supplier-scar

name	supplier-scar
description	Deterministic supplier PPM/OTIF/vendor-scorecard measurement, threshold-triggered escalation, and CQI-20/Ford Global 8D SCAR generation routing every metric, band, tier, and root-cause validation to calculate_supplier_ppm, calculate_otif, calculate_vendor_scorecard, evaluate_escalation, generate_scar, and render_sqe_canvas on quality-mcp.

Supplier Corrective Action Request (SCAR) & Vendor Rating: Measured Performance, Evidenced Escalation, Supplier-Owned Root Cause

Overview

The supplier-scar skill guides AI agents in measuring supplier performance, rating a supplier against organization-determined criteria, recommending an escalation tier, and issuing a structured Supplier Corrective Action Request. Five deterministic quantities are produced, and every one of them is produced by a tool on quality-mcp, never in prompt text: a supplier defect rate (PPM, and separately-named DPMO) via calculate_supplier_ppm; delivery performance (on-time, in-full, and the strict-conjunction OTIF) via calculate_otif; a weighted A/B/C vendor scorecard via calculate_vendor_scorecard; a threshold-triggered escalation tier via evaluate_escalation; and a CQI-20/Ford Global 8D SCAR via generate_scar. The already-evaluated results are presented as a supplier × dimension HTML matrix by render_sqe_canvas.

ISO 9001:2015 §8.4 ("Control of externally provided processes, products and services"), ISO 9001:2015 §10.2 ("Nonconformity and corrective action"), and IATF 16949:2016 §8.4 require that external providers be evaluated, selected, monitored, and re-evaluated against criteria the organization itself determines, and that nonconformity drive corrective action — and they name no PPM threshold, no OTIF window, no in-full tolerance, no scorecard weight, no rating-band boundary, and no escalation trigger. Every such number in this platform is the platform's own declared engineering heuristic, caller-configurable, labelled is_heuristic: true in the tool payload, and recorded in quality_core/sqe/ASSUMPTIONS_LOG.md (RULE-SQE-001 through RULE-SQE-013). Presenting any of them to a supplier or an auditor as an ISO 9001 or IATF 16949 requirement is a fabrication.

AIAG CQI-20 Effective Problem Solving Guide (2nd Edition, 2018) and the Ford Global 8D Manual back the structure of the corrective-action ladder and the three rendered SCAR sections — Root-Cause Requirement, Corrective-Action Requirement, and Prevention / Read-Across — and nothing else. They do not supply the escalation ladder's numeric triggers, which are heuristics like every other threshold here.

The supplier owns the root cause. This skill and generate_scar never author, infer, or paraphrase one — they request it and validate the supplier's own response by dispatching it to quality_core.rca's reversible 5-Why validator. A SCAR with no supplier response carries root_cause: null and can never reach CLOSABLE.

Two further authority boundaries hold throughout:

1. Missing data is undecided, never zero. A supplier with no receipts in the period is INDETERMINATE, not a 0 PPM "A" performer; a positively-weighted INDETERMINATE dimension suppresses the entire composite and band; an INDETERMINATE scorecard is neither escalated nor cleared.
2. Commercial authority is not a quality-engineering verdict. evaluate_escalation recommends a quality tier and always returns a commercial_authority disclaimer with it. New-business hold, de-sourcing, resourcing, and charge-back are business decisions made by authorized people, and this skill never states or implies one.

This skill equips agents to:

• Measure a supplier period from receipt lots and delivery records without ever quoting an unmeasured figure.
• Compose a defensible, fully heuristic-labelled vendor rating and a single highest evidenced escalation tier.
• Issue and track a CQI-20/Ford Global 8D SCAR whose root cause is supplied and validated, never authored.
• Delegate all measurement, scoring, adjudication, and rendering to calculate_supplier_ppm, calculate_otif, calculate_vendor_scorecard, evaluate_escalation, generate_scar, and render_sqe_canvas on quality-mcp.

When to Use

Activate this skill in the following supplier quality scenarios:

• Periodic Supplier Scorecarding: Monthly/quarterly vendor rating reviews producing an A/B/C band and a supporting evidence trail.
• Incoming-Receipt PPM Review: Reviewing a supplier's defect rate over a receiving window from inspected receipt lots.
• Delivery Performance Review: Assessing on-time, in-full, and strict-conjunction OTIF performance against promised dates and ordered quantities.
• Threshold-Triggered Escalation Decisions: Determining the quality-engineering escalation tier evidenced by a rated scorecard (and, when the caller supplies one, a recurrence count).
• Opening and Tracking a SCAR: Issuing a structured corrective action request, receiving the supplier's root cause, validating it, and tracking the request toward closure.

