fix(ParallelMuse): repair stale call sites that crash compressed_reasoning_aggregation

[Osamaali313]

Problem

WebAgent/ParallelMuse/compressed_reasoning_aggregation.py has three stale call sites left over from a refactor that removed data_path threading. They break the script's only entry point (main() → asyncio.run(main())), and each is provably wrong against the function's own definition (and, for get_llm_response, against the correct sibling call in the same file).

1. call_converge — wrong argument count (crashes main() on launch)

async def call_converge(sem, traj_group):          # line 236 — 2 params
...
tasks.append(call_converge(sem, filtered_cluster, data_path))   # line 273 — 3 args

call_converge is async, so the extra argument raises immediately when the coroutine object is created — the loop aborts on the first cluster.

2. get_llm_response — wrong argument count + undefined name

async def get_llm_response(messages, max_tokens):  # line 75 — 2 params
...
response = await get_llm_response(messages, max_tokens, data_path)   # line 175 — 3 args
...
response = await get_llm_response(messages, max_tokens)              # line 215 — correct

data_path isn't a parameter of call_state_report and isn't a module global, so line 175 hits NameError (and a 3-arg/2-param TypeError). Line 215 shows the intended 2-arg form.

3. call_info_integrate — inconsistent return type

if response is None:
    return "[Error getting integrated answer]"     # line 228 — bare str
...
return 0, "[No Valid Answer]"                        # 2-tuple
return len(report_group), final_answer               # 2-tuple

The caller unpacks two values — merge_num, prediction = await call_info_integrate(...) — so the bare-string error path raises ValueError: too many values to unpack (expected 2) whenever the integrate LLM call fails.

Fix

tasks.append(call_converge(sem, filtered_cluster))                 # line 273
response = await get_llm_response(messages, max_tokens)            # line 175
return 0, "[Error getting integrated answer]"                      # line 228

Verification

• AST check on the file confirms every call_converge / get_llm_response call now matches its 2-parameter definition, and all three call_info_integrate returns are 2-tuples.
• python -m py_compile passes.
• A standalone repro reproduces the three failures on the current code (TypeError / NameError / ValueError) and confirms the corrected calls succeed.

The full pipeline needs an LLM server (REPORT_CONVERGE_BASE_URL_POOL) and a rollout file, so I couldn't run it end-to-end — but these three defects are argument-count / return-type errors that are unambiguous against the file's own definitions and its own correct sibling call, and they sit on the script's only code path.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.