Explain a tiny slope in an exactly constant series
Retain direct evidence of constancy and the task’s error budget. Separate a tiny slope from the interpretation of its p-value.
Direct evidence of unchanged inputs
Use direct evidence that stored values are constant alongside the task’s slope-error budget. Otherwise a numerically nonzero slope can be reported as a real change and confuse teams. Link the tolerance and raw constancy evidence to the same execution. Slope-error tables derived from raw constancy can feed a Truthound trend-admission condition. [1, 2, 3]
The follow-up says this is not a recent regression
SciPy #25339 questions a tiny nonzero linregress slope for identical y values. A follow-up notes that this is not a recent regression and separates slope precision from the undefined correlation and p-value interpretation for constant data. Do not portray every NaN here as a new failure. [1]
Put the time unit in the table
Record constancy of raw values, the time unit, and the approved slope tolerance. Hold out-of-budget slopes from trend alerts while comparing stable centering or a reference calculation. Preserve observation count and set the budget before inspecting the result. [1, 3]
A constant 7 series implies 0
The illustrative y values are all 7 at distinct times measured in seconds, giving mathematical slope 0. The prior budget is 0.00000001. Absolute slope error 0 passes the <= comparison; 0.000001 fails. This is not a rule that automatically rejects the report’s 8.364264261336637e-22. [1, 3]
The slope-error boundary
ColumnPairValidator in Truthound Python SDK 3.3.2 compares operator-prepared absolute_slope_error and approved_slope_error_budget using <=. Operators provide raw constancy evidence, time units, and the reference slope. Core does not validate p-value meaning or classify every rounding difference as a defect. [1, 2, 3]
Hand off a separate constant state
Hand off constant inputs as a distinct state, or use verified centering or high-precision references with a declared tolerance. Existing independent slope checks may make another audit unnecessary. Preserve the reason for excluding a trend rather than silently rounding it away. [1, 3]
Application conditions
Retain raw constancy, observation count, time units, and a prior slope-error budget.
Alternatives
Classify constant inputs separately or use verified centering or high-precision references.
Limitations
The slope-error check does not decide p-value interpretation or classify every tiny nonzero value as a defect.
Evidence
- Event checked
- Subject checked
- Modified
