Repository navigation
A Distributional Audit of a Public Julia Comparison Artifact for Weighted Reservoir Sampling with Replacement #139
Description
Original article: A. Meligrana and A. Fazzone, “Weighted Reservoir Sampling with Replacement from Data Streams,” WWW 2026, pp. 8789–8792. https://doi.org/10.1145/3774904.3792966
PDF URL: https://github.com/mharinat/weighted-reservoir-distributional-audit/blob/main/Research_Manuscript.pdf
Metadata URL: https://raw.githubusercontent.com/mharinat/weighted-reservoir-distributional-audit/main/metadata.yaml
Code URL: https://github.com/mharinat/weighted-reservoir-distributional-audit
Scientific domain: Computer Science — randomized algorithms and sampling
Programming language: Julia, C++ and Python
Suggested editor: Nicolas P. Rougier, subject to editorial availability and assignment.
Author: Mohammad Haribat, Independent Researcher, Kuwait. Correspondence: mharibat@mew.gov.kw.
This submission is a partial replication accompanied by a source-informed distributional audit of the pinned public Julia comparison artifact. An independent C++ implementation preceded source inspection. Exact counterexamples establish two implementation defects; local repairs are evaluated with native synthetic experiments and selected reproducibility checks. The original real-data experiment and historical environment are not reproduced, and the submission does not claim that the formal algorithms or the released StreamSampling package are incorrect. Editorial assessment of this partial-replication scope is requested.
The technical supplement is available at https://github.com/mharinat/weighted-reservoir-distributional-audit/blob/main/Technical_Supplement.pdf . The repository provides a downloadable Research_Artifact.zip containing the complete source, synthetic data, protocols, licenses and reproduction instructions. Extract it and follow submission/Reproduce.md. Seven selected Python analysis scripts regenerate eight checked outputs; native sampling and timing are separate procedures.
The author reports contacting the original corresponding author on 4 October 2026. Delivery has not been independently verified, and no reply has been reported.
The author confirms sole authorship, no dedicated funding, no related conflicts of interest, no prior collaboration with the original authors and no concurrent submission elsewhere. AI assistance in code development, analysis and drafting is disclosed in the manuscript. The author has reviewed the manuscript and takes responsibility for its content.
Activity
mharinat commentedon Oct 7, 2026
Update — 7 October 2026
A reply from the original corresponding author has now been received. The initial submission's statement that no reply had been reported is therefore superseded. Private correspondence is not reproduced here.
We performed targeted checks of the public comparison/samplers.jl at upstream commit f5eef18df8c59792e4fd5dda7aebfa1dd6780c87 (fixes branch). Across 18 configurations and 180,000 coupled retrievals, the patched converter produced no output mismatch against our existing repair. Five distributional checks (one million conversion queries, three count-helper cases with one million draws each, and 200,000 short-stream retrievals) did not reject their reference laws after Holm correction. These are post-submission diagnostics, with fixed-reservoir limitations explicitly documented; non-rejection is not a general correctness or equivalence proof. No benchmark timings were rerun, and other package changes remain outside this validation.
Report: https://github.com/mharinat/weighted-reservoir-distributional-audit/blob/main/Upstream_Fixes_Addendum.md
Pinned source, scripts, environment and results: https://github.com/mharinat/weighted-reservoir-distributional-audit/blob/main/Upstream_Fixes_Verification.zip
The originally submitted manuscript and its earlier pinned source are retained for provenance. This addendum updates the correspondence status and provides independently executed evidence on the later public fixes.
mharinat commentedon Oct 9, 2026
Upstream integration update — 9 October 2026
The upstream fixes have now been merged into main in JuliaDynamics/StreamSampling.jl PR #154: JuliaDynamics/StreamSampling.jl#154 . GitHub records the merge at 2026-10-09 01:02:32 UTC, with merge commit 69cc82a4a979853c6e54249dc02e75d8e701a8b8. The public PR description credits @mharinat for reporting the WRAExp-J retrieval bias and the binomial-CDF off-by-one error. We claim no credit for the other fixes.
We downloaded comparison/samplers.jl at the merge commit and compared it byte-for-byte with the file at f5eef18df8c59792e4fd5dda7aebfa1dd6780c87 tested in our 7 October addendum. They are identical; both have SHA-256 fde8a4ef460c07035e81d1b316b4506e68193f6a7b1af07e8c0898d836d8f458. Thus the previously reported targeted checks concern the same comparison-file contents now merged upstream. No new experiment was run for this identity check, and it does not validate other changed files or establish that a new package release has been published.
Pinned merged file: https://github.com/JuliaDynamics/StreamSampling.jl/blob/69cc82a4a979853c6e54249dc02e75d8e701a8b8/comparison/samplers.jl
The submitted study remains an audit of the earlier pinned artifact; the upstream correction documents its subsequent practical outcome.