russell@unturf.com
26e3af9033
Empirical finding 2026-05-01 on 'explain spin glass modeling, tensors?': returned STRICT 1/1 with claim 'Spin glass modeling involves... mathematical tools such as tensors' cited to a chunk from the *Quantum chromodynamics* article. Token-coverage check inside the chunk passed accidentally on shared physics vocabulary (spin/model/etc.), but the article isn't about spin glasses — the claim is a retrieval-driven hallucination. Root retrieval cause: AND-mode FTS5 returned zero hits (Spin_glass article has 0 occurrences of 'modeling' or 'tensors'); OR-mode fallback ranked by raw BM25, so 'Glass' (glass*274) outranked 'Spin glass' (spin*29 + glass*15) despite Spin_glass matching 2 distinct query tokens vs Glass's 1. Verifier then accepted the hallucinated claim because the cited QCD chunk has incidental physics vocabulary overlap. New ``diagnose_title_relevance(claim_text, cited_source_titles)`` sidecar: claim's content tokens (stemmed) must overlap the union of cited source titles. Zero overlap → ``title_mismatch``. Catches the spin-tensors case (claim about spin glass, cited title 'Quantum chromodynamics' → zero stem overlap) without false-positives on legitimate cross-document grounding (Lois_Lane article cited for a Lois-Lane claim still has lois+lane overlap). Sidecar-only for now — staying out of the binary verifier output until bench evidence justifies hard-promotion. Not yet wired into inspect_cache_key (per-claim citation source mapping isn't in providence_cache yet) or the bench. Available as a utility for operators auditing specific (claim, source_title) pairs. 5 new unit tests covering match/mismatch/no-titles/no-claim-tokens and underscore-normalization for Wikipedia titles.
606 lines
24 KiB
Python
606 lines
24 KiB
Python
"""Sidecar diagnostic for unverified spans.
|
|
|
|
`aborist inspect --cache-key X` reads a providence_cache record, pulls
|
|
the same source chunks the verifier saw, and classifies each
|
|
unverified_quote into one of: verbatim_in_base, verbatim_in_raw_only,
|
|
trailing_artifact, paraphrase, partial_paraphrase, no_overlap.
|
|
|
|
These tests pin the classifier on synthetic contexts so the diagnoses
|
|
are deterministic. End-to-end coverage of `inspect_cache_key` (DB
|
|
plumbing) goes through a small fixture record below.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import time
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
|
|
from aborist.qa.inspect import (
|
|
_classify_span,
|
|
_normalize,
|
|
diagnose_deflection,
|
|
diagnose_title_relevance,
|
|
inspect_cache_key,
|
|
)
|
|
from aborist.store import append_audit, connect, transaction
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# _classify_span — pure function, no DB
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
def _norm(s: str) -> str:
|
|
return _normalize(s)
|
|
|
|
|
|
def test_classify_verbatim_in_base():
|
|
"""If the verifier flagged a span but it IS in the base context, the
|
|
diagnosis surfaces that as a likely verifier or canonicalization
|
|
bug. Deliberately distinct label so an operator notices."""
|
|
span = "the quick brown fox"
|
|
base = "lorem ipsum the quick brown fox jumps over"
|
|
raw = base
|
|
out = _classify_span(span, _norm(base), _norm(raw))
|
|
assert out["diagnosis"] == "verbatim_in_base"
|
|
|
|
|
|
def test_classify_verbatim_in_raw_only():
|
|
"""Span matches raw wikitext but base form differs — wikitext-strip
|
|
edge case worth flagging separately."""
|
|
span = "[[Cloud Strife]] is the protagonist"
|
|
base = "Cloud Strife is the protagonist" # wikitext stripped
|
|
raw = "[[Cloud Strife]] is the protagonist"
|
|
out = _classify_span(span, _norm(base), _norm(raw))
|
|
assert out["diagnosis"] == "verbatim_in_raw_only"
|
|
|
|
|
|
def test_classify_trailing_artifact_citation_appended():
|
|
"""The Pikachu case — model appended `(Source: ...)` to a verbatim
|
|
sentence; verifier flagged the whole thing. inspect should isolate
|
|
the artifact."""
