Sources behind the study.

Canonical dataset access, action indexing, and one-shot split reference.

Supports equal search-budget disclosure and reporting all development trials, eligibility and compute.

Published source for supervised contrastive learning; v3 declares the single-view temperature-0.07 pose adaptation as a descriptive comparator.

Image evidence motivates limiting synthetic-stress claims; it does not establish pose robustness.

Motivates explicit sources of variability and conditional uncertainty; does not validate this exact bootstrap assumption.

Provides precedent for class-disjoint development validation within the NTU one-shot setup.

Defines the contextual objective, controlled metric-learning comparisons, and the limits of the original robustness evidence.

Specifies the pretrained pose encoder and its NTU RGB+D 120 one-shot evaluation context.

Grounds the use of Gaussian noise and joint or frame masking as pose-robustness stress tests.

Public-safe progress, verified corrections, SCC status and remaining Week 5 admission requirements.

Binding local prospectus and scope authority for the project.

Defines holiday, class-end, study-period, and final-exam constraints on the schedule.

Historical 26-configuration v2 benchmark, preserved for provenance and future work. It is outside the December v3 study and does not authorize v3 execution.

Final December execution design: three fixed recipes, 20 epochs, three seeds, nested corruption, conditional analysis and explicit capacity/test-opening gates; decision 0004 records the scope change.