit can later report precision by field, listing type, source shape, and time period.