Audit finite-energy KL on stochastic path measures

Expand generation recurrence to parametric error families

Add universal generation recurrence certificate

Audit learned score scope for diffusion bounds

Expand influential-set path correlations

Audit influential-set bounds on non-Gaussian paths

Broaden diffusion collapse claim scope

Widen Gaussian claims to exact time-inhomogeneous paths

Replace the source-pinned claim-6 page with an independent 10-D Gaussian-mixture self-consuming-loop experiment

Remove duplicate claim-6 page (merged into the canonical one)

Claim 6: run the 10-D Gaussian-mixture self-consuming loop independently (was source-pinned only); alpha=0 degrades 9.4x vs 1.2x at alpha=1 over 80 generations

Reconcile Hub LFS rules for diffusion model collapse reproduction

Repurpose zero-loss duplicate for diffusion model collapse reproduction

Publish independent ICML 2026 reproduction: Finding Most Influential Sets

Upload app.py

Delete app.py

Upload 4 files

Delete test_app.py

Delete search_engine.py

Delete requirements.txt

Delete app.py

Upload 4 files

Delete app.py

Delete requirements.txt

Delete search_engine.py

Delete test_app.py

Upload 4 files

Delete test_app.py

Delete requirements.txt

Delete search_engine.py

Delete app.py

Upload 4 files

Delete requirements.txt

Delete test_app.py

Delete search_engine.py

Delete app.py

Upload 4 files

Delete requirements.txt

Delete app.py

Delete test_app.py

Delete search_engine.py

Upload 4 files

initial commit