Use evidence, not reputation.
Friction comes from operational signals such as access failures, waits, handoff constraints, proof issues, and tenant workflow exceptions.
Trust
Operational intelligence with strict decision boundaries.
DropScore is designed around private, auditable evidence and human-reviewed outcomes, not public labels or subjective rating loops.
No public customer score
No LLM decisioning
No camera requirement
Keep final action human-reviewed.
Pay recommendations, repair prompts, safety review, and customer-impacting workflows stay behind approval until tenant policy and legal review support more automation.
Govern LLM assistance.
LLMs can summarize verified packets or draft operator-facing text, but every prompt, model, packet scope, and output path needs audit and fallback controls.
Preserve tenant boundaries.
Cross-tenant learning is disabled by default and should use only approved aggregate or redacted scopes when tenants explicitly opt in.
No public customer score
Customers see fixable prompts, not a hidden reputation label or a driver rating feed.
No camera requirement
The core system relies on operational signals already present in last-mile workflows.
No LLM decisioning
LLMs may summarize verified evidence, but they must not directly decide pay, discipline, safety outcomes, or restrictions.
Tenant isolation
Cross-tenant learning is off by default and can use only explicitly approved privacy-preserving scopes.
Audit-first actions
Prompt changes, evidence packets, approvals, publishing attempts, and tenant acknowledgements are recorded.
Fairness review
Production calibration must track outcomes by delivery type, address type, geography, route type, language, and accessibility factors.