遇见数据集

IoTSyn v4.4.0 Smart City and Public Spaces evidence package, revision R12: stress sweeps, empirical baseline, verification audits and external-comparison data

收藏
Zenodo2026-08-03 更新2026-08-13 收录
官方服务:

资源简介:

IoTSyn Smart City and Public Spaces Edition is a public web service that generates coupled synthetic urban IoT data with known generative ground truth. Its outputs cover meteorology, thermal comfort, air quality, noise, pedestrian arrivals and aggregate presence, traffic, sensor observations, and retrospective multi-hazard decision artifacts. Researchers control all input parameters. The service operates at https://iotsyn.com/. Dataset generation requires free email registration and does not require contacting the author. This record provides the public evidence package supporting the accompanying manuscript. Contents. The package contains the public methodology specification and machine-readable contracts, including artifact schemas, a CSV data dictionary, reference test vectors, and a tolerance map. It also contains the frozen golden reference run with manifests and SHA-256 hashes; baseline and heatwave reference runs; ten distinct generated scenario profiles plus a separate UTCI clamping audit; decision-utility results across 20 seeds; and a held-out Melbourne diagnostic comparison with 30 synthetic seeds, the observed reference, and frozen source snapshots. New in this revision. This revision adds eight public-server stress runs that expand the air-quality high/severe share from 0.0% to 100.0% and the crowding share from 1.19% to 34.97%. All eight validation reports record zero contract violations. The revision also adds a day-type block-bootstrap empirical baseline; external-comparison results under both RMSE threshold policies and their decision provenance, with an explicit statement that neither policy was preregistered; disclosure of the post-result noise-channel exclusion while retaining its raw diagnostic result; a decision-utility denominator audit; a source-versus-background regime diagnostic for the Melbourne wind–PM2.5 sign; final vector artwork; and public analysis scripts for the new audits and baseline. These scripts analyse deposited results and do not implement the IoTSyn engine. Verification. PAYLOAD-SHA256SUMS.txt lists the SHA-256 digest of every other file in the extracted payload. Readers can verify the package from its extracted root with sha256sum -c PAYLOAD-SHA256SUMS.txt. The manuscript also provides a five-minute procedure for reproducing the frozen reference dataset through the public service without source-code access. Scope and limits. These artifacts establish implementation behaviour, not field validity. The held-out Melbourne diagnostic exposes limited cross-site transfer and does not establish field validity or cross-site predictive skill. The frozen reference demonstrates byte-exact reproduction only in the tested reference environment. The project specifies cross-runtime tolerance conformance but has not yet demonstrated it empirically. Terminology and provenance. The R12 external-comparison materials retain legacy predeclared terminology from an earlier manuscript draft. The updated external_comparison/METHODS.md explicitly contextualizes and corrects that terminology, while the inherited external_comparison/comparison_manifest.json and external_comparison/outputs/melbourne_2026-01_ensemble_metrics.json retain the field names predeclared_metrics and predeclared_diagnostic, respectively, for reproducibility and provenance. These labels do not indicate preregistration. evidence_updates/external_threshold_sensitivity/EXTERNAL_COMPARISON_DECISION_PROVENANCE.md records the correct terminology, both threshold policies, and the post-result timing of the noise-channel exclusion. Licensing and exclusions. The Zenodo record uses CC BY 4.0. Project-authored artifacts inherited from the earlier v4.4.0 record remain available under their original CC0-1.0 waiver. The frozen City of Melbourne snapshots under external_comparison/raw/ retain third-party CC BY 4.0 attribution. Redistributors must retain THIRD_PARTY_DATA_NOTICE.md and the provenance manifest. This record excludes the proprietary engine implementation, web-application and authentication source, deployment configuration, credentials, databases, and infrastructure material.

提供机构:
Zenodo
创建时间:
2026-08-03
二维码
社区交流群
二维码
科研交流群
商业服务