/TAG
#data-quality
24 briefs. Post it as #data-quality with this page's address. An AI that follows the link gets every brief under the tag in one fetch at /t/data-quality/index.md. Posts that carry the tag get listed here with the poster's handle: send yours to hello@videotomd.com.
- Stop AI reports inventing relationships Audit comparisons, trends, counts and sequences even when every number is real.
- Validate Feed Shapes Before Trusting Zero Inspect live responses and repair parsing before trusting an empty discovery feed.
- Compute ZIP Rent Yields With Sparse Flood Data Combine value, monthly rent, and optional flood adjustments while preserving missing and implausible results.
- Run opt-in value and schema quality probes Configure table-level value checks and compare live column types with reviewed baselines.
- Reconcile Primary Facts with Secondary Crosschecks Reconcile source coverage and surface proxy disagreements without letting them overwrite primary facts.
- Reconcile Coverage Against Record Geography Separate missing geographic mappings from out-of-footprint records before widening dataset coverage.
- Probe real records before building a pipeline Check keys, fields, pagination and runtime on a small pull before committing to ingestion.
- Parse signed percentages without hiding conflicts Accept signed percentages, reject currency-shaped strings, and expose contradictory signs.
- Register a Corridor Across Profile and City Registries Keep corridor identity, alias resolution, city membership, and counts consistent during onboarding.
- Guard Fraction and Percentage Swaps Prevent a source replacement from silently changing a share by a factor of one hundred.
- Convert Listing Totals to Unit Prices Establish whether a price covers one item or the entire listing before comparing values.
- Map a dataset to a constrained schema Separate observed mappings from grounded inference, normalize values, and validate output.
- Implement a Deterministic ZIP Resolver Resolve scope and county from static data, preserving unknown classifications and stable place choices.
- Hold Data With Conflicting Measure Definitions Block release when accurate values may represent a different measure from the one the destination consumes.
- Check Geographic Coverage of Latest Views Expose a latest-record view that silently drops part of its intended service area.
- Gate batches with content contracts Quarantine invalid rows before merge; abort only when contamination exceeds an explicit threshold.
- Assign Table Values by Header Geometry Map numeric cells to header positions so extra columns cannot silently shift field assignments.
- Classify Off-Market Listing Transitions Resolve listing departures without mistaking pending closures or old sales for current outcomes.
- Backfill Missing Listing Dates Safely Recover missing listing dates in resumable batches without generating listing activity.
- Audit Profiles With Documented Exceptions Separate complete profiles, justified nulls, and missing coverage without inventing measurements.
- Gate Catalog Promotion on a Clean Live Cycle Keep a new data product provisional until its live pipeline, independent parity, and output checks pass.
- Execute a Bounded Data Repair Establish authority, reject stale plans, change only the intended field, and verify outcomes separately.
- Validate API filters before trusting absence Check that a filter narrows results before treating an empty result as evidence.
- Freeze a Universe with a Venue Liveness Screen Remove duplicate numeraires and structurally favored pegged assets, then freeze venue-active candidates.