|
|
base = (
|
|
"Pikachu can store electricity in its cheeks and release it in "
|
|
"lightning-based attacks. Pikachu evolves from Pichu."
|
|
)
|
|
span = (
|
|
"Pikachu can store electricity in its cheeks and release it in "
|
|
"lightning-based attacks. (Source: https://en.wikipedia.org/wiki/Pikachu)"
|
|
)
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "trailing_artifact"
|
|
assert out["matched_prefix_chars"] >= 60
|
|
assert "Source:" in out["trailing_artifact"]
|
|
|
|
|
|
def test_classify_interior_elision_drops_parenthetical_aside():
|
|
"""Fox 2026-04-29 catch (Clark Kent): source has
|
|
`Clark Joseph Kent (middle name is also Jerome ...) is a fictional
|
|
character...`; model quoted `Clark Joseph Kent is a fictional
|
|
character...` — every token in source, but a `(...)` aside got
|
|
elided for prose flow. Distinct from trailing_artifact (model
|
|
APPENDS) and paraphrase (different sequence). Sidecar must
|
|
surface this as `interior_elision` with the dropped aside reported."""
|
|
base = (
|
|
"Clark Joseph Kent (middle name is also Jerome according to some "
|
|
"versions) is a fictional character created by Jerry Siegel and "
|
|
"Joe Shuster. He serves as the civilian and secret identity of "
|
|
"the superhero Superman."
|
|
)
|
|
span = (
|
|
"Clark Joseph Kent is a fictional character created by Jerry Siegel "
|
|
"and Joe Shuster. He serves as the civilian and secret identity of "
|
|
"the superhero Superman."
|
|
)
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "interior_elision"
|
|
assert out["matched_prefix_chars"] >= 17 # "clark joseph kent"
|
|
assert out["matched_suffix_chars"] >= 60
|
|
assert "jerome" in out["dropped_aside"].lower()
|
|
|
|
|
|
def test_classify_interior_elision_falls_through_when_suffix_doesnt_match():
|
|
"""Probe is conservative: suffix must land verbatim after the close
|
|
paren. A model that drops the aside AND rewords the rest should NOT
|
|
classify as interior_elision."""
|
|
base = (
|
|
"Clark Joseph Kent (middle name is Jerome) is a fictional character "
|
|
"created by Jerry Siegel and Joe Shuster."
|
|
)
|
|
span = (
|
|
"Clark Joseph Kent is the most famous superhero in comics history "
|
|
"and was invented by Jerry Siegel."
|
|
)
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] != "interior_elision"
|
|
|
|
|
|
def test_classify_synthetic_elision_caught():
|
|
"""Fox 2026-04-30 (Brachiosaurus / Jurassic Park): the model wrote
|
|
a `"..."` quote with literal `[...]` between fragments, signaling
|
|
self-elision while claiming verbatim citation. Distinct from
|
|
`interior_elision` (model dropped a `(...)` aside source carries).
|
|
Sidecar reports prefix/suffix presence in source."""
|
|
base = (
|
|
"The film centers on the fictional Isla Nublar, in Costa Rica, where "
|
|
"billionaire philanthropist John Hammond has created an amusement park "
|
|
"of cloned dinosaurs. Universal Studios acquired the rights."
|
|
)
|
|
span = (
|
|
"The film centers on the fictional Isla Nublar [...] Universal Studios "
|
|
"acquired the rights."
|
|
)
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "synthetic_elision_inside_quote"
|
|
assert out["elision_marker"] == "[...]"
|
|
assert out["prefix_in_source"] is True
|
|
assert out["suffix_in_source"] is True
|
|
|
|
|
|
def test_classify_synthetic_elision_does_not_fire_when_source_has_brackets():
|
|
"""If `[...]` literally appears in source (e.g. a citation
|
|
formatting), the substring check would have passed earlier — sidecar
|
|
falls through to its other diagnoses."""
|
|
base = "Some prose with [...] literal brackets in source."
|
|
span = "Some prose with [...] literal brackets in source."
