1.28 MB
11 files
Updated 18 days ago
Ctrl+K
| Name | Size | Uploaded | Xet hash |
|---|---|---|---|
| GATE_REPORT.json | 203 Bytes xet | 82339808 | |
| README.md | 803 Bytes xet | 08b35335 | |
| checksums.json | 632 Bytes xet | 3038fdde | |
| claim_diagnostics.png | 329 kB xet | 3d55efb5 | |
| make_upgrade_v2_figures.py | 7.23 kB xet | 81d06629 | |
| poster_v2.png | 302 kB xet | 9f669271 | |
| raw_measurements.csv | 4.35 kB xet | 9e883dc5 | |
| reproduction_bundle_v2.zip | 584 kB xet | ebdeacb6 | |
| results.json | 26.3 kB xet | 3979c776 | |
| source_audit_v2.json | 2.56 kB xet | 6af76d27 | |
| upgrade_v2.py | 22.6 kB xet | 0ed55e3b |
Full audit v2
This directory contains the complete machine-readable evidence for the
clean-room reproduction of Understanding Behavior Cloning with Action
Quantization (OpenReview 9uENnRAcSl, arXiv 2603.20538).
Run:
python upgrade_v2.py --scale 1 --output-dir upgrade_v2_outputs
python make_upgrade_v2_figures.py
results.json records every aggregate, raw_measurements.csv exposes the
underlying sweep rows, GATE_REPORT.json records pre-registered pass/fail
criteria, and source_audit_v2.json distinguishes source confirmation from
the two registered-claim errors. The full run is deterministic at seed
20260727; timing varies by machine.
The numerical experiments check consequences and explicit source constructions. They do not replace the paper's theorem proofs.
- Total size
- 1.28 MB
- Files
- 11
- Last updated
- Jul 26
- Pre-warmed CDN
- US EU US EU