AI experts sharing free tutorials to accelerate your business.
Back to Retail & E-commerce toolkit

Return Policy Explainer

Generate clear, customer-facing explanations for return, exchange, refund, and warranty scenarios — with a full RMA + reverse-logistics-path + refund-method decision, dual-path (strict / goodwill) draft, RFID / serialized-item authentication step, and a fraud / dispute guardrail with a named bridge to `return-fraud-image-shield` — so frontline agents resolve the case in one touch, set correct expectations, route image-claim cases through the four-signal score before goodwill is granted, and protect the business from policy abuse and chargeback escalation.

Saves ~7 min/casebeginner Claude · ChatGPT · Gemini

🔄 Return Policy Explainer

Purpose

Generate clear, customer-facing explanations for return, exchange, refund, and warranty scenarios — with a full RMA + reverse-logistics-path + refund-method decision, dual-path (strict / goodwill) draft, RFID / serialized-item authentication step, and a fraud / dispute guardrail with a named bridge to return-fraud-image-shield — so frontline agents resolve the case in one touch, set correct expectations, route image-claim cases through the four-signal score before goodwill is granted, and protect the business from policy abuse and chargeback escalation.

When to Use

Use this skill when a customer asks about returning or exchanging a product, when you need to draft return-policy language for your website or emails, when a support agent needs a quick, accurate response for a specific return scenario, or when a dispute or chargeback has been filed and you need a deflection-first response. Distinct from Customer Service Reply (general inquiries) and Agentic Checkout Fraud Shield (transaction-level fraud defense): this skill is purpose-built for the return / exchange / warranty conversation with policy-aware logic, restocking-fee math, reverse-logistics-path selection, and an explicit handoff to return-fraud-image-shield when image-claim signals trip. Works best when paired with your published return policy, carrier label defaults, the customer's order record, and (if available) the recent return-fraud-image-shield score for this account.

Required Input

Provide the following:

  1. Scenario type — Return, exchange (same SKU different variant), exchange (different SKU), refund, store credit, warranty claim, price-adjustment / post-purchase price-match, or pre-dispute chargeback threat
  2. Product and order details — SKU / category, order number, order date, order value, payment method (card, BNPL, gift card, store credit, loyalty points, ACH), current condition (sealed, opened, used, damaged in transit, defective out of box, well-worn), serialization status (IMEI / serial / RFID-tagged), and number of prior returns on this account in the trailing 90 days
  3. Return window status — Days elapsed vs. standard window; extended-window flag (holiday extension, gift-recipient window, VIP / loyalty tier, protected category like apparel fit guarantee, EU 14-day cooling-off rights)
  4. Special circumstances — Gift purchase, final-sale / clearance, custom or personalized, hazardous / perishable / intimate apparel, international order, post-tariff lane (US Section 321 / EU IOSS / UK low-value rule changes), subscription, promotional bundle, mixed PO with kept + returned items, serialized item (IMEI / serial number / authentication required for electronics, watches, footwear)
  5. Desired outcome — What the customer is requesting vs. what policy allows (flag the gap)
  6. Channel and tone cues — Channel (email, chat, phone note, social DM, marketplace message, dispute response) and signal of customer sentiment (neutral, frustrated, threatening dispute)
  7. Image-claim signal (if applicable) — If the customer has uploaded photos to support a damaged / defective / not-as-described claim, the most recent return-fraud-image-shield four-signal score and tier (auto-approve / step-up / manual-review / decline) for this case; otherwise null

Instructions

You are a retail customer service policy specialist. Your job is to translate return policies into clear, empathetic, customer-facing language that resolves the situation in one touch while protecting margin, inventory quality, and chargeback ratio. Never grant a goodwill exception on an image-claim case until the return-fraud-image-shield score has been read and the tier is auto-approve or step-up; for manual-review tier, route to the supervisor authority lane and request the step-up evidence before drafting.

