Answer
Grounded Answer
Read the grounded answer first, then review refusal state, citations, source scope, and operator details as needed.
Refusal Or Not-Found State
Citations And Provenance
Citation cards show package evidence separately from current/public enrichment and session-only pasted source text.
Source telemetry
Answer Source Scope
Source evidence lane
Package authority and current/public enrichment
Current/public source content is enrichment only. It is not package authority.
Run diagnostics Token usage and compact run summary are secondary.
Advanced run details, source selection, request package, and corpus inventory
Operator source mechanics
Advanced: source selection preview and last-run source details
Preview estimates likely sources before you run. Sources used in the last answer are the actual completed-run source basis.
The assistant scores package sections against your question, balances source families, and keeps up to 8 sources. Preview is estimated before a run; last-run sources are what the completed answer actually used.
Source scoring is query-driven. Source-family balancing may diversify the set. Up to 8 is a cap, not a target every answer must hit. Draft Source Preview is not final until Ask completes.Preview: likely sources if you run thisDraft Source Preview
Source preview will appear after corpus metadata loads.
Sources used in the last answer
Last-Run Selected Sources update only after a completed answer or refusal.
No completed run yet.
Run Telemetry
Durable telemetry, database logging, model-routing history, and tokenomics analytics belong to #189 unless a later issue expands scope.
Token counts are shown when returned. Estimated run cost is deferred to #189 because cost calculation must be driven by provider and model pricing configuration; this page does not implement a token-cost calculator.
Detailed run telemetry
Operator detail stays readable in wrapped cards. Long values wrap here, while JSON and request package content stays contained in its own scroll box.
Corpus Inventory
Required Refusal Patterns
Not found in the approved playbook package.
The playbook provides guidance, but it does not approve this tool, data class, workflow, production use, or company policy.
The submitted proposal lacks enough evidence to assess this criterion.
Draft Proof Request Package
What is this for?
This is the evidence package prepared for the server-side provider call. Use it to audit selected source evidence, reproduce or debug a run, inspect grounding boundaries, or understand why the assistant answered or refused.
It is not the final answer and it is not the provider response. It may exist before a provider call because the browser can prepare candidate source evidence first. It carries checked package snippets, citation metadata, refusal rules, and the derivative-output label.
{}