You are extracting structured facts from LinkedIn Experience entries.

Task:
1) Determine whether the person is currently a PhD student.
2) If NOT currently a PhD student, identify the current employer and since when.

Rules:
- Use only the provided roles.
- "Present" in the dates indicates the current role.
- If there are multiple "Present" roles, pick the one that best represents employment (not volunteering).
- Consider titles containing "PhD", "Ph.D", "Dottorando", "Doctoral", "PhD Student", "Doctoral Researcher" as PhD-student indicators.
- If current role is ambiguous, set is_current_phd to null and fill evidence.

Return ONLY valid JSON matching this schema:
{
  "is_current_phd": true|false|null,
  "current_employer": string|null,
  "current_title": string|null,
  "since_month": "YYYY-MM"|null,
  "evidence": string
}

Output rules (STRICT):
- Output ONLY the JSON object; no explanations, no markdown, no code fences.
- Do not add extra keys.
- "evidence" must be a non-empty string; if no evidence, use "Unknown".
- If a field is unknown, use null (except evidence).
- Use "YYYY-MM" for since_month or null.

Examples:
Bad:
Here is the extracted information:
{
  "is_current_phd": false,
  "current_employer": "Politecnico di Torino",
  "current_title": "Full-time",
  "since_month": "2023-11",
  "evidence": "Current role is Full-time at Politecnico di Torino, indicating employment"
}
Explanation: ...

Bad:
```json
{
  "is_current_phd": true,
  "current_employer": null,
  "current_title": "Researcher PHD Student",
  "since_month": "2022-11",
  "evidence": "Roles: Researcher PHD Student | DAUIN - Politecnico di Torino | Nov 2022 - Present"
}
```

Good:
{
  "is_current_phd": true,
  "current_employer": "Politecnico di Torino",
  "current_title": "PhD Student",
  "since_month": "2023-11",
  "evidence": "PhD Student | Politecnico di Torino | Nov 2023 - Present"
}

Roles:
{{ROLES}}
