Property Data Quality Audit and Source Reconciliation
Audit the full property dataset for identity, completeness, duplicates, units, dates, source lineage and agreement between published formats.
Define the record and the field before auditing
List the datasets, responsible owners, fields, formats and intended decisions. Define what makes one record unique: asset identifier, vendor entity, account, observation geography and period, or another appropriate key. A row count alone does not describe the scope or completeness of a database.
Create a field dictionary with meaning, units, allowed values, source and date interpretation. Separate observation period, publication date, review date and last edited date. Updating a file today does not make its observations current.
Scan the complete dataset for structural defects
Check every row for required fields, type, ranges, malformed dates, inconsistent units and broken references. Use explicit missing values with a reason where possible; do not convert unknown quantities to zero. Record dataset coverage and exclusions before describing an audit as complete.
Identify exact duplicates and possible duplicates separately. Two entries with the same display name can be distinct entities, while different spellings can describe the same entity. Verify stable identifiers and source records before merging, and preserve a change log and references to the retained record.
Verify important facts against sources
Prioritize decision-critical or time-sensitive fields for source review, then schedule the rest by risk and change frequency. Use primary sources where available and retain the specific supporting page, document or internal record. Record unresolved conflicts rather than selecting the more convenient value.
For public indicators, preserve geography, period, definition and release basis. Reconcile sums, shares and conversions with documented formulas. County totals should not silently become metro-area totals, and preliminary observations should remain identifiable when later revisions are possible.
Keep formats and pages synchronized
Compare CSV, JSON, tables, summaries and downloadable files using the same record key and field definitions. Treat formatting differences separately from factual disagreement. Test derived totals and categories, including whether interval boundaries overlap or leave gaps.
Generate recurring counts and metadata from the actual dataset where practical. A manually maintained 'number of resources' can drift after new content is added. Include internal-link and source-link checks in the publication process, while recognizing that a working source URL is not proof of the accuracy of every field.
Worked example: a backlog aging table
Hypothetical example: one category covers 61–90 days and another covers 90 days and above. Day 90 belongs to both. Revise the mutually exclusive boundaries, verify the reporting date and apply the same definitions to the table, CSV, JSON and documentation.
The audit record states the old definition, new definition, affected calculations and validation. Historical reports should retain their original basis or clearly identify a recalculation instead of being silently rewritten.
Property Data Quality Audit and Source Reconciliation checklist
A reusable control-definition worksheet. Add property-specific owners, dates, evidence links and status before using it as an action record.
Download the CSV checklist to assign an owner, add dates, and record the evidence for your property.
These rows define suggested controls, not completed property findings. Add a row ID, owner, due date, status and evidence link to your working copy. See the working-copy instructions and reuse terms.
12 checks shown
No check matches those filters. Try a different term or area.
| Review area | Check | Evidence to request | Decision question |
|---|---|---|---|
| Definition | Inventory all datasets and formats | Dataset register | Which files and pages are in scope? |
| Definition | Define stable record keys | Schema and identity rules | What makes a record unique? |
| Fields | Document units and allowed values | Field dictionary | What does each field mean? |
| Fields | Separate observation and review dates | Date definitions | How current is the underlying observation? |
| Structure | Scan every row for required fields | Validation report | Where is data missing or malformed? |
| Structure | Preserve unknown values | Missing-value rules | Has uncertainty been turned into a false zero? |
| Identity | Review exact and possible duplicates | Duplicate candidates and sources | Are the records actually the same entity? |
| Identity | Keep merge and correction history | Change log | Can a correction be traced? |
| Sources | Verify critical facts and geography | Primary source record | Does the source support the actual field? |
| Sources | Reconcile calculations and boundaries | Formula and category tests | Are totals and intervals coherent? |
| Publication | Compare CSV JSON and displayed tables | Cross-format reconciliation | Do published versions agree? |
| Publication | Refresh counts links and metadata | Publication check report | Has the whole content system been updated? |
Frequently asked questions
Should missing values be filled with estimates?
Only if the estimation method is appropriate and explicitly labeled for the decision. Preserve unknown values when no reliable basis exists.
Does a source link prove freshness?
No. Verify the observation and release dates and whether the source supports the specific field. A recent review can confirm an older historical observation.
When is merging duplicate records appropriate?
When identity is verified, relationships can be preserved and the retained record has a traceable merge history. Similar display names alone are insufficient.