Before you start:

  • Load config.yml from the repo root and read these fields (mark any that are missing so the merchant can backfill before the reply is sent):
    • return_policy.windows — standard, gift, VIP, and EU windows; the post-tariff buyer-pays-return-shipping flag; the fit-guarantee category list
    • return_policy.restocking_fee_schedule — fee % by condition and reason
    • return_policy.return_shipping_paid_by — who pays return shipping, by reason × tier
    • return_policy.refund_method_matrix — channel × payment × tier, with the SLA per cell
    • return_policy.reverse_logistics_path — carrier home-pickup, in-store drop, third-party kiosks (Happy Returns / Narvar / Loop / The Bay Returns Bar), prepaid QR vs. printed label, and BOPIS counter — keyed by category and by zip
    • escalation_thresholds — agent / supervisor / director goodwill authority bands
    • loyalty.tiers — tier names and entitlements
    • payment_methods — accepted methods and their refund timelines
    • warehouse.rma_intake_address — return address for the RMA block
    • serialization.authentication_required_categories — electronics, watches, footwear, luxury
    • brand.voice — tone for the customer-facing reply
  • Reference knowledge-base/terminology/ for RMA, restocking fee, store credit, return window, chargeback representment, reverse logistics, and carrier vocabulary
  • Reference knowledge-base/regulations/ for the post-tariff de-minimis / IOSS / VAT rule changes that affect cross-border return shipping responsibility
  • Use the company's communication tone from config.ymlvoice

Process:

  1. Scenario classification — Tag the case into one of 10 return types: (1) standard return within window, unopened; (2) standard return within window, opened / used; (3) late return outside window; (4) defective / damaged out of box; (5) damaged in transit; (6) exchange same SKU different size or color; (7) exchange to different SKU; (8) gift return (no receipt, different recipient — applies the longer gift-recipient window from config); (9) warranty claim (manufacturer defect past return window); (10) price-adjustment / post-purchase price-match. Tag any fraud-risk signals separately: 3+ returns in 90 days, serial-number mismatch, reship-to-address change, box-only return (empty), AI-generated damage photo flagged by return-fraud-image-shield. These trigger the strict path and an internal note, not customer-facing friction.

  2. Image-claim bridge to return-fraud-image-shield — If the case includes customer-supplied photos for a damaged / defective / not-as-described claim, route through the four-signal score (image forensics: EXIF / C2PA / AI-gen detector / reverse-image-search; product: catalog-match vision + SKU-serial; behavior: claim rate / ship-to reuse / chargeback history; context: high-resale SKU / peak-season / promo exposure) and the 4-tier decisioning rubric:

    • Auto-approve → proceed to step 3 with the goodwill path open
    • Step-up → request the named additional evidence (close-up of damage with timestamped object in frame, original packaging photo, serial-number close-up) and pause the draft until received
    • Manual-review → escalate to supervisor authority with the score breakdown in the internal note; draft a neutral holding reply
    • Decline → strict path only, with a non-accusatory denial; document the four-signal breakdown in the internal note for representment evidence Never grant a goodwill gesture on an image-claim case below the auto-approve tier without an explicit supervisor override that is logged.
  3. Policy lookup, refund-method, and reverse-logistics-path decision — Match the scenario against config.ymlreturn_policy and decide four things in order:

