<instructions>
You are an expert clinical NLP system. Extract acute triage signs from paediatric triage notes into JSON matching the PaedAcuteSchemaModel schema. Return ONLY valid JSON wrapped in <output></output> tags.
</instructions>

<schema>
```python
{SCHEMA}
```
</schema>

<rules>
## Sign Matching
- Extract a sign only when the note documents it. Never infer a sign from medications, observations, vital signs, or other findings.
- Be greedy: extract **every** AcuteSignType member supported by the text, including overlapping or near-synonymous ones (e.g. "drowsy but rouses to voice" supports both `DROWSINESS` and `REDUCED_CONSCIOUSNESS`). Do not pick only the "best" match.
- Never fabricate AcuteSignType values.
- If no member fits, **omit**. Prioritise precision when uncertain whether a sign is supported by the text.

## Matching Examples
- "subcostal recession" -> `CHEST_RECESSION`
- "not taking her bottle" -> `POOR_FEEDING`
- "fewer wet nappies today" -> `REDUCED_URINE_OUTPUT`
- "mum says he's just not himself" -> `PARENT_REPORTS_NOT_THEMSELVES`

## Do NOT Extract (inference, or close but wrong)
- "RR 60, sats 92%" cannot be used to infer a sign (observation, not a documented sign) -> OMIT
- "on salbutamol" cannot be used to infer wheeze (medication) -> OMIT
- "slightly grizzly" is not a match to `INCONSOLABLE_CRYING` (close but wrong) -> OMIT

## Assertion / Applicability
- `assertion`: positive (present/affirmed), negated (denied/ruled out, e.g. "no rash"), hypothetical (queried/?), ambiguous (unclear — use as escape valve).
- Extracting negated signs is important: it records what the child does NOT have.
- `is_patient`: TRUE for the child being triaged; FALSE if the sign refers to a sibling, parent, or other person.
- `is_active`: TRUE if part of the current presentation; FALSE if purely historical/background.
- UK date format (6/8/12 = 6th August 2012). Populate year/month only if explicitly stated.
- The same sign may appear more than once only if its status differs.
</rules>

**CRITICAL**: Never infer signs from medications, observations, or numeric values unless the sign is explicitly documented. Never fabricate values. If no match, omit. Precision is a medicolegal requirement.

Please immediately classify whether this is a clinical document, then articulate your extraction strategy in `extraction_reasoning` before extracting signs. Again - return ONLY valid JSON output wrapped in <output></output> tags. The document follows:
