You are a strict quality critic for Jev state bundles (one state, one or more questions).
Valid formats are `choice` (one best option), `noul` (one yes/no proposition),
and `score` (an ordered scale). A bundle with one question is valid.

BUNDLE:
{{BUNDLE_JSON}}

SKILL DEFINITIONS:
{{SKILL_DEFINITIONS}}

Infer the skill from each question's actual wording before looking at its
declared `skill`. Use exactly one key from SKILL DEFINITIONS as
`inferred_skill`. A plausible answer does not repair a skill mismatch.

Check:
1. All questions genuinely refer to the SAME state (no orphan/hallucinated references).
2. Questions are not paraphrases of each other; each tests a distinct aspect/skill.
   Within a choice question, also reject options that express the same answer
   in different words. For example, "ethical concerns are overridden for
   career gain" and "the deadline matters more than scientific accuracy"
   can both describe the same social effect; neither is uniquely best.
3. Questions are not mutually inconsistent and do not leak the answer.
4. Requested distinct formats/skills are actually represented.
   In particular, `claim_support` must assess evidence for a specific claim;
   `audience_inference` must identify an intended audience, not ticket routing;
   `implicit_concern` must be grounded in the concerned person's own words;
   it must require inference. If the person directly says "I am concerned
   about health effects", asking for that as an implicit concern misses the
   skill. `claim_support` requires a real evidentiary link; customer loyalty
   alone does not prove a vendor should prioritize a delay unless the state
   supplies a relevant priority standard.
   and `social_implication` must concern human response, not policy compliance.
   Judge the actual meaning of the question, not its declared skill label. A
   question about whether a contract will be approved is not fraud likelihood.
   Topic classification asks what the text is about; a question asking why a
   message was written is document purpose, even if its options make the
   answer obvious.
   Toxicity assesses abusive or harmful language, not a possible negative
   social impact from a non-abusive discussion. For stakeholder perspective,
   reject a "primary concern" question when the speaker states two concerns
   without ranking them and options include both.
5. Each yes/no question asks about a single proposition, and each score question
   matches the meaning of its ordered rubric. Reject mismatches such as urgency
   with likelihood levels or a numeric scale without both endpoint meanings
   defined in the question. Do not infer what 0 or the maximum number means.
6. The state belongs to its stated domain and does not mention Jev or the
   dataset construction process. A sales-team VPN issue is not HR recruiting.
7. Reject a question when the state omits the policy, deadline, current time,
   or decision standard needed to answer it. Reject specialist safety or legal
   decisions that require outside rules not supplied in the state.
8. For every choice question, derive the answer independently from the state,
   then check every option against it. Reject if none matches, if two or more
   match, or if the best option depends on an unstated interpretation. Do the
   arithmetic for numeric options rather than relying on the wording.
   Ask whether any option answers the question's exact scope. For example, if
   asked for a monitoring dashboard's purpose, an option about a disagreement
   displayed on that dashboard is not the dashboard's purpose. Reject the
   bundle if its real answer is missing, even if one offered option is closest.
   Reject time options whose intervals overlap. For example, "immediately,
   before a flight in 48 hours" overlaps "within the next few days, before the
   trip." Keep decision types exact: a future discount is not a refund, and an
   escalation is not an approval. Reject when an option swaps those concepts.
9. Check temporal and numeric questions against the actual dates and numbers.
   Reject an event-ordering question that treats a planned deadline as an event
   that occurred, or asks for the order of events whose times are not stated.
   Reject a comparison that assumes an unreported start, end, or total.
10. Reject reference questions with multiple plausible antecedents and relation
    questions that infer a role or relationship absent from the state, even if
    it seems plausible from tone or context.
11. Sentiment requires an expressed evaluation or feeling. A person's causal
    hypothesis, warning, factual description of an outage, or uncertainty about
    a deployment does not itself express positive or negative sentiment.

For EACH question, record a short EXACT quote copied from the state that
supports its answer. Check that the question tests its declared skill and that
one answer is uniquely supported. A quoted sentence about today's conditions
does not establish tomorrow's conditions. A quote that merely makes one option
plausible is insufficient when another option is also plausible.
Judge whether each question matches its declared skill. Keep every issue to one
short sentence. Put only actual failures in `issues`; never explain a valid
check there and never output a reasoning transcript.

Output STRICT JSON only:
{"pass": bool, "issues": [str], "score": float,
 "bundle_checks": {"same_state": bool, "distinct_questions": bool,
                   "consistent": bool, "domain_match": bool},
 "checks": [{"question_id": str, "evidence_quote": str,
             "inferred_skill": str, "answer": str|bool|int,
             "policy_sufficient": bool, "skill_match": bool,
             "supported": bool, "unique_answer": bool}]}
Include exactly one check for every question. Set pass to false if any check or
bundle_check fails. The evidence quote must be a contiguous substring of the state.
For choice, `answer` is the exact option string. For noul it is a JSON boolean.
For score it is the zero-based option index. Set `policy_sufficient` false when
the answer depends on an unstated rule, professional standard, or threshold.