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "verbatim_in_base"
|
|
|
|
|
|
def test_repair_synthetic_elision_split_when_both_halves_verbatim():
|
|
"""Repair plan for `"prefix [...] suffix"` where source carries
|
|
both halves verbatim: split into two quotes."""
|
|
base = (
|
|
"The film centers on the fictional Isla Nublar, in Costa Rica. "
|
|
"Universal Studios acquired the rights to the novel before publication."
|
|
)
|
|
span = (
|
|
"The film centers on the fictional Isla Nublar [...] Universal Studios "
|
|
"acquired the rights to the novel"
|
|
)
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "synthetic_elision_inside_quote"
|
|
assert out["repair"]["action"] == "split_into_two_quotes"
|
|
assert len(out["repair"]["quotes"]) == 2
|
|
|
|
|
|
def test_repair_interior_elision_includes_aside():
|
|
"""Repair plan for parenthetical-elision: rewrite quote to include
|
|
the dropped aside so it becomes verbatim."""
|
|
base = (
|
|
"Clark Joseph Kent (middle name is also Jerome according to some "
|
|
"versions) is a fictional character created by Jerry Siegel and Joe Shuster."
|
|
)
|
|
span = (
|
|
"Clark Joseph Kent is a fictional character created by Jerry Siegel "
|
|
"and Joe Shuster."
|
|
)
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "interior_elision"
|
|
assert out["repair"]["action"] == "include_aside_for_verbatim"
|
|
assert "jerome" in out["repair"]["aside_to_restore"].lower()
|
|
|
|
|
|
def test_repair_trailing_artifact_trim():
|
|
"""Repair plan for model-appended `(Source: ...)`: trim the tail.
|
|
Trailing-artifact probe requires ≥60 char matching prefix, so the
|
|
test uses a long-enough prose span."""
|
|
base = (
|
|
"Pikachu can store electricity in its cheeks and release it in "
|
|
"lightning-based attacks. Pikachu evolves from Pichu."
|
|
)
|
|
span = (
|
|
"Pikachu can store electricity in its cheeks and release it in "
|
|
"lightning-based attacks. (Source: https://en.wikipedia.org/wiki/Pikachu)"
|
|
)
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "trailing_artifact"
|
|
assert out["repair"]["action"] == "trim_trailing_artifact"
|
|
assert "(Source:" not in out["repair"]["kept_prefix"]
|
|
|
|
|
|
def test_repair_no_overlap_remove():
|
|
"""Repair plan for full-invention spans: remove the claim."""
|
|
base = "Pikachu is a Pokémon species."
|
|
span = "The Roman Senate convened in 49 BC to debate Caesar's rebellion"
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "no_overlap"
|
|
assert out["repair"]["action"] == "remove_claim"
|
|
|
|
|
|
def test_classify_paraphrase_high_token_coverage():
|
|
"""Tokens all present, sequence different — model rewrote the source."""
|
|
base = (
|
|
"Pikachu is a Pokémon species in the Pokémon franchise developed by "
|
|
"Game Freak and Nintendo. Pokémon games feature numerous creatures."
|
|
)
|
|
span = "Pikachu is a species of Pokémon creatures from the Pokémon franchise"
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "paraphrase"
|
|
assert out["token_coverage"] >= 0.85
|
|
|
|
|
|
def test_classify_partial_paraphrase_some_tokens_missing():
|
|
"""Some content overlap but a meaningful token isn't there at all —
|
|
likely model invention mixed with corpus tokens."""
|
|
base = "Pikachu is a Pokémon. Pikachu battles other creatures."
|
|
span = "Pikachu invented the lightning attack in Tokyo on Tuesday"
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] in ("partial_paraphrase", "no_overlap")
|
|
if out["diagnosis"] == "partial_paraphrase":
|
|
assert "missing_tokens" in out
|
|
|
|
|
|
def test_classify_no_overlap_full_invention():
|
|
"""Almost no content tokens shared with the corpus — pure model output."""
|
|
base = "Pikachu is a Pokémon species."
