dp-zp-agent/test/test_source_audit.py

"""Source-link availability, lifecycle qualifications and supervisor semantics."""

import io
import json
import urllib.error

import pytest

from tools.answers import reference_links as links
from tools.answers.content_selection import reference_only, evidence_phase, facets_for
from tools.knowledge.sources import profile_units
from tools.knowledge.query import answer_from_index
from tools.retrieval.person_names import resolve_person_names

REPO = "https://git.kemt.fei.tuke.sk/sh662er/Rasa"
SOURCE = "https://zp.kemt.fei.tuke.sk/students/2019/samuel_horani"


def result(text):
    return {
        "accepted_for_delivery": True,
        "status": "ok",
        "answer": text,
        "source_urls": [SOURCE],
        "claims": [],
    }


@pytest.mark.parametrize(
    "text",
    [
        "- [Repozitár s chatbotom](" + REPO + ")",
        "Repozitár: " + REPO,
        "- Výstup je tu " + REPO,
    ],
)
def test_a_bare_reference_is_not_a_completed_result(text):
    assert reference_only(text)
    assert evidence_phase(text, "progress", "result") == "unknown"
    assert facets_for(text, "progress", "Aké výsledky dosiahol?", "result") == []


def test_real_implementation_with_a_reference_remains_an_implementation():
    text = "Implementovaný chatbot používa RASA. Kód: " + REPO
    assert not reference_only(text)
    assert evidence_phase(text, "progress") == "result"


@pytest.mark.parametrize(
    "text",
    [
        "[Repozitár](" + REPO + ")",
        REPO,
        "[Repozitár][r]\n\n[r]: " + REPO,
        '<a href="' + REPO + '">Repozitár</a>',
    ],
)
def test_unverified_repository_is_not_delivered_as_a_clickable_link(text):
    value = links.checked_repository_links(result(text), check=lambda _: "not_public")
    assert REPO not in value["answer"] and value["source_urls"] == [SOURCE]
    assert value["accepted_for_delivery"] and "súkromný" in value["answer"]
    assert value["suppressed_repository_links"] == 1


def test_verified_repository_preserves_the_exact_answer_and_sources():
    value = result("[Repozitár](" + REPO + ")")
    assert links.checked_repository_links(value, check=lambda _: "verified") == value


def test_html_escaped_markdown_reference_cannot_bypass_link_suppression():
    url = "https://github.com/example/missing?x=1&amp;y=2"
    value = links.checked_repository_links(
        result("[Repozitár](" + url + ")"), check=lambda _: "unverified"
    )
    assert "github.com" not in value["answer"] and "](" not in value["answer"]


def test_unverifiable_repository_cannot_be_the_only_factual_citation():
    value = {**result("Práca je hotová."), "source_urls": [REPO]}
    checked = links.checked_repository_links(value, check=lambda _: "unverified")
    assert not checked["accepted_for_delivery"] and checked["reason"] == "reference_unavailable"
    assert not checked["source_urls"]


@pytest.mark.parametrize(
    "url",
    [
        "http://git.kemt.fei.tuke.sk/a/b",
        "https://git.kemt.fei.tuke.sk:8443/a/b",
        "https://user:password@git.kemt.fei.tuke.sk/a/b",
        "https://git.internal/a/b",
        "https://github.com/a/b?access_token=private",
        "https://git.kemt.fei.tuke.sk/../admin",
    ],
)
def test_repository_probe_never_contacts_unapproved_destinations(monkeypatch, url):
    monkeypatch.setattr(
        links.urllib.request, "build_opener", lambda *a: pytest.fail("Unsafe network request")
    )
    assert links._probe.__wrapped__(url, 0) == "unverified"


@pytest.mark.parametrize(
    "status,expected",
    [
        (404, "not_public"),
        (401, "not_public"),
        (403, "not_public"),
        (302, "unverified"),
        (503, "unverified"),
    ],
)
def test_repository_api_failures_are_not_treated_as_success(monkeypatch, status, expected):
    class Opener:
        def open(self, request, **kw):
            assert request.full_url.endswith("/api/v1/repos/sh662er/Rasa")
            assert "Authorization" not in request.headers
            raise urllib.error.HTTPError(request.full_url, status, "Unavailable", {}, None)

    monkeypatch.setattr(links.urllib.request, "build_opener", lambda *a: Opener())
    assert links._probe.__wrapped__(REPO, 0) == expected


def test_public_repo_api_metadata_is_required(monkeypatch):
    class Opener:
        def open(self, request, **kw):
            return io.BytesIO(json.dumps({"private": False, "full_name": "sh662er/Rasa"}).encode())

    monkeypatch.setattr(links.urllib.request, "build_opener", lambda *a: Opener())
    assert links._probe.__wrapped__(REPO, 0) == "verified"