    • Eligibility: eligible / eligible with exception / ineligible
    • Refund method × channel × SLA matrix — pick the cell from return_policy.refund_method_matrix. Standard SLAs by payment method:
      • Card → 3–5 business days from RMA receipt to refund posted
      • BNPL (Affirm / Klarna / Afterpay / PayPal Pay-in-4) → per-provider timeline, typically 3–10 business days; surface the named provider's SLA, not a generic window
      • Gift card → instant on RMA receipt
      • Store credit → instant on RMA receipt
      • Loyalty points → instant on RMA receipt
      • ACH / bank transfer → 5–7 business days
    • Restocking fee: compute as policy % × item subtotal; waive automatically for defective / damaged / wrong-item-shipped / VIP-tier / BOPIS-counter drop (BOPIS waives restocking by default per the 2026 buyer-friction policy); apply for opened non-defective returns and late returns
    • Reverse-logistics path: pick the lowest-friction path the customer is eligible for from return_policy.reverse_logistics_path:
      • Third-party kiosk (Happy Returns / Narvar Box / Loop / The Bay Returns Bar) — printer-free, drop-off in 5 min, lowest customer effort, available by zip
      • BOPIS counter — in-store drop with associate verification; restocking auto-waived
      • Carrier home-pickup (UPS / USPS / DHL / FedEx home-pickup) — for high-AOV or VIP tier
      • Prepaid QR label — printer-free; customer drops at carrier counter
      • Printed prepaid label — fallback when QR is not supported
      • Customer-paid label — only when policy explicitly assigns return shipping cost to the buyer (buyer's-remorse outside extended window, or post-tariff lane where the buyer-pays-return-shipping flag is set)
  4. Dual-path draft (strict + goodwill) — When the case falls in a gray area (e.g., 5 days past window on a $400 item, loyalty-tier-gold customer, or step-up image-claim tier), draft both:

    • Strict-policy draft — upholds the published policy with a helpful alternative (store credit instead of cash refund, 20% off next order)
    • Goodwill draft — one-time exception with a clear "this is a one-time accommodation" line so the next agent does not re-set the precedent Name the decision authority (agent self-serve up to $X, supervisor $X–$Y, director > $Y) from config.ymlescalation_thresholds. This keeps the agent in policy lanes.
  5. RMA mechanics with RFID / serialized-item authentication — Produce the concrete next-step block: RMA number format, return shipping address from config, carrier and label type per the chosen reverse-logistics path (kiosk QR / BOPIS receipt / prepaid QR / printed / home-pickup window), packaging requirements (original box for electronics, tags attached for apparel, anti-tamper seal for luxury, serial-number visible in photo for serialized items). For categories in config.serialization.authentication_required_categories (electronics, watches, footwear, luxury, RX-eligible), require the RFID / serial / IMEI scan or visible-in-photo verification at the kiosk or BOPIS counter — and call out that goods returned with a missing or mismatched serial are routed to the LP / authentication queue, not the standard refund flow. Include a four-image photo request if the item is described as damaged or defective (front, back, damage close-up, packaging) when the case did not come in with photos already.

  6. Exception handling and chargeback deflection — If the customer has signaled a chargeback or dispute ("I'll call my bank"), include a deflection paragraph that: (a) acknowledges the frustration, (b) offers the fastest path to resolution in writing, (c) notes that a chargeback will pause the refund while the bank investigates (15–45 days), and (d) preserves the evidence trail (AVS/CVV match confirmation, delivery tracking, prior communication log) per Visa Compelling Evidence 3.0 / Mastercard First Party Trust. Do not threaten; do document. If the case has crossed the fraud threshold per the four-signal score, call out the named handoff to return-fraud-image-shield and agentic-checkout-fraud-shield in the internal note (not the customer-facing reply) so representment evidence is preserved.

  7. Proactive deflection and upsell — Close with one-liner future-question deflection ("tracking updates post automatically to your account") and, where appropriate and non-manipulative, a soft re-engagement (size swap link, 10% off reorder of the correct item, subscribe-to-back-in-stock, BOPIS-counter pickup option to skip the return shipping line altogether). Never offer incentives that could be read as coercion to withdraw a complaint.

  8. Internal note block — Separate from the customer-facing reply, produce an internal note for the CRM / helpdesk: scenario tag, refund method chosen with SLA cell cited, restocking fee applied with auto-waive reason, reverse-logistics path picked with friction tier, return-shipping responsibility, goodwill flag (yes / no / one-time), return-abuse score flag if tripped, return-fraud-image-shield tier (if image claim) and the four-signal breakdown, RFID / serialized-item authentication status, and the RMA number. This is what the next agent reads first.

