Methodology · v2.10

SDPDelivery Promise

Match between {legal_name}'s stated delivery / fulfillment SLAs and observed real-world lag. SAMPLING IS A HARD CONTRACT (server-enforced): the payload MUST contain ≥3 ROUTINE on-time delivery events (lag ≤ 5 days) AND ≤2 delay events. The server now COUNTS routine events (response_at − date ≤ 5d) and BAILS with `unbalanced_sample` when fewer than 3 routine events are present — the previous prompt-only contract was ignored by the LLM. If you cannot source ≥3 routine on-time events from major map / local-business directory store reviews, Trustpilot resolved-quickly threads, retail reviews mentioning timely arrival, or company-published delivery testimonials, RETURN type="absent" with reason="balanced_sample_unavailable" rather than score on a delay-only cherry-pick. Cherry-picking complaints from BBB/ConsumerAffairs/Reddit is forbidden — it drags scores artificially low for mega-brands with high overall on-time rates. REQUIRED: cite the company's published delivery SLA URL (the official shipping/delivery policy page on the company's own domain, or its carrier's SLA page) as one of the source_urls. Return time_responsiveness with events: exactly ≥3 routine fulfillment + ≤2 delay events. Each event: date = order timestamp, response_at = actual delivery timestamp, severity = order magnitude, url = public source. Anchor the sample to the company's STATED shipping SLA — niche/edge-case long delays exist for every fulfillment operation but are NOT representative of the median. Server applies time_responsiveness_score (median lag → bucket).