call_gpt_4o_with_ocr blocks the event loop with time.sleep(1)

Description

The asynchronous call_gpt_4o_with_ocr function begins its main work with time.sleep(1). Because this synchronous sleep runs on the event-loop thread, every invocation pauses unrelated coroutines and timers for one second before the model setup proceeds.

Replacing it with await asyncio.sleep(1) would preserve the intended delay without blocking the loop. If the delay is unnecessary, removing it would avoid the pause entirely. Other synchronous operations in this function may also merit review separately.

Observed Behavior

The release-source harness exercised the exact time.sleep call expression with a controlled delay of about 30 ms. The call took about 30 ms, and no event-loop heartbeat ran during it despite six expected opportunities. No external service was contacted.

Affected Version

Confirmed in self-operating-computer 1.5.8 at operate/models/apis.py:317, in call_gpt_4o_with_ocr. The current repository version retains time.sleep(1) in the asynchronous function.

Reproduction

Run the PoC below from its material directory with Python 3.10, alongside the shared _site_harness.py, inputs/sites.json, and result.json. A reproduced result records approximately 30 ms in the blocking call and zero heartbeats while it runs.

PoC Source Code

#!/usr/bin/env python3
from pathlib import Path
import sys

POC_ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(POC_ROOT))
from _site_harness import run

REPO_ROOT = next(
    parent
    for parent in Path(__file__).resolve().parents
    if (parent / "artifacts").is_dir()
)

if __name__ == "__main__":
    raise SystemExit(run(Path(__file__).with_name("inputs") / "sites.json", Path(__file__).with_name("result.json"), REPO_ROOT))