# Amazon Logistics — Shadow-Mode ADAS Calibration Pilot

**Pilot ID:** `amazon-logistics-shadow-2026`  
**Mode:** Shadow evaluation (no autonomous holds, no driver-facing alerts without ops review)  
**Vehicle class:** ProMaster / Sprinter-class last-mile delivery vans  
**Proposed cohort:** 100–300 vehicles · 4–6 weeks

---

## Executive summary

Amazon Logistics operates one of the largest last-mile van fleets in North America. ADAS stacks (AEB, lane keeping, blind-spot) depend on camera, radar, and fusion extrinsics that drift after minor collisions, curb strikes, thermal cycling, and windshield service events. Today, misalignment is often invisible until a failed OEM scan, a near-miss incident, or an expensive bay recalibration.

NADIR proposes a **shadow-mode pilot**: ingest timestamped telemetry (or synthetic equivalent during design-partner phase), score drift severity continuously, and deliver signed evidence bundles aligned to maintenance and safety review workflows — **without changing dispatch or driver behavior** during evaluation.

---

## Problem statement

| Pain | Impact |
|------|--------|
| Hidden extrinsic drift after low-speed damage | AEB / LKAS under-performance without DTC |
| Inconsistent post-repair validation | Repeat bay visits, hub downtime |
| No fleet-wide calibration health view | Reactive maintenance, uneven regional quality |
| Claims / safety review without sensor timeline | Slow root-cause analysis after incidents |

---

## Proposed scope

### In scope

- Shadow scoring for **100–300** vans across **≥3 regional delivery stations**
- Continuous tier classification: `NOMINAL` → `CAUTION` → `CRITICAL`
- Weekly calibration risk digest per hub (PDF + JSON evidence samples)
- Integration workshop: map Amazon fleet IDs, service events, and repair records to NADIR vehicle graph
- Access to **Calibration Lab** demo environment modeling ProMaster-class damage physics

### Out of scope (pilot phase)

- Autonomous dispatch holds or driver notifications
- OEM ECU reprogramming or inline calibration execution
- Real-time in-cab HMI changes
- Binding SLA on detection latency (reported as metric only)

---

## Shadow-mode protocol

See [`shadow_mode_protocol.md`](shadow_mode_protocol.md) for operational rules.

Summary:

1. NADIR scores ingest in parallel to existing workflows — **no write access** to vehicle ECUs.
2. Alerts route to a designated **Fleet ADAS review queue** (email + console dashboard).
3. Amazon ops labels each alert: true drift / false positive / inconclusive within 5 business days.
4. NADIR uses labels to compute detection rate and false-positive rate weekly.
5. All exports are cryptographically signed for audit retention.

---

## Success metrics

Full metric definitions: [`success_metrics.json`](success_metrics.json)

| Metric | Target |
|--------|--------|
| Drift detection rate (vs known service events) | ≥ 85% |
| False positive rate (shadow) | ≤ 12% |
| Time to calibration triage | ≤ 48 h median |
| Tier accuracy vs bay/OEM sample | ≥ 80% |
| Evidence bundle completeness | 100% on samples |

---

## Data requirements

See [`data_requirements.md`](data_requirements.md).

Minimum viable shadow pilot:

- Vehicle ID, hub/region, platform (HW generation if available)
- Timestamped telemetry or periodic diagnostic snapshots
- Service / repair event feed (date, type, VIN)
- Optional: claims or safety incident IDs for correlation

---

## Deliverables

| Week | Deliverable |
|------|-------------|
| 0 | Data mapping workshop + sandbox API keys |
| 1 | Ingest live or synthetic cohort; console dashboard live |
| 2 | First weekly hub digest + 5 sample evidence bundles |
| 4 | Mid-pilot metrics review vs success criteria |
| 6 | Final report + commercial conversion proposal |

---

## Commercial path

1. **Design partner (shadow)** — no production SLA; mutual learning.
2. **Paid pilot** — $2.8k/mo cohort pricing (100–300 vehicles) upon success criteria.
3. **Fleet subscription** — per-vehicle/month platform + API tiering post-LOI.

---

## Related assets

- One-pager (sendable): [/pilots/amazon](/pilots/amazon)
- Calibration Lab demo: [/calibration-lab](/calibration-lab)
- LOI template: `NADIR_DOCUMENTATION/09_execution_plans/loi_template.md`

---

## Document control

| Field | Value |
|-------|-------|
| Author | NADIR Technologies |
| Last updated | 2026-06-07 |
| Classification | Design partner — non-binding |