|
|
span = "The Roman Senate convened in 49 BC to debate Caesar's rebellion"
|
|
out = _classify_span(span, _norm(base), _norm(base))
|
|
assert out["diagnosis"] == "no_overlap"
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# inspect_cache_key — end-to-end with a tiny fixture
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
def _seed_record(qa_db: Path, *, cache_key: str, sources: list[dict],
|
|
unverified: list[str]) -> None:
|
|
"""Insert a minimal providence_cache row with a merkle_proof that
|
|
points at the given sources. Caller has already populated each
|
|
shard with the actual document + chunks."""
|
|
conn = connect(qa_db)
|
|
try:
|
|
with transaction(conn):
|
|
event_hash = append_audit(
|
|
conn,
|
|
event_type="providence_query",
|
|
subject_root=cache_key,
|
|
body={"cache_key": cache_key},
|
|
)
|
|
conn.execute(
|
|
"INSERT INTO providence_cache "
|
|
"(cache_key, source_root, document_uri, question_hash, "
|
|
" question_text, answer_text, merkle_proof, "
|
|
" model_profile_hash, conversation_hash, "
|
|
" governance_policy_hash, schema_version, "
|
|
" canonicalization_version, chunking_version, "
|
|
" falsification_state, chain, audit_event_hash, "
|
|
" created_at, hit_count, audit_mode, n_quotes, "
|
|
" n_verified, unverified_quotes, verifier_method) "
|
|
"VALUES (?, ?, 'corpus://multi', ?, ?, ?, ?, ?, ?, ?, "
|
|
" ?, ?, ?, 'live', 'private', ?, ?, 0, 'HYBRID', ?, ?, ?, 'span')",
|
|
(
|
|
cache_key,
|
|
"00" * 32,
|
|
"qh",
|
|
"what is foo?",
|
|
"answer here",
|
|
json.dumps({"sources": sources}, ensure_ascii=False),
|
|
"mh",
|
|
"ch",
|
|
"gh",
|
|
"v9.8.0",
|
|
"norm-v1",
|
|
"tok-512-v1",
|
|
event_hash,
|
|
int(time.time()),
|
|
len(unverified) + 1, # n_quotes (one verified plus N unverified)
|
|
1, # n_verified
|
|
json.dumps(unverified, ensure_ascii=False),
|
|
),
|
|
)
|
|
finally:
|
|
conn.close()
|
|
|
|
|
|
def _seed_doc(shard: Path, *, document_root: str, document_uri: str,
|
|
chunk_text: str) -> None:
|
|
"""Insert a documents row + one hot chunk with the given text."""
|
|
conn = connect(shard)
|
|
try:
|
|
from aborist.compress import pack_chunk
|
|
|
|
with transaction(conn):
|
|
conn.execute(
|
|
"INSERT INTO documents "
|
|
"(document_root, document_uri, source_type, kind, "
|
|
" compression_depth, chunking_version, "
|
|
" canonicalization_version, schema_version, ingest_ts, hit_count) "
|
|
"VALUES (?, ?, 'html', 'surface', 0, 'tok-512-v1', "
|
|
" 'norm-v1', 'v9.8.0', ?, 0)",
|
|
(document_root, document_uri, int(time.time())),
|
|
)
|
|
conn.execute(
|
|
"INSERT INTO chunks (document_root, idx, leaf_hash, content, tier) "
|
|
"VALUES (?, 0, ?, ?, 'hot')",
|
|
(document_root, "ff" * 32, pack_chunk(chunk_text)),
|
|
)
|
|
finally:
|
|
conn.close()
|
|
|
|
|
|
def test_inspect_cache_key_classifies_each_unverified(tmp_path):
|
|
qa_db = tmp_path / "qa.db"
|
|
shard = tmp_path / "001.db"
|
|
DOC = "ab" * 32
|
|
_seed_doc(
|
|
shard,
|
|
document_root=DOC,
|
|
document_uri="https://example.com/x",
|
|
chunk_text=(
|
|
"Pikachu can store electricity in its cheeks. "
|
|
"Pikachu evolves from Pichu."
