[atlaie]

Summary

Found while testing #9566 end to end in Colab: when a policy refuses or holds back a call, the evaluator gets messages that point to the wrong place, and the DO doesn't see an api's arguments or a job's parameters. This only changes messages and displays; no behaviour changes.

Changes

• job.output_paths: a call refused by a policy prints the reason and points to job.policy_error, not to job.stderr (which the evaluator doesn't get). Other failures still point to job.stderr.
• job.output_paths: a done job whose outputs the output policy held back says so, and that the data owner can still share them, instead of blaming the script.
• A blocking api call prints why it ended (the refusal reason, or that the outputs were held back), through a small _print_call_result helper.
• The api table has a Policies column (input, output, input, output or -), and the text view shows the same.
• ApiResult lists the api's arguments and says when set_policies cleared the state; print(result) now shows the same text as repr(result).
• The job view shows code/params.json in a Parameters section, pretty-printed and escaped.

Testing

• Ran uv sync --all-packages, just test-unit-job (157 passed), just test-unit-rds (155 passed), uv run pytest tests/unit/syft_bg (174 passed) and pre-commit run --all-files (passes).
• Added capsys checks for the output-policy and rate-limit refusals, a held-back test and a Policies column test in syft-rds/tests/test_api_policies.py; tests/unit/syft_bg/test_api_result.py for the args and the cleared state; a Parameters test in syft-job/tests/test_policies.py. The new tests fail on dev.
• tests/integration/syft_bg/test_cli.py gives 24 passed and 2 failed; the 2 remove-peer failures happen the same way on dev without this change (the test sets up an api without run.sh), so I left them alone.

Asana task

• https://app.asana.com/1/1185126988600652/task/1219834421197793

🤖 Generated with Claude Code