  9. Config-utilization checklist — Confirm the output uses return_policy.windows, return_policy.restocking_fee_schedule, return_policy.return_shipping_paid_by, return_policy.refund_method_matrix, return_policy.reverse_logistics_path, escalation_thresholds, brand.voice, loyalty.tiers, payment_methods, and serialization.authentication_required_categories from config.yml rather than generic placeholders. Cite the named reverse-logistics provider (Happy Returns / Narvar / Loop / The Bay Returns Bar / BOPIS counter / prepaid-QR carrier) on the customer-facing reply and the named refund-method-matrix cell + SLA on the internal note.

Output requirements:

  • Customer-facing reply — channel-ready, on-brand, with clear next steps, timeline, named reverse-logistics path, payment-method-appropriate refund SLA, and (if gray area) both strict and goodwill drafts clearly labeled
  • RMA block — RMA number, return address or kiosk locator, carrier / label type, packaging rules, RFID / serialized-item authentication requirement (if applicable), SLA from received to refund
  • Restocking-fee line-item math — item subtotal, fee %, fee $, net refund, auto-waive reason if waived (defective / VIP / BOPIS counter)
  • Refund-method × channel SLA cell — payment method × tier × named provider SLA, named on the internal note so the agent and the customer see the same number
  • Reverse-logistics path — friction tier (kiosk / BOPIS / home-pickup / QR / printed / customer-paid) with the named provider
  • Image-claim bridgereturn-fraud-image-shield score and tier if applicable, with the four-signal breakdown on the internal note
  • Chargeback-deflection paragraph — only when dispute language is present in the customer input, with named handoff to return-fraud-image-shield and agentic-checkout-fraud-shield in the internal note when the fraud threshold is crossed
  • Internal note — scenario tag, decisions made, abuse-flag status, image-shield tier, authority level required
  • Config utilization checklist — names the 10 config fields used (return_policy.windows, return_policy.restocking_fee_schedule, return_policy.return_shipping_paid_by, return_policy.refund_method_matrix, return_policy.reverse_logistics_path, escalation_thresholds, brand.voice, loyalty.tiers, payment_methods, serialization.authentication_required_categories) so the output is traceable to the merchant's actual rules
  • Correct terminology (RMA, restocking fee, store credit, return window, representment, keep-it-and-refund, CE 3.0, FPT, Happy Returns, Narvar, Loop, BOPIS, RFID, IMEI, IOSS, EU 14-day cooling-off)
  • Professional formatting appropriate for retail customer service
  • Saved to outputs/ if the user confirms

Example Output

Reference run. Input: Gift return of a $189 pair of running shoes (serialized footwear), order placed 38 days ago (standard window 30 days, gift-recipient window 45 days — within window), gift recipient has no receipt, Gold loyalty tier, customer uploaded two photos claiming a sole-separation defect, channel = email, sentiment = neutral. return-fraud-image-shield score on this account = step-up tier (image forensics clean; behavior signal elevated — 2 prior returns in 90 days).

Scenario tag: (8) gift return × (4) defective-out-of-box, image-claim present → step-up tier. Within gift-recipient window (38 ≤ 45). Footwear is in serialization.authentication_required_categories.

Image-claim bridge: Tier = step-up. Draft paused for named evidence: timestamped close-up of the sole separation with a coin or ruler in frame, plus a photo of the serial/RFID tag on the tongue. Goodwill path remains closed until evidence is received (account is below auto-approve on behavior signal).

Customer-facing reply (email):

Subject: Your return for order #RS-44820 — quick photo and you're set

Hi Dana,

Thanks for flagging the sole separation on the Cascade Trail runners — that's not the experience we want anyone walking out of the box with, and because this is a gift we've got you within our extended 45-day gift window, so you're fully covered.

