tests/test_chain_live.py

"""Chained test — extract_claims -> fetch_citation -> verify_dual, run for real.

Uses a specific, checkable factual claim (not a motivation-style sentence,
unlike the earlier test) attributed to a real, findable paper.

Requires GROQ_API_KEY and S2_API_KEY set in your environment.

Run from the veriscite/ root:
    python -m tests.test_chain_live
"""

import asyncio
from app.graph.nodes import extract_claims, fetch_citation, verify_dual

SAMPLE_TEXT = """
Body:
Attention mechanisms allow transformer models to capture long-range
dependencies in sequential data without relying on recurrence [1].

References:
[1] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez,
A. N., Kaiser, L., & Polosukhin, I. (2017). Attention is all you need.
In Advances in Neural Information Processing Systems (NeurIPS 2017).
"""


async def main():
    state = {
        "source_text": SAMPLE_TEXT,
        "claims": [],
        "audits": [],
        "current_index": 0,
        "report": None,
    }

    state = await extract_claims(state)
    print("STEP 1 — extract_claims:")
    for c in state["claims"]:
        print(" ", c)
    print()

    if not state["claims"]:
        print("No claims extracted — stopping.")
        return

    state = await fetch_citation(state)
    print("STEP 2 — fetch_citation:")
    cc = state["claims"][0]
    print("  resolved_paper_id:", cc["resolved_paper_id"])
    print("  evidence_text (first 250 chars):", (cc["evidence_text"] or "")[:250])
    print()

    if not cc["evidence_text"]:
        print("No evidence resolved — stopping.")
        return

    state = await verify_dual(state)
    print("STEP 3 — verify_dual:")
    audit = state["audits"][0]
    print("  winner_sentence:", audit["winner_sentence"])
    print("  deberta_verdict:", audit["deberta_verdict"])
    print("  llm_verdict:", audit["llm_verdict"])
    print("  agreement:", audit["agreement"])


if __name__ == "__main__":
    asyncio.run(main())