---
name: research-ops
description: Evidence-first research workflow. Use when the user wants fresh facts, comparisons, enrichment, recommendations, or any task requiring current public information. Triggers on 'research', 'look up', 'compare', 'what's the latest', 'who should I talk to'. Separates sourced facts from inference from recommendations.
---

# Research Ops

Evidence-first current-state research workflow. Do not answer current questions from stale memory when fresh search is cheap.

## When to Use

- User says "research", "look up", "compare", "who should I talk to", "what's the latest"
- The answer depends on current public information
- User supplied evidence and wants it factored into a fresh recommendation
- The task may be recurring enough to become a monitor instead of one-off

## Workflow

### 1. Start From What the User Already Gave

Normalize supplied material into: already-evidenced facts, needs verification, open questions. Do not restart from zero if the user built part of the model.

### 2. Classify the Ask

| Type | Description | Approach |
|------|-------------|----------|
| Quick factual | Single answer needed | One search, verify, report |
| Comparison/decision | Multiple options to evaluate | Multi-source, score matrix |
| Lead/enrichment | People or company targeting | Signal scoring, warm path |
| Monitoring candidate | Recurring question | Recommend automation |

### 3. Take the Lightest Evidence Path First

Escalate only when needed:
1. Quick web search for fast discovery
2. Deep multi-source synthesis when multiple sources matter
3. Market research when outcome should end in a recommendation
4. Lead intelligence when the real ask is target ranking

### 4. Report With Explicit Evidence Boundaries

For every important claim, label it:
- **Sourced fact** — with date and source
- **User-supplied context** — from what user provided
- **Inference** — logical conclusion from evidence
- **Recommendation** — your suggested action

Freshness-sensitive answers must include concrete dates.

### 5. Decide If This Should Stay Manual

If the user is likely to ask the same research question repeatedly, say so explicitly and recommend a monitoring/automation layer instead of repeating manual search.

## Output Format

```
QUESTION TYPE: factual / comparison / enrichment / monitoring

EVIDENCE:
- [sourced facts with dates]
- [user-provided context]

INFERENCE:
- [what follows from evidence]

RECOMMENDATION:
- [answer or next move]
- [whether this should become a monitor]
```

## Guardrails

- Do not answer current questions from stale memory when fresh search is cheap
- Separate sourced fact from inference from recommendation — always
- Do not spin up heavyweight research if the answer is already in local docs
- Specificity beats vague summaries