|
|
),
|
|
)
|
|
_seed_record(
|
|
qa_db,
|
|
cache_key="01" * 32,
|
|
sources=[{
|
|
"document_root": DOC,
|
|
"document_uri": "https://example.com/x",
|
|
"title": "X",
|
|
"shard": shard.name,
|
|
"chunk_idx": 0,
|
|
}],
|
|
unverified=[
|
|
"Pikachu can store electricity in its cheeks.", # verbatim
|
|
"Pikachu fought in the 1988 Olympics in Seoul", # invention
|
|
],
|
|
)
|
|
|
|
result = inspect_cache_key(
|
|
"01" * 32, qa_db=qa_db, shards_dir=tmp_path
|
|
)
|
|
assert result["status"] == "ok"
|
|
assert result["record"]["cache_key"] == "01" * 32
|
|
assert len(result["sources"]) == 1
|
|
assert result["sources"][0]["chunk_count"] == 1
|
|
assert len(result["unverified"]) == 2
|
|
diags = [d["diagnosis"] for d in result["unverified"]]
|
|
assert "verbatim_in_base" in diags
|
|
# The Olympics span should be flagged as no_overlap or partial_paraphrase
|
|
# depending on token coincidences — both honest classifications.
|
|
assert any(d in ("no_overlap", "partial_paraphrase") for d in diags)
|
|
|
|
|
|
def test_inspect_cache_key_unknown_returns_not_found(tmp_path):
|
|
qa_db = tmp_path / "qa.db"
|
|
# touch DB so connect() initialises schema
|
|
connect(qa_db).close()
|
|
result = inspect_cache_key(
|
|
"00" * 32, qa_db=qa_db, shards_dir=tmp_path
|
|
)
|
|
assert result["status"] == "not_found"
|
|
|
|
|
|
def test_inspect_does_not_mutate_state(tmp_path):
|
|
"""Sidecar invariant: inspect must not write to providence_cache,
|
|
audit_events, or anything else. Pin so a future refactor can't
|
|
silently add side effects."""
|
|
qa_db = tmp_path / "qa.db"
|
|
shard = tmp_path / "001.db"
|
|
DOC = "cd" * 32
|
|
_seed_doc(
|
|
shard, document_root=DOC, document_uri="https://example.com/y",
|
|
chunk_text="hello world",
|
|
)
|
|
_seed_record(
|
|
qa_db, cache_key="02" * 32,
|
|
sources=[{
|
|
"document_root": DOC, "document_uri": "https://example.com/y",
|
|
"title": "Y", "shard": shard.name, "chunk_idx": 0,
|
|
}],
|
|
unverified=["something the model said"],
|
|
)
|
|
|
|
conn = connect(qa_db)
|
|
try:
|
|
before_audit = conn.execute("SELECT COUNT(*) FROM audit_events").fetchone()[0]
|
|
before_prov = conn.execute("SELECT COUNT(*) FROM providence_cache").fetchone()[0]
|
|
finally:
|
|
conn.close()
|
|
|
|
inspect_cache_key("02" * 32, qa_db=qa_db, shards_dir=tmp_path)
|
|
|
|
conn = connect(qa_db)
|
|
try:
|
|
after_audit = conn.execute("SELECT COUNT(*) FROM audit_events").fetchone()[0]
|
|
after_prov = conn.execute("SELECT COUNT(*) FROM providence_cache").fetchone()[0]
|
|
finally:
|
|
conn.close()
|
|
assert after_audit == before_audit
|
|
assert after_prov == before_prov
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# deflection diagnostic
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
def test_deflection_mars_bdfl_to_python_guido():
|
|
"""Empirical 2026-04-30: 'who is a benevolent dictator for life
|
|
for mars?' returned STRICT with answer 'Guido van Rossum is a BDFL
|
|
for the Python programming language.' Verifier did its job (claims
|
|
grounded) but answer never mentions mars — pure topic-shift.
|
|
|
|
Subject-anchor heuristic catches this: 3/4 of question tokens
|
|
(benevolent, dictator, life) match the answer, but the SUBJECT
|
|
anchor (last content token, 'mars') is missing. That's the signal
|
|
that a generic-vocabulary overlap can't paper over."""