Input Requirements

• Supplier Period (period): supplier_id, period_start, period_end (inclusive window), optional period_label.
• Receipt Lots (lots): supplier_id, lot_id, quantity_received, receipt_date, defect_count (null = not yet counted, an undecided sentinel — never a zero), optional opportunities_per_unit (required on every in-scope lot for DPMO to be reported).
• Delivery Records (deliveries): supplier_id, order_id, quantity_ordered, quantity_delivered, optional requested_date, promised_date, actual_delivery_date.
• Cost Dimension (optional): copq_items plus a revenue_base, only when the caller has configured a positive cost weight and a cost curve.
• SCAR Request (request): supplier_id, issue_description, optional scar_id, linked_ncr_id, date_issued, due_date, requested_by.
• SCAR Evidence (optional): linked_ncr_evidence, supplier_root_cause_evidence (the supplier's own 5-Why chain), cost_impact_evidence, and a verification_of_effectiveness statement.

Prerequisites

• Active quality-mcp server connection providing calculate_supplier_ppm, calculate_otif, calculate_vendor_scorecard, evaluate_escalation, generate_scar, and render_sqe_canvas.

Step-by-Step Methodology

Follow the 5-step supplier rating and corrective-action methodology:

┌─────────────────────────────────────────────────────────────────────────┐
│           5-STEP SUPPLIER SCAR & VENDOR RATING METHODOLOGY              │
├───────────────────────────────────┬─────────────────────────────────────┤
│ 1. Evidence Collection            │ Gather receipts, deliveries, period │
│ 2. Deterministic Measurement      │ Execute calculate_supplier_ppm /    │
│                                   │ calculate_otif on MCP               │
│ 3. Composite Rating               │ Execute calculate_vendor_scorecard  │
│ 4. Escalation & Corrective Action │ Execute evaluate_escalation /       │
│                                   │ generate_scar                       │
│ 5. Visual Canvas & Reporting      │ Render via render_sqe_canvas        │
└───────────────────────────────────┴─────────────────────────────────────┘

1. Step 1: Evidence Collection & Period Definition

• Establish the exact evaluation window (period_start, period_end) and the supplier_id the evidence belongs to; lots and deliveries are matched on both.
• Collect inspected receipt lots and delivery records. Leave a value undecided (null) when it is genuinely unknown — an uninspected lot's defect_count is null, never 0.
• Strict Invariant: Never substitute, impute, or estimate a missing receipt date, defect count, promised date, or delivered quantity. The engines resolve undecided evidence to INDETERMINATE by design; that verdict is the answer, not an obstacle to work around.

2. Step 2: Deterministic Measurement

• Invoke calculate_supplier_ppm and calculate_otif on quality-mcp with the period and its evidence.
• Strict Invariant: Never compute a PPM, DPMO, on-time percentage, in-full percentage, or OTIF conjunction inline. All measurement must execute through calculate_supplier_ppm / calculate_otif, and every figure reported to the user must be quoted from the tool's returned payload.
• Read verdict first. On INDETERMINATE, quote the tool's reason and stop — do not report a rate.
• Read sample_adequacy.meets_minimum and the warnings list. A below-minimum received quantity is a signal to gather more evidence; the minimum itself is a declared heuristic with no standards basis.
• Treat on_time_pct, in_full_pct, and otif_pct as three separate figures. otif_pct is the strict conjunction (a delivery counts only when it is both on-time and in-full) and is never the average of the other two.

3. Step 3: Composite Rating

• Invoke calculate_vendor_scorecard on quality-mcp with the same period and evidence, plus copq_items/revenue_base when a cost dimension is configured.
• Strict Invariant: Never assign a composite score or an A/B/C band from memory, and never re-derive a sub-score from a raw metric. All scoring must execute through calculate_vendor_scorecard.
• If verdict is INDETERMINATE, report composite_score: null and band: null with the tool's reason: a positively-weighted undecided dimension suppresses the whole rating and its weight is never redistributed across the remaining dimensions.
• When quoting weights, curves, or band boundaries, quote them from heuristic_configuration together with their is_heuristic / basis labels — they are this platform's engineering heuristic, not a standards requirement.