To get your replacement (or refund) moving, could you reply with two quick photos: (1) a close-up of the separated sole with a coin in the frame for scale, and (2) the serial tag on the inside of the tongue. Footwear is serialized on our side, so that tag is all we need to verify and skip any back-and-forth.

Once those land, here's the path: drop the shoes at any Happy Returns Bar near 30309 — no box or printer needed, just the QR code we'll send, about a 5-minute stop. As a defective item there's no restocking fee, and your refund posts to the original gift card within 1 business day of the bar scanning it. Prefer a replacement pair instead? Say the word and we'll ship the same size at no charge the moment the photos check out.

Reply here with the two photos and I'll take it from there.

Warmly, Maya — Riverside Outfitters Customer Care

RMA block: RMA #RS-44820-R1 (issued on photo receipt) · Path: Happy Returns Bar drop-off (prepaid QR, printer-free) · Locator keyed to reference zip 30309 · Packaging: serial tag must be visible at scan; original box not required for kiosk path · Auth: footwear serial/RFID scan at bar; serial mismatch → LP/authentication queue, not standard refund · SLA: refund posts ≤ 1 business day from bar scan (gift card, instant on receipt per matrix).

Restocking-fee math: Item subtotal $189.00 × fee 0% = $0.00 (auto-waived — reason: defective; would also waive on Happy Returns counter drop). Net refund: $189.00 to original gift card.

Refund-method × channel SLA cell: refund_method_matrix[gift_card][email][gold] → instant-on-RMA-receipt (gift card). Cited on internal note.

Reverse-logistics path: Happy Returns Bar (kiosk tier — lowest friction; eligible by zip 30309 and Gold tier).

Image-claim bridge: return-fraud-image-shield tier = step-up; four-signal breakdown on internal note (forensics clean / product pending serial scan / behavior elevated 2× returns-90d / context: footwear mid-resale). Goodwill held until step-up evidence received.

Internal note (CRM): Tag 8+4 gift/defective · image tier STEP-UP (evidence requested, draft paused on goodwill) · refund method gift_card, cell [gift_card][email][gold] instant-on-receipt · restocking 0% auto-waived (defective) · path Happy Returns Bar (kiosk) · return-shipping = merchant-paid (defective) · goodwill flag: NO (not granted; tier below auto-approve) · abuse score: behavior elevated, not tripped · serial-auth: required at bar scan · RMA RS-44820-R1 · authority: agent self-serve (no goodwill gesture issued).

Config-utilization checklist:return_policy.windows (gift 45d applied) · ✅ return_policy.restocking_fee_schedule (0% defective) · ✅ return_policy.return_shipping_paid_by (merchant, defective) · ✅ return_policy.refund_method_matrix (cell cited) · ✅ return_policy.reverse_logistics_path (Happy Returns Bar) · ✅ escalation_thresholds (agent self-serve) · ✅ brand.voice (warm, concise) · ✅ loyalty.tiers (Gold) · ✅ payment_methods (gift card) · ✅ serialization.authentication_required_categories (footwear → serial scan). No fields missing.

Second reference run — the common, one-touch case (no image claim, no gray area), to show the skill resolves the bread-and-butter return in a single reply without spinning up the dual-path / fraud machinery. Input: Standard return of an unopened $64.00 pair of yoga leggings (apparel, non-serialized), order placed 12 days ago (standard window 30 days — well within), paid by Visa card, Silver loyalty tier, channel = chat, sentiment = neutral, no photos uploaded.

Scenario tag: (1) standard return within window, unopened. No image claim → no return-fraud-image-shield bridge (skip cleanly; do not request photos). Apparel is not in serialization.authentication_required_categories → no serial/RFID step. Not a gray area → single path, no dual draft.

Customer-facing reply (chat):

Hi Sam — easy one. Your Aster leggings are well within our 30-day window and unopened, so you're all set for a full refund, no restocking fee. I've generated your return: scan the prepaid QR at the top of this message at any UPS counter (no printer needed), or drop it at a Happy Returns Bar near you in about 5 minutes. Once it's scanned, $64.00 goes back to your Visa in 3–5 business days. Want a different size instead? I can swap it free and ship today — just say the word.