|
|
d = diagnose_deflection(
|
|
"who is a benevolent dictator for life for mars?",
|
|
"Guido van Rossum is a Benevolent Dictator For Life (BDFL) "
|
|
"for the Python programming language.",
|
|
)
|
|
assert d["kind"] == "deflection", d
|
|
assert "mars" in d["question_tokens"]
|
|
assert d["subject_anchor"] == "mars"
|
|
assert d["subject_in_answer"] is False
|
|
assert "mars" not in d["overlap"]
|
|
# Generic-vocabulary overlap is high (3/4) but subject is missing
|
|
# — that's the deflection signal subject-anchor catches.
|
|
assert d["overlap_ratio"] > 0.5
|
|
|
|
|
|
def test_deflection_partial_overlap_when_some_subjects_match():
|
|
"""Some content tokens overlap, others don't — soft signal."""
|
|
d = diagnose_deflection(
|
|
"what is the relationship between linux and unix?",
|
|
"Linux is a Unix-like operating system inspired by Unix design.",
|
|
)
|
|
assert d["kind"] == "partial_overlap" or d["kind"] == "on_topic", d
|
|
assert "linux" in d["overlap"]
|
|
assert "unix" in d["overlap"]
|
|
# 'relationship' is a question-shape word that may or may not appear
|
|
# in answer; key is that core subjects overlap.
|
|
assert d["overlap_ratio"] > 0.0
|
|
|
|
|
|
def test_deflection_on_topic_when_all_subjects_present():
|
|
"""All question content tokens appear in answer — clean on-topic."""
|
|
d = diagnose_deflection(
|
|
"who painted the mona lisa?",
|
|
"Leonardo da Vinci painted the Mona Lisa.",
|
|
)
|
|
assert d["kind"] == "on_topic"
|
|
assert d["overlap_ratio"] == 1.0
|
|
|
|
|
|
def test_deflection_handles_empty_question_or_answer():
|
|
"""Vacuous case: question with no content tokens — no signal to give."""
|
|
d = diagnose_deflection("?", "Some answer text here.")
|
|
assert d["kind"] == "no_question_tokens"
|
|
assert d["overlap_ratio"] == 0.0
|
|
|
|
|
|
def test_deflection_strips_possessive_s_for_overlap():
|
|
"""Possessive 'mars's' should match 'mars' in the answer."""
|
|
d = diagnose_deflection(
|
|
"what is mars's atmosphere?",
|
|
"Mars has a thin atmosphere of carbon dioxide.",
|
|
)
|
|
assert "mars" in d["overlap"]
|
|
assert "atmosphere" in d["overlap"]
|
|
|
|
|
|
def test_deflection_suppressed_for_when_year_questions():
|
|
"""Bench finding 2026-05-01: 'what year did the berlin wall fall?'
|
|
answer '1989' has zero subject overlap (subject anchor 'fall' is
|
|
a verb the answer doesn't echo). Pre-suppression, classified as
|
|
deflection across all three modes (false positive)."""
|
|
d = diagnose_deflection(
|
|
"what year did the berlin wall fall?",
|
|
"1989.",
|
|
)
|
|
assert d["shape_suppressed"] is True
|
|
# Even with shape_suppressed, zero overlap on a date answer falls
|
|
# back to overlap-ratio = 0 → deflection. The shape_suppressed
|
|
# field tells the operator the signal was a known false-positive
|
|
# shape, not an actual topic shift.
|
|
assert d["kind"] == "deflection" # overlap-ratio path still fires
|
|
# But the operator can filter on shape_suppressed when scoring.
|
|
|
|
|
|
def test_deflection_suppressed_when_numeric_answer_overlaps():
|
|
"""When the date answer DOES include topic words ('Berlin Wall
|
|
fell in 1989'), shape suppression doesn't matter — overlap is
|
|
high enough for partial_overlap or on_topic."""
|
|
d = diagnose_deflection(
|
|
"what year did the berlin wall fall?",
|
|
"The Berlin Wall fell in 1989, ending the Cold War.",
|
|
)
|
|
assert d["shape_suppressed"] is True
|
|
assert d["kind"] in ("on_topic", "partial_overlap")
|
|
|
|
|
|
def test_deflection_suppressed_for_why_questions():
|
|
"""WHY-shape questions: answer discusses causes, may or may not
|
|
echo the question's verb. Bench: pointer mode deflected 3/3 on
|
|
'why did the titanic sink?' but answer was a valid cause."""