4. Step 4: Escalation & Corrective Action

• Invoke evaluate_escalation on quality-mcp with the scorecard result, and a caller-supplied recurrence_count only when the user has actually stated one.
• Strict Invariant: Never assign an escalation tier from memory and never state a commercial action as this skill's own recommendation. All escalation must execute through evaluate_escalation, and any root cause must come from the supplier via generate_scar's linkage — never authored here.
• Present evaluated_triggers (the complete set, fired or not) alongside selected_evidence (only the winning tier's triggers), and repeat the returned commercial_authority disclaimer verbatim whenever a tier is communicated.
• Invoke generate_scar to issue the request. Route the supplier's returned 5-Why chain into supplier_root_cause_evidence and let the tool adjudicate it; if a linkage slot resolves EVIDENCE_INVALID, surface that sub-engine's own findings text unchanged.

5. Step 5: Visual Canvas & Reporting

• Invoke render_sqe_canvas on quality-mcp with the already-evaluated scorecard and escalation results — never with raw lots, deliveries, or metrics.
• The canvas performs no arithmetic and no adjudication; an INDETERMINATE supplier renders explicitly unrated rather than as a blank or zeroed cell.
• Retain the tool payloads alongside the rendered report so every published figure remains traceable to the call that produced it.

Tool Invocations

calculate_supplier_ppm

• MCP Server: quality-mcp
• Purpose: Deterministic supplier defect rate (PPM) and, when every in-scope lot carries an opportunity count, DPMO for one supplier period from receipt lots. A zero denominator, an undated in-scope lot, or an undecided defect_count resolves INDETERMINATE with ppm: null — never 0.0.
• Parameters:
  • period (dict[str, Any], required): Supplier identity and inclusive window.
    • supplier_id (string, required): Supplier identifier; lots are matched on it exactly.
    • period_start (string, required): ISO-8601 window start (inclusive).
    • period_end (string, required): ISO-8601 window end (inclusive).
    • period_label (string | null, optional): Human-facing label such as "2026-07".
  • lots (list[dict[str, Any]], optional, default []): Receipt lots — supplier_id, lot_id, quantity_received, receipt_date, defect_count (null = undecided), opportunities_per_unit.
  • config (dict[str, Any] | null, optional): sample_adequacy_minimum (integer, default 1000) — a declared engineering heuristic labelled is_heuristic: true, with no standards basis.
• Return Type: dict[str, Any]
• Return Schema:
  • supplier_id, period_start, period_end, period_label: Echoed period identity.
  • verdict (string): "MEASURED" or "INDETERMINATE".
  • ppm (float | null): Defect rate in parts per million; on every verdict.

---

calculate_otif

• MCP Server: quality-mcp
• Purpose: On-time, in-full, and strict-conjunction OTIF delivery performance for one supplier period. otif_pct counts a delivery only when it is on-time and in-full; it is never an average of on_time_pct and in_full_pct. No matched delivery, or any absent date or undecided delivered quantity, resolves the whole period INDETERMINATE.
• Parameters:
  • period (dict[str, Any], required): Same shape as calculate_supplier_ppm's period.
  • deliveries (list[dict[str, Any]], optional, default []): Delivery records — supplier_id, order_id, quantity_ordered, quantity_delivered (null = undecided), requested_date, promised_date, actual_delivery_date.
  • config (dict[str, Any] | null, optional): OTIFConfig — early_tolerance_days (default 0), late_tolerance_days (default 2), early_counts_as_on_time (default false), in_full_tolerance_pct (default 0.0), over_delivery_counts_as_in_full (default true). All five are declared engineering heuristics with no standards citation.
• Return Type: dict[str, Any]
• Return Schema:
  • supplier_id, period_start, period_end, period_label: Echoed period identity.
  • verdict (string): "MEASURED" or "INDETERMINATE".
  • delivery_count (integer): Matched deliveries — the single shared denominator of all three percentages.
  • , , (): Component counts.

---

calculate_vendor_scorecard

• MCP Server: quality-mcp
• Purpose: Weighted composite of the PPM (quality) and OTIF (delivery) dimensions, plus an optional COPQ (cost) dimension, mapped to an A/B/C band. Each source engine is invoked once and stays authoritative for its own arithmetic. Any positively weighted INDETERMINATE dimension suppresses the entire composite and band, and its weight is never redistributed.
• Parameters:
  • period (dict[str, Any], required): Same shape as above.
  • lots (list[dict[str, Any]], optional): Receipt lots for the quality dimension.
  • deliveries (list[dict[str, Any]], optional): Delivery records for the delivery dimension.
  • copq_items (list[dict[str, Any]] | null, optional): Cost-of-poor-quality items, delegated to the COPQ engine — never re-implemented here.
  • revenue_base (float | null, optional): Revenue base for the cost dimension's percentage-of-revenue metric.
  • config (dict[str, Any] | null, optional): ScorecardConfig — quality_weight (default 0.60), delivery_weight (default 0.40), cost_weight (default 0.0, and a cost_curve is required when it is positive), quality_curve / delivery_curve / cost_curve (best_value, worst_value), a_band_minimum (default 90.0), b_band_minimum (default 75.0), plus nested ppm_config and otif_config. The three weights must sum to 1.0 or the tool raises. Every value is a declared engineering heuristic.
• Return Type: dict[str, Any]
• Return Schema:
  • supplier_id, period_start, period_end, period_label: Echoed period identity.

