Truthound · Storage and recovery consistency · 3/3
Check rows, values and metadata before and after storage.
55 articles
Validating distinct counts after variable-offset Polars rolling slices
A selected sequence can lose its original sort order even when cached metadata says otherwise.
Checking Iceberg delete positions before trusting a head limit
A delete at position 100 cannot address a four-row file; validate the metadata before checking output size.
Checking Delta deletion-vector references when encoded paths disagree
Separate object identity from deletion application when only escaped partition paths return deleted rows.
Check whether Iceberg WAP data is visible on main before approval
Actual branch visibility and the existence of the WAP snapshot define the approval boundary more clearly than a stage label.
Check that an Iceberg UUID composite delete is attached to scan tasks
A committed delete file can exist without reaching the reader's applicable scan plan.
Iceberg nested equality deletes: verify survivors as well as deletion targets
Preservation of nontarget rows is a separate acceptance condition, not an optional consequence of deleting the intended keys.
When Iceberg fixed-binary delete keys collapse, inspect buffer ownership
Compare the distinct keys written to the file with those retained in the reader's in-memory set.
Iceberg expireSnapshots inside a transaction: check files before catalog commit
A successful maintenance call is less informative than whether still-published snapshots retain available file references.
After copying an Iceberg DeleteFile, verify equalityFieldIds survive
A copied path and successful commit do not establish that the metadata defining deletion semantics was retained.
DuckDB ENUM Parquet distinct_count: allowed values are not observed values
The number of declared ENUM choices is different from the number of distinct values actually written to a file.
Validate a ClickHouse patch recovery through the merge after ATTACH
Seeing the updated value immediately after reattachment is not enough to close a lightweight-update recovery.
Dask and Zarr parallel writes: inspect shard ownership before trusting a readable file
Overlapping partial updates to one shard can separate successful completion from pixel preservation.
Investigating Dask task fusion when independent chains receive the same key
Preserve the mapping between logical and executable work before relying on output-file existence.
Checking value parity for V4 Arrow IPC union subset reads
Bind full-read and subset-read controls to the same file, metadata version, union kind and row identity.
Airbyte MariaDB CDC restart failures: inspect the TableMap context behind the offset
A restart offset must identify a decodable event boundary, not merely an increasing number.
