RH AI Research Starting Point v2.2: Adversarial Reproducible Audit

Connections

Regarding its relationship with other packages, I try to use the words from its own documents, not my interpretation. The W-19 content completely merged into this package is identical to its independent page and is not rendered repeatedly here.

"16 classes of rejectable error certificates; 1 class of expected-to-survive verifier collusion attack; exact Hilbert-14 floating-point false negative; external signatures, reproducible builds, and independent verifier routes." — Excerpt from the "Additions" section of the main document in this package. Deliberately designed a class of "expected to pass" attacks to verify that the audit itself is not overly sensitive.

Version Summary · README v2.2

The true change summary for this version (README_v2.2.md); this package also retains the old README files for v2.0 and v2.1, reproduced as-is without deletion. Only the version truly corresponding to v2.2 itself is rendered here.

v2.2 Core Content

Nodes, new content list, Batch 01 progress, and declarations.

Files

The complete RH_W_19_工程包_v0.1/ folder incorporated into this package is identical to the W-19 page; only the filenames are listed here for verification. W-18 and W-19 each repackaged earlier historical documents into their own records/ subfolders, which constitute the vast majority of this package's volume, and similarly, individual files are not listed repeatedly.

sha256 b47e38b8706b7d0beef2383f83eaa811d4c97e1c8380cc48f2d893b174422a67