---

evaluate_escalation

• MCP Server: quality-mcp
• Purpose: Recommend the single highest evidenced quality-engineering escalation tier — NONE → MONITOR → SCAR_REQUIRED → CONTAINMENT_REQUIRED → EXECUTIVE_REVIEW, or INDETERMINATE when the scorecard itself is INDETERMINATE (the supplier is then neither escalated nor cleared). Highest tier wins on ties. Never a commercial disposition.
• Parameters:
  • scorecard (dict[str, Any], required): A calculate_vendor_scorecard result. Only a RATED or INDETERMINATE verdict is accepted.
  • config (dict[str, Any] | null, optional): EscalationConfig — monitor_score_maximum (default 89.0), scar_score_maximum (default 74.0), containment_score_maximum (default 59.0), executive_score_maximum (default 39.0), monitor_recurrence_minimum (default 1), scar_recurrence_minimum (default 2), containment_recurrence_minimum (default 3), executive_recurrence_minimum (default 4). All eight are declared engineering heuristics.
  • recurrence_count (integer | null, optional): Caller-supplied only; recurrence is never inferred, and its triggers are evaluated only when a count is actually supplied.
• Return Type: dict[str, Any]
• Return Schema:
  • supplier_id (string), tier (string), scorecard_verdict (string).
  • evaluated_triggers (list[dict]): The full set, fired or not — each with tier, , , , , , , .

---

generate_scar

• MCP Server: quality-mcp
• Purpose: Generate a CQI-20/Ford Global 8D structured Supplier Corrective Action Request and adjudicate its linked evidence. The tool requests and validates a supplier root cause and never authors, infers, or paraphrases one: root_cause is only ever a verbatim copy of the terminal cause in a supplier-returned chain that quality_core.rca accepts.
• Parameters:
  • request (dict[str, Any], required): supplier_id, issue_description, optional scar_id, linked_ncr_id, date_issued, due_date, requested_by. An empty request returns an INDETERMINATE SCAR naming the fields a usable request must carry; a non-empty request holding an invalid value returns a clean error.
  • config (dict[str, Any] | null, optional): SCARConfig — evaluate_vendor_scorecard_linkage (bool, default true).
  • linked_ncr_evidence (dict[str, Any] | null, optional): Nonconformance evidence, dispatched to quality_core.ncr.
  • supplier_root_cause_evidence (dict[str, Any] | null, optional): The supplier's own 5-Why chain, dispatched to quality_core.rca's reversible validator.
  • cost_impact_evidence (dict[str, Any] | null, optional): Cost evidence, dispatched to quality_core.copq.
  • verification_of_effectiveness (string | null, optional): The supplier's stated verification; a blank statement normalises to null and never satisfies closure.
• Return Type: dict[str, Any]
• Return Schema:
  • supplier_id, scar_id, issue_description: Echoed request identity.
  • status (string): "DRAFT", , , , , or . requires an accepted supplier root cause a stated no other linkage resolving .

---

render_sqe_canvas

• MCP Server: quality-mcp
• Purpose: Render a single-writer HTML canvas presenting the supplier × dimension matrix — PPM, OTIF on-time/in-full, cost, composite score, band, and escalation tier. It performs no arithmetic and no adjudication: it presents already-computed results only, and an INDETERMINATE supplier renders explicitly unrated rather than as a blank cell.
• Parameters:
  • rows (list[dict[str, Any]] | null, optional): One entry per supplier — supplier_id, scorecard (a calculate_vendor_scorecard result), escalation (an evaluate_escalation result), optional supplier_name. Never raw lots, deliveries, or metrics.
  • theme (string, default "dark"): Colour theme palette ("dark" or "light").
  • standalone (boolean, default true): If true, returns a standalone HTML5 document; if false, an embeddable container.
  • title (string, default "SQE Vendor Scorecard Canvas"): Canvas header title.
• Return Type: dict[str, Any]
• Return Schema:
  • title (string): Canvas title.
  • verdict (string): "RENDERED", or "INDETERMINATE" when rows was supplied as an empty list — zero supplier results is never presented as a supplier population with nothing wrong in it.
  • reason (string | null): The explanation of an INDETERMINATE verdict; null when rendered.
  • rows_count (integer): Number of supplier rows rendered.
  • html (string): Rendered HTML string.

