Truthound · Numerical and statistical validity · 1/3
Separate finite values, units and precision from statistical validity.
58 articles
upb JSON nanos: distinguish Timestamp -1 from valid negative Duration values
The same nanos field has different allowed ranges depending on message type.
SOC Hunt time discrepancies: do not adjust an already correct indexed UTC instant
Reconcile the raw offset string, indexed instant and rendered instant separately.
ProtoJSON e0 notation for large integers: quotes alone do not promise exact identity
Quotes alone do not guarantee exactness for every numeric notation.
An OpenSearch date_nanos bucket returns 0: inspect range-rewrite units before changing source data
Separate search results from the aggregation fast path.
MySQL snapshot shifts: separate DATETIME fields from the JVM calendar
Separate stored fields of a timezone-free DATETIME from the JVM calendar conversion.
Beam YAML 1d: align the day label with an actual 86,400-second window
A fixed UTC day requires 86,400 seconds; a configuration label alone does not establish that duration.
Large IDs in Debezium asNumber: a Number object does not establish exact identity
Creating a numeric object does not establish preservation of an exact identifier.
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.
Verify the UTC instant after a future DST roundtrip
A retained timezone name does not establish the same instant. Keep the local representation beside the epoch value.
Before handing off sparse unpenalized regression with a constant feature
Hand off the intercept, constant-feature, and penalty settings together. Distinguish path comparisons from an independently known simple solution.
Trace disappearing LOF outliers through the shared distance buffer
Do not treat a worker-count change as model improvement. Distinguish the retracted pickle explanation from the later BallTree reproducer.
Hand off residuals alongside the rank cutoff in large float32 regression
Record sample size, precision, and cutoff in the same run. Evaluate residual quality alongside weight consistency.
Reconcile Welch ANOVA significance after a change of scale
Retain the original data and common scale. Do not selectively report a decision flipped at a numerical boundary.
A reported distance of zero can conceal weighted tail mass
A tiny weight can retain a material contribution over a large distance. Check the meaning of zero using transport definitions and reflected controls.
Do not read a two-point regression’s zero standard error as certainty
An exact line through the points need not leave degrees of freedom for estimating residual variance.
Check information lost near a Student-t CDF of 0.5
A finite CDF within range can still lose small distinctions. Use the df=1 closed form and an error budget appropriate to the task.
Hand off a sparse matrix exponential with its unitary residual
Derive the property and norm tolerance from the Hermitian generator. A returned matrix alone cannot admit a conserved-quantity calculation.
Before admitting a finite scalar sqrtm result as a square root
Separate the scalar-input policy from an explicit 1×1 matrix. Verify the returned value against the original equation.
Admit a budget-limited trust-region step using its model cost
Compare a returned step with a feasible interior candidate using the original rank-deficient subproblem objective.
