Marketplace Trust and Operations

Dispute Resolution Workbench

A case workbench for marketplace disputes that turns chats, photos, milestones, and agent notes into a structured evidence timeline and appeal-friendly workflow.

Evidence-first disputes Ops workflow design

Problem

Marketplace disputes are messy because the inputs are messy: partial chats, conflicting photos, incomplete delivery history, and emotional narratives from both sides. Without structure, agents make inconsistent calls and appeals become difficult to manage well.

Notes

What it does

Dispute Resolution Workbench is a marketplace operations surface for handling claims that need more than a status field and a notes box. It turns fragmented evidence into a timeline that agents can actually work from.

The product centers on three things: collecting the right proof early, helping agents make consistent decisions, and preserving enough context for appeals to be meaningful instead of starting from scratch.

Operator workflow

A dispute is opened against an order, listing, or handoff event.
The system collects available milestones, conversations, and proof artifacts.
The agent works from a single timeline and a structured resolution worksheet.
If the decision is appealed, new evidence is attached without rewriting the earlier record.
Supervisors can compare first-pass and appeal decisions to find consistency gaps.

Product philosophy

Preserve context

A lot of dispute pain comes from losing the sequence of events. Once chats, images, and milestones are stitched together, the shape of the case becomes easier to understand.

Keep decisions explainable

Agents should be able to point to the checklist, the evidence, and the relevant policy framing behind a decision. That is better for internal QA and better for end users when outcomes are challenged.

Treat appeals as first-class

Appeals are not just a status transition. They are a second review with new evidence, and the system should make that difference visible.

Suggested metrics

Time to resolution by dispute type
Reopen or appeal rate
Evidence completeness before first decision
Agent consistency across similar cases

Impact

Improves consistency by capturing proof early, preserving decision context, and making appeals easier to audit and compare.

Stack

FastAPI
Postgres
Python
Object Storage
Notification workflows
Celery
HTMX

Technical design

Case API for disputes, parties, item references, and outcome states
Append-only evidence timeline for chats, photos, milestones, and notes
Decision templates and checklists for policy-guided resolution
Appeals workflow with new evidence ingestion and decision diffing
Operations analytics for backlog, reopen rate, and evidence completeness

Engineering decisions

Treat evidence as immutable artifacts referenced by metadata and hashes
Use structured decision templates with room for free-text nuance
Model dispute type explicitly instead of forcing all flows into one status enum
Keep notifications and nudges queue-driven so workflow state remains clean

Tradeoffs

More structure reduces ambiguity, but the UI has to stay lightweight for operators
Photo-based evidence helps outcomes, but it raises retention and privacy concerns
Appeals make the system fairer, but they need rate-limiting and abuse controls

Outcome / impact

Decisions become more consistent because evidence is organized and comparable
Appeals are easier to re-review because the system preserves what changed and why
Operations teams gain clearer visibility into which dispute types are slow or under-documented