Framework of Spatiotemporal Traffic Flow Forecasting using Self-Evolving Graph Neural Networks: Public T_freeze Protocol and Reproducibility Package

Abstract

This record is the public T_freeze research artifact for the PhD research project “Framework of Spatiotemporal Traffic Flow Forecasting using Self-Evolving Graph Neural Networks,” conducted by Shahzaib Asghar under the supervision of Husnain Saleem. It freezes the architecture, development-stage evaluation protocol, reliability analysis, checkpoint index, bootstrap/statistical locks, Blind-B eligible pool and generator, the locked Blind-B reveal rule, the post-D7 stage status, and the frozen master evaluation engine before any Blind-B beacon reveal or final one-shot evaluation. The deposited ZIP is intended to serve as a timestamped, immutable pre-reveal evidence package. Its required SHA-256 is:b0e41920aabdf9bcc73d51d2683cc08082f827bf87c8429ce7b69ec706735afd The development program includes METR-LA, PEMS-BAY and PEMS08 evidence, multiple model families and seeds, controlled sensor/topology/regime disturbances, reliability analysis, ablations, and evolving-topology validation. D7 was prospectively amended before D7 execution to use TEAM PeMS4_Evolve as the disclosed evolving-topology substitute; the amendment, provenance, nine trajectory-level comparisons, 57 completed internal fits, and inferential result are preserved in D7_STATUS.json. The locked +1% non-inferiority test against Online-AN did not pass (one-sided p = 0.7338152919974188); this negative/mixed result is retained without post-hoc alteration. No Blind-B condition is to be generated, previewed, simulated, or evaluated until after this exact artifact has been made public and its deposited bytes have been verified against the SHA-256 above.