---

Example 1: Rated Supplier — PPM → OTIF → Scorecard → Escalation → SCAR

Every figure and verdict below is quoted from the tool response immediately above it. Nothing in this example is derived in prose.

Step 1 Invocation — calculate_supplier_ppm

{
  "name": "calculate_supplier_ppm",
  "arguments": {
    "period": {
      "supplier_id": "SUP-4410",
      "period_start": "2026-07-01",
      "period_end": "2026-07-31",
      "period_label": "2026-07"
    },
    "lots": [
      {
        "supplier_id": "SUP-4410",
        "lot_id": "LOT-7701",
        "quantity_received": 18000,
        "receipt_date": "2026-07-06",
        "defect_count": 42,
        "opportunities_per_unit": 4
      },
      {
        "supplier_id": "SUP-4410",
        "lot_id": "LOT-7702",
        "quantity_received": 22000,
        "receipt_date": "2026-07-17",
        "defect_count": 58,
        "opportunities_per_unit": 4
      }
    ]
  }
}

Successful Response

{
  "supplier_id": "SUP-4410",
  "period_start": "2026-07-01",
  "period_end": "2026-07-31",
  "period_label": "2026-07",
  "verdict": "MEASURED",
  "ppm": 2500.0,
  "numerator": 100,
  "denominator": 40000,
  "lot_count": 2,
  "dpmo": 625.0,
  "dpmo_opportunity_count": 160000,
  "sample_adequacy": {
    "minimum": 1000,
    "meets_minimum": true,
    "is_heuristic": true,
    "basis": "declared engineering default, no standards citation — see ASSUMPTIONS_LOG.md"
  },
  "reason": null,
  "warnings": [],
  "recommendations": [],
  "standards_basis": "No published AIAG/ISO/IATF standard defines a PPM formula, DPMO opportunity model, or sample-adequacy threshold; the arithmetic here is generic industry practice and the sample-adequacy minimum is a declared engineering heuristic (see ASSUMPTIONS_LOG.md)."
}

Report the returned verdict of "MEASURED", the returned ppm figure, and the separately named dpmo figure exactly as the tool emitted them. sample_adequacy.meets_minimum came back true, so no adequacy warning applies — and that minimum is a heuristic, not a standards threshold.

Step 2 Invocation — calculate_otif

{
  "name": "calculate_otif",
  "arguments": {
    "period": {
      "supplier_id": "SUP-4410",
      "period_start": "2026-07-01",
      "period_end": "2026-07-31",
      "period_label": "2026-07"
    },
    "deliveries": [
      {
        "supplier_id": "SUP-4410",
        "order_id": "PO-55120",
        "quantity_ordered": 4000,
        "quantity_delivered": 4000,
        "promised_date": "2026-07-08",
        "actual_delivery_date": "2026-07-08"
      },
      {
        "supplier_id": "SUP-4410",
        "order_id": "PO-55121",
        "quantity_ordered": 4000,
        "quantity_delivered": 3800,
        "promised_date": "2026-07-15",
        "actual_delivery_date": "2026-07-15"
      }
    ]
  }
}

Successful Response (abridged — delivery_breakdown shows the two orders passed above; the full period contained 16 matched deliveries)

{
  "supplier_id": "SUP-4410",
  "period_start": "2026-07-01",
  "period_end": "2026-07-31",
  "period_label": "2026-07",
  "verdict": "MEASURED",
  "delivery_count": 16,
  "on_time_count": 15,
  "in_full_count": 15,
  "otif_count": 14,
  "on_time_pct": 93.75,
  "in_full_pct": 93.75,
  "otif_pct": 87.5,
  "delivery_breakdown": [
    {
      "order_id": "PO-55120",
      "is_on_time": true,
      "is_in_full": true,
      "is_otif": true,
      "shortfall_qty": 0
    },
    {
      "order_id": "PO-55121",
      "is_on_time": true,
      "is_in_full": false,
      "is_otif": false,
      "shortfall_qty": 200
    }
  ],
  "heuristic_configuration": {
    "early_tolerance_days": 0,
    "late_tolerance_days": 2,
    "early_counts_as_on_time": false,
    "in_full_tolerance_pct": 0.0,
    "over_delivery_counts_as_in_full": true,
    "is_heuristic": true,
    "basis": "declared engineering default, no standards citation — see ASSUMPTIONS_LOG.md"
  },
  "reason": null,
  "warnings": [],
  "recommendations": [],
  "standards_basis": "No published AIAG/ISO/IATF standard defines an on-time window, an in-full tolerance, or whether early delivery counts as on-time; every OTIFConfig value here is a declared engineering heuristic, caller-configurable (see ASSUMPTIONS_LOG.md, RULE-SQE-001/002)."
}

