Styxis
Company
Knowledge

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.

Finiteness versus useful resolution

A finite CDF within range can lose the resolution an analysis needs. For df=1, use the independent closed form and an approved error budget. Analyses of small effects may require that distinction; tasks that do not need it should not inherit the same precision cost automatically. Truthound can compare a CDF error derived from an external closed form with the task budget. [1, 2, 3]

The df=1 scope of the report

SciPy #25667 reports stdtr quantization near t=0 for df=1, including outputs fixed at 0.5. The report discloses AI-assisted investigation with limited checking, and a follow-up adds high-precision negative-t controls. The claim does not extend to the same failure at every degree of freedom. [1]

An error budget from the purpose

Record df, t, the observed CDF, closed-form reference, and tolerance. Hold interpretation of small effects when the error exceeds the budget, then reconsider precision or the calculation path. Operators prepare the actual error and independently approved limit, preserving calculation status separately. [1, 3]

An independent closed-form reference

The illustrative budget is fixed in advance at 0.000000001. Compute the reference for df=1 using 0.5+atan(t)/pi. An absolute error of 0 passes the <= comparison; 0.000000003 exceeds the budget and fails. Increasing the limit after observing the error does not establish preservation of a small effect. [1, 3]

Core compares the error table

ColumnPairValidator in Truthound Python SDK 3.3.2 compares operator-computed absolute_cdf_error and approved_error_budget using <=. Operators establish t, df, and reference precision. The budget is a task requirement, not a universal CDF standard; Core does not calculate stdtr or atan or verify a patch. [1, 2, 3]

Do not extend the formula to other df

For df=1, compare a verified closed-form path or a high-precision reference. Do not extend that expression to other degrees of freedom. If independent regressions already cover the required region and error budget, another audit can be unnecessary. [1, 3]

Application conditions

Join df, t, an independent closed-form reference, and a prior error budget by evaluation ID.

Alternatives

Use a verified closed form limited to df=1 or a high-precision reference.

Limitations

  • The task budget is not a universal CDF standard, and the check does not execute stdtr or atan.

Evidence

Event checked
Subject checked
Modified