|
|
d = diagnose_deflection(
|
|
"why did the titanic sink?",
|
|
"The Titanic struck an iceberg and sank in 1912.",
|
|
)
|
|
assert d["shape_suppressed"] is True
|
|
# Has overlap on titanic+sink, classifies as on_topic or partial.
|
|
assert d["kind"] in ("on_topic", "partial_overlap")
|
|
|
|
|
|
def test_deflection_not_suppressed_for_who_questions():
|
|
"""Mars-BDFL pattern still fires on who-shape questions."""
|
|
d = diagnose_deflection(
|
|
"who is a benevolent dictator for life for mars?",
|
|
"Guido van Rossum is a Benevolent Dictator For Life for Python.",
|
|
)
|
|
assert d["shape_suppressed"] is False
|
|
assert d["kind"] == "deflection"
|
|
assert d["subject_anchor"] == "mars"
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# title-relevance sidecar
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
def test_title_mismatch_on_qcd_cited_for_spin_glass_claim():
|
|
"""Empirical 2026-05-01: 'explain spin glass modeling, tensors?'
|
|
returned STRICT 1/1 with claim 'Spin glass modeling involves...
|
|
mathematical tools such as tensors' cited to a chunk from
|
|
Quantum_chromodynamics. Claim tokens have zero overlap with QCD
|
|
title — sidecar should flag TITLE_MISMATCH so an operator can
|
|
see the retrieval-driven hallucination signature."""
|
|
d = diagnose_title_relevance(
|
|
"Spin glass modeling involves using the concept of spin glasses, "
|
|
"which are disordered magnetic systems, to study complex systems "
|
|
"and phenomena in physics and other fields. Spin glasses are "
|
|
"characterized by random interactions. Tensors are used to "
|
|
"represent the interactions between spins.",
|
|
["Quantum_chromodynamics"],
|
|
)
|
|
assert d["kind"] == "title_mismatch"
|
|
assert d["overlap"] == []
|
|
# Claim has 'spin', 'glass', 'modeling', etc. — title has 'quantum',
|
|
# 'chromodynamics'. Zero stem overlap.
|
|
assert "spin" in d["claim_tokens"]
|
|
assert "quantum" in d["title_tokens"]
|
|
|
|
|
|
def test_title_match_on_lois_lane_cited_for_supermans_girlfriend():
|
|
"""Sanity: legit citation (Lois Lane article cited for Lois-Lane
|
|
claim) passes title-relevance even though Superman isn't in the
|
|
title."""
|
|
d = diagnose_title_relevance(
|
|
"Lois Lane is the longtime love interest of Superman.",
|
|
["Lois_Lane"],
|
|
)
|
|
assert d["kind"] == "title_match"
|
|
assert "lois" in d["overlap"]
|
|
assert "lane" in d["overlap"]
|
|
|
|
|
|
def test_title_match_underscores_normalized():
|
|
"""Wikipedia titles use underscores; normalization to spaces lets
|
|
multi-word titles tokenize correctly."""
|
|
d = diagnose_title_relevance(
|
|
"The Berlin Wall fell in 1989.",
|
|
["Berlin_Wall"],
|
|
)
|
|
assert d["kind"] == "title_match"
|
|
assert "berlin" in d["overlap"]
|
|
assert "wall" in d["overlap"]
|
|
|
|
|
|
def test_title_mismatch_no_titles():
|
|
"""Defensive: empty title list returns no_titles."""
|
|
d = diagnose_title_relevance("Some claim.", [])
|
|
assert d["kind"] == "no_titles"
|
|
|
|
|
|
def test_title_mismatch_no_claim_tokens():
|
|
"""Defensive: empty/stopword-only claim returns no_claim_tokens."""
|
|
d = diagnose_title_relevance("the a an", ["Anything"])
|
|
assert d["kind"] == "no_claim_tokens"
|