Quote all three returned percentages separately. The returned otif_pct is lower than both the returned on_time_pct and the returned in_full_pct because the tool applied the strict conjunction — do not restate it as an average, and do not reconcile the three figures in prose.

Step 3 Invocation — calculate_vendor_scorecard

{
  "name": "calculate_vendor_scorecard",
  "arguments": {
    "period": {
      "supplier_id": "SUP-4410",
      "period_start": "2026-07-01",
      "period_end": "2026-07-31",
      "period_label": "2026-07"
    },
    "lots": "<the same receipt lots passed to calculate_supplier_ppm>",
    "deliveries": "<the same delivery records passed to calculate_otif>"
  }
}

Successful Response (each dimension's source_evidence is omitted here for brevity — it repeats the calculate_supplier_ppm and calculate_otif payloads above verbatim)

{
  "supplier_id": "SUP-4410",
  "period_start": "2026-07-01",
  "period_end": "2026-07-31",
  "period_label": "2026-07",
  "verdict": "RATED",
  "composite_score": 80.0,
  "band": "B",
  "dimensions": [
    {
      "name": "quality",
      "source_metric_name": "ppm",
      "raw_metric": 2500.0,
      "sub_score": 75.0,
      "weight": 0.6,
      "weighted_contribution": 45.0,
      "source_verdict": "MEASURED",
      "source_reason": null,
      "warnings": [],
      "recommendations": [],
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    },
    {
      "name": "delivery",
      "source_metric_name": "otif_pct",
      "raw_metric": 87.5,
      "sub_score": 87.5,
      "weight": 0.4,
      "weighted_contribution": 35.0,
      "source_verdict": "MEASURED",
      "source_reason": null,
      "warnings": [],
      "recommendations": [],
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    }
  ],
  "heuristic_configuration": {
    "weights": {
      "quality": {"value": 0.6, "is_heuristic": true, "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"},
      "delivery": {"value": 0.4, "is_heuristic": true, "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"},
      "cost": {"value": 0.0, "is_heuristic": true, "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"},
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    },
    "rating_bands": {
      "a_band_minimum": {"value": 90.0, "is_heuristic": true, "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"},
      "b_band_minimum": {"value": 75.0, "is_heuristic": true, "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"},
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    },
    "is_heuristic": true,
    "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
  },
  "omitted_dimensions": [
    {"name": "cost", "reason": "cost_weight is 0.0; not scored"}
  ],
  "reason": null,
  "warnings": [],
  "recommendations": [],
  "standards_basis": "ISO 9001:2015 section 8.4 and IATF 16949:2016 section 8.4 require supplier evaluation against criteria determined by the organization; those clauses do not define any scorecard weight, scoring curve, or A/B/C band."
}

Report the returned verdict of "RATED", the returned composite_score, and the returned band of "B" exactly as emitted, and state that the band boundaries quoted from heuristic_configuration.rating_bands are this platform's engineering heuristic — not an ISO 9001 or IATF 16949 requirement. The cost dimension appears under omitted_dimensions because it carried zero weight; it was not silently folded into the other two.

Step 4 Invocation — evaluate_escalation

{
  "name": "evaluate_escalation",
  "arguments": {
    "scorecard": "<the calculate_vendor_scorecard result above, passed through unmodified>"
  }
}

Successful Response (heuristic_configuration omitted here for brevity — it lists all eight thresholds, each labelled is_heuristic)

