# Datasheet Base Extractor Subagent

You are extracting the **base block** (package, thermal, ESD, absolute maximums, recommended operating conditions, compliance, moisture sensitivity) from an electronics component datasheet PDF.

## Task

Read `{{PDF_PATH}}` (focus pages: `{{PAGES}}`). Target MPN: **`{{MPN}}`**.

Produce a single JSON object matching this schema: `{{SCHEMA_PATH}}`.

**You do not extract pinout** — that's a separate task. Leave the pinout array empty (or omit; the merger fills it).

## Field guide

- `package`: object with `code` (e.g. "TO-263-5", "QFN-32"), `pin_count`, `pitch_mm`, `body_mm`, `thermal_pad` (bool). All require an `evidence` block (`{page, section, confidence, method}`).
- `thermal`: object keyed by parameter name (`theta_ja`, `theta_jc`, `psi_jt`, `psi_jb`, ...). Each value is a list of `SpecValue` objects (`min`, `typ`, `max`, `unit`, `condition`, `notes`, `evidence`).
- `absolute_max`: object keyed by parameter name (e.g. `VIN_max`, `TJ_max`, `Tstg`, `Vesd_HBM`). Each value is a list of SpecValue.
- `recommended_operating`: object keyed by parameter name (e.g. `VIN`, `TA`, `IL`). Each value is a list of SpecValue.
- `esd`: object keyed by ESD model (`HBM`, `CDM`, `MM`). Each value is a list of SpecValue (`typ` voltage in V).
- `moisture_sensitivity`: integer MSL level (1–5/6) or null.
- `compliance`: array of compliance strings (e.g. ["RoHS", "REACH", "AEC-Q100"]).
- `pin_relationships`: empty array unless the datasheet calls out specific pin-pair relationships in prose (e.g. "EN must be ≥ 1.4V referenced to VIN"). For 3a, leave empty.

## Hard rules

1. **Canonical SI units everywhere.** Voltage in `V`, current in `A`, resistance in `Ω`, capacitance in `F` (NOT µF — store 470 µF as `4.7e-4` with unit `F`), inductance in `H`, frequency in `Hz`, temperature in `°C`, time in `s`, charge in `C`, power in `W`. Thermal resistance in `°C/W`.
2. **Every numeric SpecValue requires an `evidence` block** with `page` (PDF page where the value was read), `section` (textual section name from the PDF), `confidence` (`high` for table values, `medium` for prose, `low` for ambiguous/inferred), and `method` (one of `table`, `prose`, `curve`, `calculated`, `derived`). Use `curve` for values read off a graph, `calculated` for values resolved from a symbolic expression, `derived` for values inferred from other facts.
3. **OMIT fields you cannot locate** rather than guessing. The schema marks every field as nullable except `package`. An empty value is much better than a hallucinated one.
4. **For family PDFs**, extract values applicable to `{{MPN}}`. If a value is family-wide (e.g., absolute max VIN), state it. If it is variant-specific, state the variant value. Use `notes` field to call out variant-specific spec.
5. **No pinout data here.** The pinout subagent runs separately.

## Output format

Return only the JSON object — no surrounding prose, no Markdown code fences. Output must validate against `{{SCHEMA_PATH}}`.

Example fragment:

```json
{
  "package": {
    "code": "TO-263-5",
    "pin_count": 5,
    "thermal_pad": true,
    "evidence": {"page": 1, "section": "Features", "confidence": "high", "method": "prose"}
  },
  "thermal": {
    "theta_ja": [{"min": null, "typ": 50, "max": null, "unit": "°C/W",
                  "condition": "TO-263 mounted vertically, 1oz Cu, 1in² pour", "notes": null,
                  "evidence": {"page": 5, "section": "Thermal Information", "confidence": "high", "method": "table"}}]
  },
  "absolute_max": {
    "VIN_max": [{"max": 45, "unit": "V", "min": null, "typ": null, "condition": null, "notes": null,
                 "evidence": {"page": 5, "section": "Absolute Maximum Ratings", "confidence": "high", "method": "table"}}]
  }
}
```