Workflows and examples

Begin with a decision, not a source quota.

The most useful request names who will use the result, what they must decide, what belongs in scope, what evidence matters, and what the deliverable must expose.

See a complete result

The worked local/global sea-level inquiry tests whether apparently conflicting measurements share a reference frame and scope. It includes rejected source history and a precise next evidence request. Open the room to follow claims, read notes and return to the answer.

A representative first request

Use $omnara-deep-research. Investigate whether small organizations should adopt passkeys for customer accounts in 2026. Audience: a product and security lead deciding what to ship. Scope: consumer web accounts in the United States and European Union. Include implementation tradeoffs, recovery, accessibility, adoption evidence, and credible objections. Deliver a decision brief with recommendation, evidence limits, contradictions, source-state counts, audit dispositions, and refresh triggers.

Before retrieval, expect the preserved inquiry, audience and decision use, scope and exclusions, coverage areas, source ecosystems, evidence burden, access boundary, and next useful move.

Create a durable campaign vault

python -B scripts/research_campaign.py init campaigns/passkeys --title "Passkey adoption decision" --query "Should small organizations adopt passkeys for customer accounts in 2026?" --tier focused

python -B scripts/research_campaign.py validate campaigns/passkeys

Expected output includes INITIALIZED and VALID. That proves an internally consistent empty campaign, not completed research.

Advance from the first unverified edge

Record query families and actual searches.
Ledger every retained candidate before counting inspection.
Write a substantive source note before marking deeply read.
Link claims to source IDs and record scope and confidence.
Update coverage and contradictions as the field changes.
Draft from an evidence digest and ordered section briefs.
Assemble and structurally audit the report.
Run a separate semantic claim review.

Expected campaign outputs

Artifact	Purpose
Research brief and campaign.json	Inquiry, decision, scope, state, budget, blockers, and resume point.
Query, source, and claim ledgers	Search path, source states, claims, support, confidence, and contradictions.
Source notes	Full-reading evidence, method, scope, dates, limits, and useful locations.
Coverage and contradiction maps	What matters, what is supported, what disagrees, and which gaps remain.
Report and audits	Bounded synthesis, structural resolution, semantic dispositions, limits, and refresh triggers.

Other realistic uses

Technical comparison

Compare two deployment approaches using official documentation, failure modes, migration costs, operational evidence, and version cutoffs.

Policy decision

Map statutes, guidance, empirical evidence, stakeholder incentives, competing interpretations, jurisdiction, and change triggers.

Historical investigation

Separate primary records, later scholarship, provenance disputes, missing archives, inference, and unresolved explanations.

Market or product inquiry

Distinguish vendor claims, customer evidence, public records, adoption proxies, counterexamples, and decision criteria.

Assemble and audit

python -B scripts/assemble_report.py campaigns/passkeyspython -B scripts/citation_audit.py campaigns/passkeys

A passing structural audit means markers resolve to eligible records. It does not prove that each source entails the exact sentence. Use the vault reference, prompt recipes, and trust boundary for the full operating model.

Review the actual answer

Assembly preserves differing reports until explicit replacement. Current completion binds actual review to the inquiry, sources, notes and all report prose. A word count or topic-matched citation cannot establish the inference. Commands and limits.