{
  "supplier_id": "SUP-4410",
  "tier": "MONITOR",
  "scorecard_verdict": "RATED",
  "evaluated_triggers": [
    {
      "tier": "MONITOR",
      "metric": "composite_score",
      "comparison": "<=",
      "observed_value": 80.0,
      "threshold": 89.0,
      "fired": true,
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    },
    {
      "tier": "SCAR_REQUIRED",
      "metric": "composite_score",
      "comparison": "<=",
      "observed_value": 80.0,
      "threshold": 74.0,
      "fired": false,
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    },
    {
      "tier": "CONTAINMENT_REQUIRED",
      "metric": "composite_score",
      "comparison": "<=",
      "observed_value": 80.0,
      "threshold": 59.0,
      "fired": false,
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    },
    {
      "tier": "EXECUTIVE_REVIEW",
      "metric": "composite_score",
      "comparison": "<=",
      "observed_value": 80.0,
      "threshold": 39.0,
      "fired": false,
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    }
  ],
  "selected_evidence": [
    {
      "tier": "MONITOR",
      "metric": "composite_score",
      "comparison": "<=",
      "observed_value": 80.0,
      "threshold": 89.0,
      "fired": true,
      "is_heuristic": true,
      "basis": "caller-configurable engineering heuristic with no standards citation — see ASSUMPTIONS_LOG.md"
    }
  ],
  "recurrence_count": null,
  "reason": null,
  "standards_basis": "AIAG CQI-20 corrective-action escalation discipline; organizational tier structure only, not numeric thresholds.",
  "commercial_authority": "Any commercial response remains a business decision made by authorized people; this result recommends only a quality-engineering tier."
}

Report the returned tier of "MONITOR", show the full evaluated_triggers list so the reviewer can see which triggers did not fire, and repeat the returned commercial_authority string verbatim. Do not translate the tier into a commercial action of any kind.

Step 5 Invocation — generate_scar

{
  "name": "generate_scar",
  "arguments": {
    "request": {
      "supplier_id": "SUP-4410",
      "scar_id": "SCAR-2026-018",
      "issue_description": "Bearing journal diameter above drawing limit on receipt lot LOT-7702, 58 units rejected at incoming inspection.",
      "linked_ncr_id": "NCR-2026-311",
      "date_issued": "2026-08-03",
      "due_date": "2026-08-31",
      "requested_by": "Supplier Quality Engineering"
    }
  }
}

Successful Response (section content and linkage.raw_result bodies abridged; linkage.linked_ncr and linkage.cost_impact resolved EVIDENCE_NOT_SUPPLIED because no evidence was passed for them)

{
  "supplier_id": "SUP-4410",
  "scar_id": "SCAR-2026-018",
  "issue_description": "Bearing journal diameter above drawing limit on receipt lot LOT-7702, 58 units rejected at incoming inspection.",
  "status": "AWAITING_SUPPLIER_RESPONSE",
  "sections": [
    {
      "heading": "Root-Cause Requirement",
      "rule_id": "RULE-SQE-011",
      "content": "State the systemic root cause of this nonconformity. ... The root cause is stated by the supplier and validated here; this generator never authors, infers, or substitutes one."
    },
    {
      "heading": "Corrective-Action Requirement",
      "rule_id": "RULE-SQE-012",
      "content": "Define and implement the permanent corrective action(s) that resolve the established systemic root cause. ..."
    },
    {
      "heading": "Prevention / Read-Across",
      "rule_id": "RULE-SQE-013",
      "content": "Identify every other part, product, line, and process to which the same systemic root cause applies, ..."
    }
  ],
  "linkage": {
    "linked_ncr": {
      "linkage_key": "linked_ncr",
      "verdict": "EVIDENCE_NOT_SUPPLIED",
      "engine": "quality_core.ncr",
      "findings": [],
      "rationale": "no nonconformance evidence was supplied for this slot",
      "raw_result": null
    },
    "supplier_root_cause": {
      "linkage_key": "supplier_root_cause",
      "verdict": "EVIDENCE_NOT_SUPPLIED",
      "engine": "quality_core.rca",
      "findings": [],
      "rationale": "no supplier root-cause response has been received for this slot",
      "raw_result": null
    },
    "cost_impact": {
      "linkage_key": "cost_impact",
      "verdict": "EVIDENCE_NOT_SUPPLIED",
      "engine": "quality_core.copq",
      "findings": [],
      "rationale": "no cost evidence was supplied for this slot",
      "raw_result": null
    },
    "vendor_scorecard": {
      "linkage_key": "vendor_scorecard",
      "verdict": "LINKAGE_NOT_AVAILABLE",
      "engine": null,
      "findings": [],
      "rationale": "vendor scorecard linkage is deferred this release and is never verdict-affecting",
      "raw_result": null
    }
  },
  "root_cause": null,
  "verification_of_effectiveness": null,
  "due_date": "2026-08-31",
  "date_issued": "2026-08-03",
  "reason": "this SCAR has been issued and no supplier root-cause response has been received.",
  "warnings": [],
  "recommendations": [],
  "standards_basis": "AIAG CQI-20 Effective Problem Solving (2nd Edition, 2018) and the Ford Global 8D Manual back the three rendered section headings only ..."
}

Report the returned status of "AWAITING_SUPPLIER_RESPONSE" and the returned root_cause of null. Do not fill that field in. The SCAR requests a root cause from the supplier; when the supplier returns a 5-Why chain, pass it as supplier_root_cause_evidence and let generate_scar adjudicate it — a chain the validator rejects yields RESPONSE_REJECTED with the validator's own findings, which you surface unchanged.

