1. Signal discovery
Public trend signals are collected from configured sources and grouped into topics. Signals are ranked using factors such as freshness, cross-source presence, engagement, and available evidence.
2. Pre-generation screening
Before a topic or question can enter the model-backed research pipeline, local policy rules screen for medical, legal, and financial professional content. Blocked items do not create a search run and do not consume search-generation model tokens.
3. Research and evidence collection
Eligible questions are expanded into complementary research queries. Omniracle gathers source material, groups evidence by claim, records source domains, and keeps the source-to-claim relationship available to the final report.
4. AI-assisted synthesis
Models help structure and synthesize the collected evidence. Each model stage retains its request messages, structured inputs, raw response, parsed output, model identity, token usage, latency, prompt version, and errors so the complete request chain can be reconstructed for evaluation.
5. Publication and indexing gates
A completed report must meet minimum evidence-source, independent-domain, confidence, and result-length thresholds before it is eligible for indexing. Category and topic collections must also contain enough indexable reports before those aggregate pages may enter the sitemap.
Limitations
Sources can be incomplete, incorrect, revised, or unavailable. Automated classification may also make mistakes. Readers should open the cited sources and treat the report as a research starting point rather than a substitute for primary material or professional advice.