The generated answer text, grounded in retrieved sources.
The classification result that determined routing behaviour.
Citations for the sources used in generating the answer.
Optional researchSynthesized narrative from the deep research phase, when tier-3 routing exercised external or host-provided research.
Total wall-clock duration of the entire query pipeline in milliseconds.
Which tiers were actually exercised during this query.
[0] for trivial, [1, 2] for multi-source with fallback
Names of fallback strategies that were activated during this query. Empty array if no fallbacks were needed.
['keyword-fallback', 'tier-escalation']
Optional groundingCitation verification results. Populated when deep research runs
or when verifyCitations is explicitly requested in config or query options.
Optional recommendationsRecommended skills, tools, and extensions based on query analysis.
Populated when the plan-aware classifier (classifyWithPlan) produces
capability recommendations. When no recommendations are made (or the
plan-aware classifier is not used), this field is undefined.
Each recommendation includes a confidence score (0-1) and a human-readable reasoning string explaining why the capability was recommended.
Final result returned by the QueryRouter after classification, retrieval, and answer generation.
This surface is intentionally provenance-oriented: it includes not only the generated answer and citations, but also the tier path actually exercised and the fallback names that were activated during routing.