Import upstream Verifiers episodes into Bench

README.md

@@ -38,6 +38,7 @@ kbench snapshot your-node        # live serving flags, for models.yaml
 
 kbench perf brain              # perf sweep only (no task authoring needed)
 kbench run brain --limit 50    # quality + perf -> results/*.json
+kbench import-verifiers brain <run-dir> --task <environment-name>
 kbench results                 # every score card recorded so far
 kbench card results/<file>     # re-render one
 ```
@@ -48,6 +49,13 @@ Raw Inspect logs land in `logs/` (gitignored). View them with:
 .venv/bin/inspect view --log-dir logs
 ```
 
+Prime Verifiers v1 `traces.jsonl` can enter the same scorecard with
+`kbench import-verifiers`. Bench reads the upstream episode artifact; it does
+not execute or copy Verifiers' harness. The importer stores episode scores,
+task-content fingerprints and short assistant excerpts, never task data, tool
+arguments or full conversations. Inspect AI remains the native runner for
+Bench tasks.
+
 ## Layout
 
 ```
@@ -58,6 +66,7 @@ kbench/
   perf.py       async load generator, any OpenAI-compatible endpoint
   results.py    the committed result schema
   run.py        orchestration: target + tasks + perf -> score card
+  verifiers_import.py  upstream Verifiers episode JSONL -> QualityResult
 data/
   public/       a few example samples, shown on the site
   private/      the real holdout — never leaves this repo