| """Four budgets bound a run, and the run gets the smallest of them. |
| |
| Raising one of them is indistinguishable, from the outside, from raising none. |
| That is not a hypothetical: the harness limit was raised from three to ten, |
| every test passed, and the next benchmark behaved exactly as before, because |
| the worker's hard cap and the server's per-run config were both still three. |
| |
| So the numbers are asserted to agree here rather than trusted to stay in step. |
| If a future change means one of them genuinely should differ, this test is the |
| place to say so, in writing, on purpose. |
| """ |
|
|
| from __future__ import annotations |
|
|
| from distinct_agent.harness import StructuredToolHarness |
| from distinct_agent.tools import HARD_TOOL_LIMITS |
| from distinct_server.ui import PER_RUN_TOOL_LIMITS |
| from distinct_tools.core import JobPolicy |
|
|
| EXPECTED = 10 |
|
|
|
|
| def test_every_budget_that_bounds_a_run_agrees_on_the_number() -> None: |
| assert StructuredToolHarness.hard_max_tool_calls == EXPECTED |
| assert HARD_TOOL_LIMITS["max_calls"] == EXPECTED |
| assert PER_RUN_TOOL_LIMITS["max_calls"] == EXPECTED |
| assert JobPolicy(allowed_tools=frozenset()).max_calls == EXPECTED |
|
|
|
|
| def test_the_number_is_high_enough_for_the_request_that_exposed_it() -> None: |
| """Plan, convert, scale, write the document, build the spreadsheet, answer.""" |
|
|
| assert EXPECTED >= 6 |
|
|
