Repository navigation
Release per-row detector decisions for the WAITI 2026 paper - #125
Closed
architsharm wants to merge 2 commits into
Closed
architsharm wants to merge 2 commits into
architsharm wants to merge 2 commits into
Conversation
benchmarks/releases/waiti2026/ holds one JSONL file per dataset the paper scores. Each row has the gold label and every tier's decision and stored score: code, PIGuard, ProtectAI, the judgment model and the LLM. Each row also has the dev/test half used to choose thresholds and, for the injection sets, whether its text is in PIGuard's training data. Rows are keyed by the dataset's own ID or by source file and index at 723fcec, plus a SHA-256 of the normalised row text. No dataset text is included. CoCoNot (conflicting licences on its card) and R-Judge (no licence) are left out per row; their aggregates stay in the paper. verify.py (stdlib) recomputes the headline numbers and checks them against the paper's generated macros and appendix cells: 97 checks pass.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
…ipts; no SQL text)
Owner
Author
|
Moved to https://github.com/architsharm/waiti2026-artifact (standalone artifact repo). |
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
benchmarks/releases/waiti2026/has the per-row detector decisions behind the WAITI 2026 paper "No Tier Wins Everywhere". There are 19 files inrows/, one per dataset, about 16 MB in total; SafePyramid is gzipped. Each row has:aggregates/holds the counts the paper reports where per-row data was not stored, such as Table 3 and the AgentDojo provenance summary.prompts.jsonhas our prompts verbatim. The README maps each file to the paper's tables and documents the fields, thresholds, sources and licences.No third-party text
Rows are keyed by the dataset's own ID where it has one. Otherwise they are keyed by the source file and its 0-based index at
723fcec, plus a SHA-256 of the normalised row text so others can join the rows to their own copy. No dataset text is included.I checked each source's licence on its card or repository. Two datasets are not released per row:
Their aggregate numbers are unchanged.
How to verify
The script uses only the standard library. It recomputes the headline numbers and compares them with the paper's generated macros (
paper_numbers.tex) and with the appendix cells (aggregates/camera_ready.json). Current result:97 checks passed, 0 failed.This adds files under
benchmarks/only, so the vendored wheels don't need a rebuild.