Truthound · All articles · 20/24
Evidence-led Truthound Core guidance on data contracts, check inputs and operational limits.
480 articles
NULLs after a DataFusion Parquet Variant merge: check presence by layout
Distinguish a field missing in the source from a reader-created NULL caused by representation handling.
DataFusion UNION ALL: validate declared schemas and executed batches separately
At export time, check that the schema observed in planning matches the type contract of the batch actually received.
After adding a computed column to a DataFusion sort, recheck export order
A sort in the plan does not establish that the final exported rows remain globally ordered.
Before a DataFusion Decimal comparison, check the integer-digit budget
A large precision value alone does not mean a promoted type has room for the required integer part.
DataFusion keys after GROUP BY: validate joins at the aggregate grain
An original primary key may no longer be unique by itself after aggregation, allowing repeated contributions to disappear.
When ClickHouse shardNum shifts: validate both the range and server mapping
An integer-typed shard number does not by itself establish the provenance of distributed results.
Missing ClickHouse shards: check responding servers, not just DNS
Successful connections and queries can still read the same shard repeatedly and produce an incomplete aggregate.
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.
Why ClickHouse NATS checks must separate messages from rows
A successful message receipt does not establish complete landing when one message expands into several rows.
ClickHouse MATERIALIZED CTEs: verify the final consumer still receives data
A successful first read does not establish that later consumers receive the same shared intermediate result.
Reused ClickHouse–Iceberg column names: compare filters by field ID
A familiar name can attach a current predicate to statistics belonging to a different historical field.
Iceberg binary truncate counts bytes: checking ClickHouse partition prefixes
Bytes resembling UTF-8 characters still require byte-based truncation in a binary partition.
DecisionTree check_input=False: a validation flag cannot replace the missing-value mask
Skipping input checks requires knowing which information the skipped stage normally creates for the algorithm.
scikit-learn class_weight string keys: preserve the type of numeric-looking labels
Parsing a string label as an integer can create a different dictionary key even when the printed characters look identical.
An all-NaN partial_fit batch is not permission to erase accumulated statistics
Treating a feature's lack of new observations as invalidation of earlier statistics can turn later finite inputs into NaN.
Interpreting RFECV support: selected count is not the original feature axis
A support result for one subset is still a vector over all original input features. Do not use an incorrect shape annotation as an aggregation contract.
KBinsDiscretizer on near-constant features: detect a final bin count of 0
Include features with a tiny positive range, as well as truly constant features, in discretization tests.
NCA rejects string labels: preserve class meaning through an encoding workaround
Check supported inputs and a reversible mapping before changing label representation.
Checking input ownership when IterativeImputer changes the original missing values
If imputation is allowed to change only the output, the source missingness mask must survive. Check that shared-data contract across the call.
HGB errors on all-missing features: check observation semantics before filling with 0
Replacing unobserved values with arbitrary numbers to avoid a training failure changes the data's meaning.
