Truthound · All articles · 11/24
Evidence-led Truthound Core guidance on data contracts, check inputs and operational limits.
480 articles
Retain a shared Puffin file while any active deletion vector uses it
Retain a shared Puffin container when another active reference survives a deletion-vector replacement. Reconcile cleanup candidates with the full reference graph, then check offsets and contents separately.
Check Trino unsigned Parquet decoding beyond the signed boundary
Reading a Parquet unsigned integer into bigint does not establish value preservation. Reconcile keys against independently decoded annotations to detect sign changes at the signed boundary.
Test Trino timestamp casts near their representation limit
Check timestamp precision conversion near the upper bound against a reference preserving the full year and fractional seconds. Coercing inputs into a narrower date range can erase the discrepancy under investigation.
Bound Trino spooling URLs by signing-credential lifetime
Calculate a spool URL's usable delivery window from the earlier of URL and signing-credential expiry. Include delivery delays when reporting the time left to the client, and keep raw tokens out of the record.
Check reserved-name readiness in Trino metadata queries
Exercise each consumer-facing metadata query for a table with a reserved-word column. Success on another metadata path, or an abbreviated SQL description, cannot stand in for the required query parsing successfully.
Read final Pinot metadata before accepting an empty Trino split
Exclude a Pinot split from successful aggregation when final metadata reports an error, even if its data blocks are empty. Keep missing metadata distinct from a valid empty result so a low count cannot conceal a missing segment.
Declare Parquet statistic order alongside Trino min and max
Before consuming Parquet min/max, check each leaf's column_orders declaration against schema order. Passing the declaration check still leaves the numerical correctness of the bounds to independent validation.
Delta deletion vectors do not make a live file disappear from partition inventory
Reconcile partition statistics with active log entries before cleanup touches deletion-vector files. Keeping live-row counts separate from physical file counts prevents an empty-partition misclassification.
Test Trino bloom filters with the opposite sign of zero
Compare keys at one snapshot to check whether a bloom filter drops an oppositely signed zero that SQL treats as equal. Matching counts can conceal changed identities, so reconcile key sets with pruning enabled and disabled.
Separate local file durability from PyIceberg commit acknowledgment
Read the durability report within the scope of the deprecated filesystem-table commit path. In an isolated recovery test, reconcile acknowledged references with file contents and distinguish process termination from power loss.
Require Iceberg v2 snapshot fields beyond one reader’s acceptance
Check required fields for the Iceberg format version separately from reader acceptance. Verify manifest-list and summary presence in a new snapshot before proceeding to the contents of its references.
Check current-schema-id before combining Iceberg REST cleanup and schema changes
When cleanup and column addition share a REST request, the current schema reference must still resolve to a retained definition.
Reconcile acknowledged concurrent Iceberg REST changes after commit
When concurrent REST updates each receive a success response, check that fresh table state retains every acknowledged property key. Judge preservation by the identity of the changes rather than the number of successful codes.
Refresh writer identity when an Iceberg table name is reused
After recreating a table under the same name, check its UUID and actual data and manifest write targets. Matching names and schema IDs cannot approve an old factory location for the new table.
Require repeated progress from combined Iceberg maintenance tasks
Check repeated progress after the first completion when maintenance tasks share a lock. Count completions by task in a window based on the actual schedule to distinguish a stall from work that is not yet due.
Review the coordinate-axis contract behind Iceberg default CRS compatibility
Evaluate a default CRS and table CRS through the file format's coordinate order and a reviewed pair, rather than names alone. A documented compatibility convention does not establish interchangeability for every CRS.
Project non-null union branches when validating the Hudi file-group reader
Reconcile file-group projection of an Avro union with the legacy path using branch values and member representations. Success for a top-level name cannot approve an untested nested branch.
Classify missing Hudi clean instants by the sync window
Classify an absent archived clean by whether it belongs to the window required for the current sync. If required evidence is missing, reconcile a verified full partition inventory at the same boundary before declaring completion.
Test savepoint reads after Hudi cleanup, beyond file retention
Exercise the specified commit's AS OF read after cleaning, even when savepoint files remain. Keep retained inventory separate from consumer-read status to distinguish absent files from a rejected read boundary.
Keep nested delete ordering values in Hudi event-time reconciliation
Replay a late delete against a newer update for the same key. Record the nested ordering path and table version, then check the survivor against the event-time policy.