def index():
    doc = {
        "title": "Alena Nová",
        "path": "pages/students/2022/alena/README.md",
        "content": "# Diplomová práca 2024\nNázov: Overené grafy\nVedúci: Daniel Hládek\n"
        "Stav:\n- Implementovaný systém RAG.\nUkončil predčasne\n",
    }
    rows = profile_units(doc)
    return {"units": rows, "version": "fixture", "index_version": {"version": "fixture"}}


def test_lifecycle_qualification_is_preserved_even_when_it_has_only_two_words():
    rows = index()["units"]
    note = next(r for r in rows if r["kind"] == "lifecycle")
    assert note["text"] == "Ukončil predčasne" and note["work_year"] == 2024
    r = answer_from_index("Ako pokročila Alena Nová?", index(), engine="rag")
    assert "Ukončil predčasne" in r["answer"] and any(
        e["evidence_role"] == "lifecycle" for e in r["evidence"]
    )


def test_supervisor_year_is_never_satisfied_by_a_historical_work():
    r = answer_from_index("Ktoré práce vedie Daniel Hladek v roku 2026?", index(), engine="rag")
    assert r["status"] == "no_answer" and r["reason"] == "supervision_not_documented"
    assert "2026" in r["answer"] and "autor dokumentu" in r["answer"].lower()
    assert not r["evidence"] and not r["source_urls"]


def test_current_supervision_is_not_inferred_from_a_formal_supervisor_line():
    r = answer_from_index("Ktoré práce aktuálne vedie Daniel Hladek?", index(), engine="graphrag")
    assert r["status"] == "no_answer" and r["reason"] == "active_supervision_unknown"
    assert "historickom" in r["answer"] and not r["evidence"]


def test_missing_supervisor_in_followup_requires_a_name_and_keeps_the_year():
    r = answer_from_index("Ktoré práce sú teraz vedené v roku 2026?", index(), engine="rag")
    assert r["status"] == "ambiguous" and r["reason"] == "missing_supervisor"
    assert "2026" in r["answer"]


def test_title_typo_is_not_misclassified_as_a_person():
    catalog = [{"name": "Daniel Hládek", "kind": "Person"}]
    r = resolve_person_names("Ktoré práce aktuálne vedie Daniel Hladek a nazv prác?", catalog)
    assert r["status"] == "ok" and [m["name"] for m in r["matches"]] == ["Daniel Hládek"]


def test_dated_status_overrides_previous_meeting_instead_of_inheriting_its_year():
    from tools.knowledge.sources import meeting_date

    assert meeting_date("Stav 8.11.2024:")["date"] == "2024-11-08"
    rows = profile_units(
        {
            "title": "Alena Nová",
            "path": "pages/students/2022/alena/README.md",
            "content": "# Bakalárska práca 2025\nStretnutie 4.2.2025\nStav:\n- Implementovaný systém RAG.\n\n"
            "Stav 8.11.2024:\n- Urobené poznámky o RAG.\n",
        }
    )
    assert next(r for r in rows if "poznámky" in r["text"])["date"] == "2024-11-08"


def test_new_lifecycle_note_does_not_reextract_unchanged_later_profiles(tmp_path):
    import hashlib
    from tools.knowledge.build import build
    from tools.sync.sync_indexes import source_files
    from tools.sync.index_state import durable_json

    wiki = tmp_path / "wiki"
    data = tmp_path / "data"
    data.mkdir()
    paths = []
    for slug, name in [("alena", "Alena Nová"), ("boris", "Boris Starý")]:
        p = wiki / f"pages/students/2022/{slug}/README.md"
        p.parent.mkdir(parents=True)
        p.write_text(
            f"---\ntitle: {name}\n---\n# Diplomová práca 2026\nStav:\n- Implementovaný systém RAG.\n"
        )
        paths.append(p)

    def version(number):
        return {
            "version": str(number),
            "source_sha256": hashlib.sha256(
                json.dumps(source_files(wiki), sort_keys=True).encode()
            ).hexdigest(),
        }

    def complete(messages):
        payload = json.loads(messages[1]["content"])
        return {"choices": [{"message": {"content": json.dumps({"items": [
            {"evidence_id":row["evidence_id"], "decision":"knowledge", "concepts":["RAG"], "relations":[]}
            for row in payload
        ]})}}]}

    durable_json(data / "index_version.json", version(1))
    build(wiki, data, complete, {"model": "fixture"}, graph=False, vectors=False)
    old = json.loads((data / "knowledge/index.json").read_text())
    old_ids = {(u["person"], u["text"]): u["id"] for u in old["units"]}
    paths[0].write_text(paths[0].read_text() + "\nUkončil predčasne\n")
    durable_json(data / "index_version.json", version(2))
    report = build(wiki, data, complete, {"model": "fixture"}, graph=False, vectors=False)
    new = json.loads((data / "knowledge/index.json").read_text())
    assert report["llm_batches"] == 0
    assert all(
        u["id"] == old_ids[(u["person"], u["text"])]
        for u in new["units"]
        if u["kind"] != "lifecycle"
    )
    assert any(u["kind"] == "lifecycle" for u in new["units"])