[SYSTEM]
You are a biotech competitive intelligence analyst. Your task is to identify drug 
assets that satisfy a structured sourcing thesis query.

[INPUT]
Query JSON:
{QUERY_JSON}

Rules:
- Use the as_of_date in the Query JSON as a hard cutoff.  
- Return assets that satisfy ALL constraints (target, modality, stage 
window, indication, and exclusions).
- Do not guess. Only include an asset if you can provide dated evidence for each 
constraint dimension that is explicitly present in the Query JSON (e.g., 
primary_target, modality, stage_as_of_date, indication if specified, and any geography 
or sponsor/disclosure constraints if specified).
- Exclusions: exclude an asset only if you find evidence it violates an exclusion. If 
you cannot find any evidence of a violation, do not invent one.
- Prefer canonical asset identity (INN/USAN if known) and include common code names in 
the asset_name field when helpful for disambiguation.
- If the asset identity is ambiguous (cannot be reliably distinguished from another 
program), exclude it.

[OUTPUT]
Return a JSON array of objects with the schema below.

Schema:
{
  "asset_name": ".",
  "sponsor": ".",
  "modality": ".",
  "primary_target": ".",
  "indications": ["."],
  "stage_as_of_date": ".",
  "evidence": [
    {
      "source_type": "patent | company_website | registry | publication | poster | 
thesis | tto | grant | other",
      "source_id": ".",
      "date": "YYYY-MM-DD",
      "supporting_span": "short quoted span or precise section reference"
    }
  ]
}

CASBench scoring template (query-specific rubric)

For each returned candidate asset A and query Q:

1) Let D(Q) be the set of constraint dimensions explicitly present in Q's structured 
constraint specification. Examples of dimensions include target, modality, stage 
window, indication, sponsor/disclosure geography, or other structured constraints 
included in the query.

2) For each dimension d in D(Q), compute:
match_d = 1 if A satisfies Q's constraint for dimension d else 0

3) If Q specifies exclusions, compute:
exclusion_compliance = 1 if A does not violate any Q.exclusions else 0
If Q specifies no exclusions, exclusion_compliance is omitted from scoring.

4) Define the constraint score using only the active dimensions for that query:
constraint_score = average of {match_d for d in D(Q)} plus exclusion_compliance if 
applicable.
(Equivalently, constraint_score is normalized to [0,1] for every query by dividing by 
the number of scored sub-components.)

Ranking for R@P≥0.95:
- Rank candidates by constraint_score descending.
- Break ties deterministically (canonical asset_id ascending).
- Scan the ranked list and select the longest prefix with precision >= 0.95.
- Report recall on that prefix: TP_prefix / |gold_assets(Q)|.

{
  "query_id": "CAS-DEV-017",
  "thesis": "NaV1.8 (SCN10A) small-molecule inhibitor assets that are 
preclinical in China (PRC development control and/or PRC-first disclosure), as 
of 2025-11-01",
  "constraints": {
    "target": ["NaV1.8", "SCN10A"],
    "modality": "small_molecule",
    "stage_min": "discovery",
    "stage_max": "late_preclinical",
    "geography": "China"
  },
  "exclusions": [
    "biologics (peptides, antibodies, toxins)",
    "non-selective sodium channel blockers where Nav1.8 is not a primary 
mechanistic driver",
    "purely upstream/downstream pain-pathway programs without NaV1.8 target 
engagement evidence",
    "non-PRC development control as of as_of_date"
  ],
  "as_of_date": "2025-11-01",
  "difficulty": "Hard"
}

{
  "asset_id": "CAS-NAV18-004",
  "drug_name": "Hengrui Amidine-Modified VX-548 Analog",
  "evidence_artifacts": [
    {
      "artifact_id": "EV-NAV18-004-001",
      "source_type": "patent",
      "source_id": "WO2024041613A1 (family; PRC-origin series)",
      "language": "Chinese/English",
      "date": "2024-02-01",
      "claims_supported": ["target", "modality", "mechanism/selectivity", 
"geography"],
      "supporting_span": "series described as NaV1.8 inhibitors with pain-use 
claims; includes selectivity panel"
    },
    {
      "artifact_id": "EV-NAV18-004-002",
      "source_type": "other",
      "source_id": "AdisInsight drug profile (ID: 800057997)",
      "language": "English",
      "date": "2024-02-15",
      "claims_supported": ["sponsor", "stage", "indication"],
      "supporting_span": "lists sponsor (Hengrui), modality (small molecule), 
target (NaV1.8), and development stage (preclinical)"
    }
  ]
}