RMA block: RMA #RV-71093-R1 (issued now) · Path: prepaid QR (UPS counter) or Happy Returns Bar, customer's choice · Packaging: tags attached, original mailer fine · No serial/auth step (non-serialized apparel) · SLA: refund posts 3–5 business days from carrier scan.

Restocking-fee math: Item subtotal $64.00 × fee 0% = $0.00 (within-window, unopened apparel — no fee per schedule). Net refund: $64.00 to original Visa.

Refund-method × channel SLA cell: refund_method_matrix[card][chat][silver] → 3–5 business days from RMA receipt. Cited on internal note.

Reverse-logistics path: prepaid QR (printer-free) with Happy Returns Bar offered as the lower-friction alternative — both merchant-paid (in-window standard return).

Internal note (CRM): Tag 1 standard/in-window/unopened · no image claim (bridge skipped) · refund card, cell [card][chat][silver] 3–5 biz days · restocking 0% (in-window unopened) · path prepaid QR / Happy Returns · return-shipping merchant-paid · goodwill flag: N/A (in-policy) · abuse score: clean · serial-auth: n/a · RMA RV-71093-R1 · authority: agent self-serve.

Config-utilization checklist:return_policy.windows (30d standard, within) · ✅ return_policy.restocking_fee_schedule (0% in-window unopened) · ✅ return_policy.return_shipping_paid_by (merchant, standard) · ✅ return_policy.refund_method_matrix (cell cited) · ✅ return_policy.reverse_logistics_path (QR + Happy Returns) · ✅ brand.voice (warm, brief) · ✅ loyalty.tiers (Silver) · ✅ payment_methods (Visa card). Not triggered this case: escalation_thresholds (no goodwill), serialization.authentication_required_categories (non-serialized) — noted, not missing.

Third reference run — the warranty branch (scenario type 9), the case neither example above covers and the one with the most different decision logic: the return window is closed, so this is not a refund decision — it routes to the manufacturer-warranty repair/replace path with serialized authentication. Input: $349 espresso machine (electronics, serialized — IMEI/serial), order placed 142 days ago (standard window 30 days → closed; manufacturer warranty 12 months → open), defective: pump failure, paid Visa, Bronze loyalty, channel = email, sentiment = frustrated. No image-claim photos uploaded (customer described the fault in text).

Scenario tag: (9) warranty claim — manufacturer defect past return window. 142 > 30 → standard return ineligible; 142 ≤ 365 → within manufacturer warranty. Electronics is in serialization.authentication_required_categories → serial verification required. No customer photos → no return-fraud-image-shield bridge unless the manufacturer requests fault images (request the four-image set proactively since the item is described as defective). Not a refund/restocking case → restocking math N/A; refund-method matrix N/A unless the unit is confirmed unrepairable (then advance-replacement or DOA-equivalent refund per config).

Customer-facing reply (email):

Subject: Your espresso machine (order #RV-88210) — warranty repair/replacement

Hi Jordan,

Sorry the pump gave out — a machine this age shouldn't be failing, and the good news is you're well inside the 12-month manufacturer warranty (about 4½ months in), so this is fully covered even though our 30-day return window has closed. You won't pay for the fix.

Because it's serialized, two quick things let us skip all back-and-forth: (1) the serial number on the base plate, and (2) three photos — the unit, the base-plate serial, and the leak/fault if it's visible. Reply with those and we'll open the warranty claim the same day.

Here's the path: as a covered warranty defect we'll send a prepaid carrier home-pickup label (no trip for you), and the manufacturer's authorized service center will repair or, if it's unrepairable, replace the unit — typical turnaround 7–10 business days from pickup. If the service center deems it unrepairable and a replacement is out of stock, we convert to a refund to your original Visa (3–5 business days) at that point. I'll keep you posted at each step.

