Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 11 additions & 1 deletion docs/benchmarks.md
Original file line number Diff line number Diff line change
Expand Up @@ -76,7 +76,17 @@ The same showcase on ALFWorld, Blackjack and Millionaire is under `results/<eval

## The injection guard rail

The rail scores 20 of 20 on its labelled set at a median of 464 ms (`s1a run injection_guard`).
The shipped public fixture contains the ten original negative controls plus 1,054 InjecAgent cases and 629
AgentDojo security cases (1,703 records total). The source cases are retained as compact JSONL with their benchmark
name and source category in `note`; the original records remain the negative control so precision is measurable.

| labelled set | records | positive | negative | Jev precision | Jev recall |
|---|---:|---:|---:|---:|---:|
| Public fixture (`injection-public.jsonl`) | 1,703 | 1,693 | 10 | pending Jev key | pending Jev key |

Run the benchmark with `uv run s1a run injection_guard --slot jev --labelled-set
evals/labelled/injection-public.jsonl`. The source releases are [InjecAgent](https://github.com/uiuc-kang-lab/InjecAgent)
and [AgentDojo](https://github.com/ethz-spylab/agentdojo).

## Series

Expand Down
Loading