WS-DREAM Dataset #1 v4: a provenance-tracked consolidation

A provenance-tracked consolidation of WS-DREAM Dataset #1 (Zheng, Zhang & Lyu,IEEE TSC 2014), built exclusively from the official Zenodo deposit(DOI 10.5281/zenodo.1133476, CC BY-SA 4.0, md5 verified before building). The QoS matrices are byte-for-byte copies of the source. The measurement isuntouched. What v4 adds is provenance. --- THREE DIFFERENCES FROM THE OFFICIAL RELEASE, and only three 1. A [position] column. The official [Service ID] cannot serve as a join key: ID 4702 is carried by three rows (positions 4700, 4701, 4702) while IDs 4700 and 4701 are absent, 5823 distinct IDs for 5825 rows. The anomaly is native, present in the 2009 release too. It is preserved as-is; the IDs are NOT renumbered. [position] is the key that actually works. 2. Six Country values corrected, the only change of value in the dataset. Five services labelled "United States" are natively "Argentina" in the 2009 release; one is natively "Japan". The original value is preserved in a [country_officiel] column, so the correction is fully reversible by anyone who disagrees with it. 3. Provenance columns: [country_status], [country_officiel], [country_v1], always populated, so every correction is auditable row by row. No coordinate was modified (5825/5825 unchanged). Six services therefore carrya Country that is inconsistent with their placeholder coordinate, "Argentina"with coordinates in Kansas. This is deliberate. No trustworthy replacementcoordinate exists, and imputing one would fabricate a value indistinguishablefrom a real observation. Visible inconsistency over silent invention. --- TWO WARNINGS FOR ANYONE WRITING CODE AGAINST THIS DATASET The service list has TWO header lines: column titles, then a "=====" separatorrow. Some circulating copies have had the separator stripped. Code that skips asingle header row will swallow the separator as if it were a service, shiftingevery service by one position relative to the QoS matrix columns, silently,with no error raised. Skip two rows, and verify afterwards that [position] runscontiguously from 0. That column exists so this class of error becomesdetectable instead of invisible. The readme.txt included here, copied verbatim from the source archive, describes the service list as having eleven columns. The file has nine. Writeyour code against the file, not against the notice. We left the readme untouchedrather than silently editing a document we did not write. --- ON THE GEOGRAPHIC LAYER The geographic fields (coordinates, AS) are a post-hoc IP geolocationenrichment, the source readme credits IPLocation for the location data andProf. Mingdong Tang for the AS data. They are not described in the canonical2014 TSC article, and the readme announcing them is dated 2016, seven yearsafter the QoS measurements themselves. More than one edition of this layercirculates, and they do not agree with each other. The QoS signal is a different matter: the response-time and throughput matricesare identical, byte for byte, across every version we have examined. Themeasurement is solid; the geography is an annotation. v4 does not fix thegeographic layer, placeholder coordinates remain in the data. It makes thelayer's provenance legible, and corrects the six values for which independentevidence exists. --- ON LICENSING The readme.txt in this deposit carries a 2016 MIT License notice. The Zenodorecord this archive came from, the deposit bearing the DOI, made by theauthors, and the one their own project page redirects to, states CC BY-SA 4.0. We follow the terms of the deposit: it is the act of distribution we actuallyrelied on, it is the later of the two, and it carries a persistent identifier.CC BY-SA imposes ShareAlike, so v4 is released under CC BY-SA 4.0 as well.Where two notices differ, we have taken the more restrictive reading. The readme is preserved exactly as received, MIT notice included. It is part ofthe archive we were given, and editing it would be a modification we could notjustify. An eleven-column variant of the service list exists elsewhere, under terms thatdo not permit redistribution. No byte of it enters v4: the build script neveropens it, and v4 contains no Continent, Region or City column. --- CONTENTS wslist.txt 13 columns (9 official + 4 added), 5825 services userlist.txt 8 columns (7 official + [position]), 339 users rtMatrix.txt 339 x 5825 response-time matrix (byte-identical to source) tpMatrix.txt 339 x 5825 throughput matrix (byte-identical to source) readme.txt original WS-DREAM readme (byte-identical to source) LICENSE CC BY-SA 4.0 full text build_v4.py the build script MANIFEST_v4_en.md the manifest, read this one REPRODUCIBILITY The build script refuses to run if the source md5 does not match, and halts onany contract violation: positional misalignment, a correction count other thansix, an encoding replacement character, a modified coordinate, a renumberedService ID, or an altered 4702 anomaly. Re-running it reproduces all six SHA-256checksums identically. v4 is reconstructible, not merely stored. The manifest documents every checksum, every deviation with its justification,and an explicit list of what was NOT done. ATTRIBUTION Original dataset: WS-DREAM Team, The Chinese University of Hong Kong. Please cite: Zheng, Z., Zhang, Y., & Lyu, M. R. (2014). Investigating QoS ofReal-World Web Services. IEEE Transactions on Services Computing, 7(1), 32-39.DOI: 10.1109/TSC.2012.34