[kgryte]

Resolves a part of #11352.

Description

What is the purpose of this pull request?

This pull request:

• migrates the tests for stats/base/dists/t/pdf from relative tolerance testing (delta <= tol, where tol = {10.0,40.0} * EPS * abs( expected[ i ] )) to ULP difference testing using @stdlib/assert/is-almost-same-value.

Changes are limited to test files:

• test/test.pdf.js
• test/test.factory.js
• test/test.native.js

In each of the four fixture loops in each file, the if ( y === expected[i] ) { ... } else { delta/tol ... } branch was replaced with

t.strictEqual( isAlmostSameValue( y, expected[ i ], N ), true, 'returns expected value' );

The @stdlib/assert/is-almost-same-value require was added, and the now-unused @stdlib/math/base/special/abs and @stdlib/constants/float64/eps requires (along with the delta and tol locals) were removed. Unlike some sibling packages, EPS is not otherwise referenced in any of these three files, so it is removed from all three.

test/test.js is unchanged, as it contains no tolerance-based assertions.

ULP constants

The bound differs per fixture, matching the per-fixture tolerances it replaces. The measured minimum ULP values are:

Fixture	Previous tolerance	ULP
small_small.json	10.0 * EPS	9
large_small.json	10.0 * EPS	12
small_large.json	40.0 * EPS	52
large_large.json	40.0 * EPS	57

The same four values apply to all three converted files (test/test.pdf.js, test/test.factory.js, test/test.native.js).

Each bound was tightened by measuring the exact worst-case ULP difference over the full Julia fixture set (4 fixtures × 1001 values = 4004 values). For every bound, exactly one assertion fails at N-1, confirming each value is the minimum admissible integer bound:

Fixture	Values	Exact (0 ULP)	Max observed ULP	Cases at max	Failures at N-1
small_small.json	1001	194	9	1	1
large_small.json	1001	154	12	1	1
small_large.json	1001	74	52	1	1
large_large.json	1001	68	57	1	1

Notably, the JavaScript implementation, the factory-generated function, and the native implementation return bit-identical results for all 4004 fixture values, so the same bounds are the measured minima for all three files.

For reference, the previous 10.0 * EPS and 40.0 * EPS relative tolerances correspond to roughly 10–20 ULP and 40–80 ULP respectively, so the new bounds are consistent with the tolerances they replace.

Related Issues

Does this pull request have any related issues?

This pull request has the following related issues:

• [RFC]: Migrate math/base/special packages from relative tolerance testing to ULP difference testing (tracking issue) #11352

Questions

Any questions for reviewers of this pull request?

No.

Other

Any other information relevant to this pull request? This may include screenshots, references, and/or implementation notes.

Verification performed locally on linux/x64, Node.js v22:

• make test TESTS_FILTER=".*/stats/base/dists/t/pdf/.*" — 12057 assertions passing, 0 failing (test.pdf.js 4018, test.factory.js 4018, test.js 3, test.native.js 4018). The native add-on was compiled locally via node-gyp rebuild, so the test.native.js assertions actually executed rather than being skipped.
• The full suite was run twice at the final ULP values with identical results, confirming no FMA/architecture-dependent flakiness locally.
• make lint-javascript-tests TESTS_FILTER=".*/stats/base/dists/t/pdf/.*" — clean.
• The only files changed are the three test files.

One environment note: the lint-editorconfig-files pre-commit hook could not run locally, as it downloads the editorconfig-checker binary from a GitHub release that was not reachable from the sandbox. The changed files were instead checked by hand for tab indentation, absence of trailing whitespace, LF line endings, and a final newline — all clean, and consistent with the files' pre-existing formatting.

Checklist

Please ensure the following tasks are completed before submitting this pull request.

• Read, understood, and followed the contributing guidelines.

AI Assistance

When authoring the changes proposed in this PR, did you use any kind of AI assistance?

• Yes
• No

If you answered "yes" above, how did you use AI assistance?

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding

Disclosure

If you answered "yes" to using AI assistance, please provide a short disclosure indicating how you used AI assistance. This helps reviewers determine how much scrutiny to apply when reviewing your contribution. Example disclosures: "This PR was written primarily by Claude Code." or "I consulted ChatGPT to understand the codebase, but the proposed changes were fully authored manually by myself.".

This PR was written by Claude Code running as an unattended scheduled task. The tolerance-to-ULP conversion mirrors the idiom of already-migrated packages in the same family (e.g., stats/base/dists/chi/pdf and stats/base/dists/gamma/pdf); the ULP bound search and the local verification runs were performed by the agent.

---

@stdlib-js/reviewers

---

Generated by Claude Code

[stdlib-bot]

Coverage Report

Package	Statements	Branches	Functions	Lines
stats/base/dists/t/pdf	$\\color{green}275/275$ $\\color{green}+100.00\\%$	$\\color{green}17/17$ $\\color{green}+100.00\\%$	$\\color{green}4/4$ $\\color{green}+100.00\\%$	$\\color{green}275/275$ $\\color{green}+100.00\\%$

The above coverage report was generated for the changes in this PR.

[Planeshifter]

⚠️ Issue Reference Review

An automated check found potentially unrelated issue/PR references in this PR:

Reference	Assessment	Reasoning
#11352	suspicious	#11352 is an RFC/tracking issue explicitly scoped to migrating math/base/special packages to ULP-based testing, but this PR migrates tests for stats/base/dists/t/pdf, which is outside that issue's stated scope.

Why this matters: GitHub automatically closes issues referenced with
closing keywords (Resolves, Closes, Fixes) when the PR is merged. Incorrect
references can accidentally close unrelated issues.

What to do:

• If the reference is correct, no action needed. This check may produce
  false positives.
• If the reference is incorrect, please update your PR description.

This assessment was generated by an AI model and is informational only.

---

Generated by Claude Code