On 102/144 eval tasks, `test.sh` pipes `grader.py --json` to `reward.json`. Its null `tool_efficiency` (no tool-call count is written) makes Harbor reject the whole reward, correctness included, for every answered trial; only empty submissions grade as 0. The other 42 tasks write a plain number to `reward.txt` without `--json` and are unaffected.
This copies the already-fixed test-split `grader.py` byte for byte to all 144 tasks. Only `tool_efficiency` is removed; `grade()`, tolerances and normalisation stay the same. `test.sh` and all other files are unchanged.
### Testing
JSON after the change: correct `{"correctness": 1.0, "submission": 1.0}`; empty `{"correctness": 0.0, "submission": 0.0}`.