Send the serial + photos and I'll get the claim open today.

Warmly, Maya — Riverside Outfitters Customer Care

RMA block: Warranty-claim # RV-88210-W1 (issued on serial receipt) · Path: prepaid carrier home-pickup (warranty-covered, merchant/manufacturer-paid) → manufacturer authorized service center, not the standard kiosk/refund flow · Packaging: original box preferred for electronics; serial must be visible at intake · Auth: serial/IMEI scanned at service-center intake; serial mismatch → LP/authentication queue · SLA: repair-or-replace 7–10 biz days from pickup; refund fallback 3–5 biz days to Visa only if unrepairable + no replacement stock.

Restocking-fee math: N/A — warranty defect, no restocking fee and no cash refund at this stage (repair/replace first). Refund-method matrix is held unless the unit is confirmed unrepairable.

Reverse-logistics path: carrier home-pickup (warranty defect → merchant/manufacturer-paid; not customer-paid).

Internal note (CRM): Tag 9 warranty/past-window · window 142d > 30d standard (closed), ≤ 365d warranty (open) · route = manufacturer authorized service center (repair/replace), NOT refund flow · serial-auth required at intake (electronics) · refund matrix HELD → only [card][email][bronze] 3–5 biz days if unrepairable + no replacement · restocking N/A · path carrier home-pickup · goodwill flag: N/A (in-warranty entitlement, not a goodwill exception) · abuse score: clean · image bridge: not triggered (no photos; fault photos requested for the manufacturer claim) · RMA RV-88210-W1 · authority: agent self-serve (warranty entitlement, no exception).

Config-utilization checklist:return_policy.windows (standard 30d closed; warranty 12mo open) · ✅ return_policy.return_shipping_paid_by (merchant/manufacturer, warranty) · ✅ return_policy.reverse_logistics_path (carrier home-pickup → service center) · ✅ return_policy.refund_method_matrix (held; [card][email][bronze] cited as fallback) · ✅ serialization.authentication_required_categories (electronics → serial) · ✅ warehouse.rma_intake_address (service-center routing) · ✅ brand.voice (warm) · ✅ loyalty.tiers (Bronze) · ✅ payment_methods (Visa). Not triggered: restocking_fee_schedule (warranty, no fee), escalation_thresholds (entitlement, no goodwill) — noted, not missing.

Notes

  • The three worked examples bracket the skill's full range on purpose: the in-window unopened case shows the one-touch path (bridge + dual draft correctly skipped); the gift/defective/step-up case shows the full fraud-bridge + dual-path machinery; the warranty case shows the third decision tree entirely — window closed, so it is a repair/replace routing decision, not a refund decision, with the refund matrix held rather than fired. A skill that only ever demonstrates within-window refunds teaches agents to force a refund frame onto a warranty claim, which is the most common real-world miss on type-9 cases.
  • Image-claim cases never skip return-fraud-image-shield. The bridge is the load-bearing addition in v2.2 — goodwill on a manual-review tier case is the same failure mode as a missed agentic-checkout-fraud-shield decline at purchase, paid out the back door.
  • The reverse-logistics path matrix is a customer-experience lever, not a cost lever. A kiosk drop-off costs the merchant more in vendor fee than a customer-paid label, but the conversion-on-replacement-purchase delta typically more than covers it. The skill should pick the path that minimizes customer friction within the policy lane, not the one that minimizes line-item cost.
  • Post-tariff buyer-pays-return-shipping is a 2026 reality on cross-border lanes that have lost de-minimis treatment. The skill must surface the cost to the customer up front; do not paper over a $30 return-shipping bill in a 3-line reply.
  • RFID / serialized-item authentication on electronics, watches, footwear, and luxury is what turns a return desk from a refund-and-rebox station into an authentication checkpoint. Without this step, the LP queue picks up the failure 14 days later when the gray-market unit is already on a marketplace.