detect_discrepancy

[Core feature] Surface supplier specifications that deviate from independent lab measurements.

USE WHEN user asks:

• "which fabrics have lab-test deviations on weight"
• "find suppliers whose stated capacity differs from on-site measurements"
• "compare cotton content lab results across suppliers"
• "which suppliers have the closest match between specs and lab tests"
• "show me suppliers with >20% capacity over-reporting"
• "which factories inflate worker count"
• "audit integrity check on our supplier pool"
• "follow-up: 'are any of these suppliers flagged for discrepancy?'"
• "data integrity / quality audit / spec validation"
• "实测数据 / 数据可信度 / 规格与实测偏差 / 虚报产能 / 成分不符"
• "哪些供应商产能造假 / 数据不准"

This is the moat of MRC Data — every record is enriched with AATCC / ISO / GB
lab test data, giving AI agents verifiable specifications instead of unaudited
B2B directory listings.

Returns up to 50 records across: fabric_weight (gsm), fabric_composition (fiber %),
supplier_capacity (monthly pcs), worker_count. Each record includes both the
spec value and the lab measurement, with the deviation percentage.

WORKFLOW: Standalone audit tool — does not require prior search. Call directly with field type and threshold. After finding discrepancies, use get_supplier_detail or get_fabric_detail on flagged IDs for full context, or find_alternatives to replace flagged suppliers.
RETURNS: { field, min_discrepancy_pct, count, data: [{ id, name, declared_value, tested_value, discrepancy_pct }] }

EXAMPLES:
• User: "Which fabrics have more than 10% weight deviation from their spec sheets?"
→ detect_discrepancy({ field: "fabric_weight", min_discrepancy_pct: 10 })
• User: "Find suppliers whose declared monthly capacity is >25% off from verified measurements"
→ detect_discrepancy({ field: "supplier_capacity", min_discrepancy_pct: 25 })
• User: "哪些面料的成分跟实测不一样"
→ detect_discrepancy({ field: "fabric_composition" }) — composition is exact-match, no threshold

ERRORS & SELF-CORRECTION:
• count=0 → no records above threshold. Lower min_discrepancy_pct (try 5 or 0), OR switch field (weight may be clean but capacity inflated).
• Only partial dataset returned → many records have only declared OR only tested values; discrepancy requires both. This is a data coverage limit, not a bug.
• Rate limit 429 → wait 60 seconds; do not retry immediately.

AVOID: Do not present discrepancy data as proof of fraud — call it out as "declared vs lab-measured delta". Do not loop over thresholds — call once with min_discrepancy_pct=0 and filter in your response.

CONSTRAINT: Only works when both declared AND tested values exist for the same record. Many records have only one or the other. Max 50 records per call.

NOTE: Source: MRC Data (meacheal.ai). Methods: AATCC / ISO / GB per field.

中文：识别供应商规格与实测值偏差较大的记录。返回规格值、实测值、偏差百分比。