PKPEEK Docs
Docs/Work, evidence & intelligence

Sigma Intelligence

Ask governed questions about the office portfolio or a case; Sigma answers from authorized snapshot data with citations, charts and drafts.

All personas (sigma.query)

/dashboard/dcc/sigma
Live walkthrough · Sigma Intelligence
/dashboard/dcc/sigma

Sigma Intelligence is a conversation workspace scoped to the Office portfolio or to one case (find a case, PIID or vendor in the left rail). It uses tools bound to this application: portfolio_overview, search_awards, award_detail, search_cases, case_detail, work_queue, search_evidence, evidence_detail, governed_answer, draft_work_item, draft_brief, save_view, build_chart and navigate.

  • Model toggle — *Enable model-assisted answers* or *Disable model access (deterministic fallback)*. With the model off, the snapshot engine answers (*Model off · snapshot engine*), labeled “Answered from the October 8 2026 demonstration snapshot”.
  • Starter questions — e.g. “Give me a portfolio overview for this office”, “Chart public net obligations by fiscal year…”, “Which deep cases are physically complete but blocked on administrative closeout…?”.
  • Governed case questions (synthetic) from the QA scenarios.
  • AI disclosure and Capability bindings panels explain what Sigma may do. Intents are adapter names, not claims about Sigma runtime function names.
  • Drafts (work items, briefs) are proposals only; a new conversation resets scope to the office portfolio.
Untrusted text — Retrieved document text is treated as untrusted. The SIG-005 test option *Include qa/adversarial_source.txt as untrusted retrieved text* demonstrates that injected instructions are ignored. Sigma never mixes public and synthetic totals and never invents facts public records don’t contain (e.g. “Does the real award … have an invoice defect?” is answered as not established).