---

Example 2: Negative Control — Zero Receipts in Period

The supplier has no matched receipt lots in the requested window (the lots on hand belong to a different supplier or a different period).

Invocation

{
  "name": "calculate_supplier_ppm",
  "arguments": {
    "period": {
      "supplier_id": "SUP-9004",
      "period_start": "2026-07-01",
      "period_end": "2026-07-31",
      "period_label": "2026-07"
    },
    "lots": []
  }
}

Indeterminate Response

{
  "supplier_id": "SUP-9004",
  "period_start": "2026-07-01",
  "period_end": "2026-07-31",
  "period_label": "2026-07",
  "verdict": "INDETERMINATE",
  "ppm": null,
  "numerator": null,
  "denominator": 0,
  "lot_count": 0,
  "dpmo": null,
  "dpmo_opportunity_count": null,
  "sample_adequacy": {
    "minimum": 1000,
    "meets_minimum": false,
    "is_heuristic": true,
    "basis": "declared engineering default, no standards citation — see ASSUMPTIONS_LOG.md"
  },
  "reason": "no in-scope received quantity: 0 receipt lot(s) matched supplier_id 'SUP-9004' in window [2026-07-01, 2026-07-31], totalling 0 unit(s) received; PPM is undefined over a zero denominator and is not reported as 0.0",
  "warnings": [],
  "recommendations": [
    "Supply the receipt lots for this supplier and window before quoting a PPM figure."
  ],
  "standards_basis": "No published AIAG/ISO/IATF standard defines a PPM formula, DPMO opportunity model, or sample-adequacy threshold; the arithmetic here is generic industry practice and the sample-adequacy minimum is a declared engineering heuristic (see ASSUMPTIONS_LOG.md)."
}

Correct agent behaviour. Report the period as INDETERMINATE, quote the tool's own reason and its recommendations entry, and ask the user to supply a valid receipt period or the missing lot data before any figure is published. Stop there.

The behaviour this negative control rules out. Reporting 0 PPM and inferring an A rating. A supplier that shipped nothing — or whose receipts were simply not provided — has an undefined defect rate, not a perfect one. Do not proceed to calculate_vendor_scorecard, evaluate_escalation, or a SCAR on this evidence: the correct next move is a question to the user, not another tool call.

Best Practices

1. Strict Invariant: Zero Inline Math / Zero Inline Adjudication. Never compute a PPM, DPMO, OTIF, or composite score inline — always delegate to calculate_supplier_ppm, calculate_otif, or calculate_vendor_scorecard on quality-mcp. Never assign a rating band or escalation tier from memory — delegate to calculate_vendor_scorecard and evaluate_escalation. Never present a weight, band boundary, or threshold as an ISO 9001 or IATF 16949 requirement — every one is this platform's own declared engineering heuristic, not a standards requirement (see ASSUMPTIONS_LOG.md). Never state or imply a commercial action (new-business hold, de-sourcing, resourcing, charge-back) as a recommendation — that is a business decision, not a quality-engineering verdict. Never author, infer, or paraphrase a supplier's root cause — generate_scar requests one and validates the supplier's own response; the supplier owns the root cause.
2. Sample-Adequacy & Conjunction Discipline: Treat sample_adequacy.meets_minimum: false and a suppressed scorecard band as signals to gather more evidence, never as licence to force a rating. Report on_time_pct, in_full_pct, and otif_pct as three distinct figures, and never present the conjunction as an average of the other two.
3. Full-Trigger-Set Review: Always show evaluated_triggers, not just selected_evidence, so a reviewer can see which thresholds did not fire — that completeness is the escalation engine's own design intent, and it is what makes a tier auditable.
4. SCAR Closure Discipline: Never mark or imply a SCAR closed without both a quality_core.rca-accepted supplier root cause and a supplier-stated verification_of_effectiveness. CLOSABLE is the tool's verdict to issue, not the agent's.
5. Verbatim Findings: When any linkage slot resolves EVIDENCE_INVALID, surface the sub-engine's own findings text to the user rather than re-explaining, summarizing, or softening it — the owning engine's wording is the record.