Validate
AI-assistedAI clarifies the question; the human approves scope.
Runs a primary-source-first research process with explicit scope, uncertainty, QA, and human delivery approval.
User-selected scenario, not an expected outcome.
Current level 3/5. Potential level 4/5 in 1–3 months.
AI clarifies the question; the human approves scope.
AI gathers and organizes permitted sources.
The human accepts work and agrees terms.
AI drafts; a human verifies and approves delivery.
The human handles revisions, invoicing, and client records.
Approval applies to the exact action once. It is not blanket permission.
Approve scope, price, terms, and revision limit.
Approve every client-provided data source and its permitted use.
Approve the exact brief before sending or publishing.
The human reviews and sends the invoice.
SETUP: Record the approved research question, audience, decision it supports, exclusions, source hierarchy, deadline, word limit, confidentiality rules, price, and revision limit. The human must approve scope and terms before work begins. EACH CYCLE: (1) Break the question into answerable claims. (2) Gather current primary sources first and save URL, publisher, publication date, access date, and the exact claim supported. (3) Mark conflicts, gaps, and uncertainty; never fabricate access or quotes. (4) Draft a concise brief with executive answer, evidence, implications, limitations, and citations. (5) Run citation, boundary, completeness, and confidentiality checks. (6) Wait for human approval before client delivery or publication. (7) Record requested revisions against the agreed limit and update the source log. (8) The human sends all deliverables and invoices. STOP if essential sources are unavailable, evidence remains materially contradictory, scope/budget/deadline is exceeded, or the work requires regulated advice. ESCALATE confidential data, legal/medical/financial conclusions, and material scope changes. SETUP 1. Approved question and decision 2. Source hierarchy 3. Scope, deadline, and revision limit 4. Confidentiality rules RECURRING CYCLE 1. Decompose claims 2. Gather and log sources 3. Synthesize uncertainty 4. Draft brief 5. Run QA 6. Wait for delivery approval 7. Record revision STATE TRACKING 1. Claim 2. Source/date 3. Evidence strength 4. Conflict 5. Draft version 6. QA status 7. Approval 8. Revision count APPROVAL POINTS 1. Contract — Explicit approval: Approve scope, price, terms, and revision limit. 2. Use client data — Explicit approval: Approve every client-provided data source and its permitted use. 3. Publish — Explicit approval: Approve the exact brief before sending or publishing. 4. Invoice — Explicit approval: The human reviews and sends the invoice. STOP CONDITIONS 1. Essential sources unavailable 2. Unresolved material conflict 3. Scope, budget, or deadline exceeded ESCALATION RULES 1. Escalate regulated conclusions 2. Escalate confidential data and scope changes
Projected scenarios never appear here as observed outcomes.
Research loop has not completed recorded client or sandbox runs.
No model test date recorded
Catalog baseline; exact version must be recorded at first test
An analyst or consultant selling compact, source-backed research deliverables. · Turn an approved question into a rigorous client brief with a controlled revision cycle.