Deep research

Skill wme3/DeliberateDecisions/skills/deep-research

Gate 3 - Web search, document analysis, reference class, gather evidenceFrom its SKILL.md

Install

npx -y skills add wme3/DeliberateDecisions --skill deep-research

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.4 KB, as published. Nobody here has run it

Gate 3: Deep Research

Purpose: Gather evidence to fill knowledge gaps and ground the decision in facts.

Announce: "Moving to Research Gate - let's gather evidence."

Entry Criteria

Landscape Gate completed
Unknown-Knowables identified

Research Sources

Use all available sources:

Conversation - Ask human for internal knowledge, context, documents
Documents - Analyze provided reports, financials, memos
Web research - Search for market data, competitor info, industry analysis

Process

1. Prioritize Research Questions

Review Unknown-Knowables from Landscape Gate. Prioritize by:

Impact on decision (high impact first)
Feasibility of finding answer
Time available

2. Gather Evidence

For each research question:

Ask human first:

"Do you have internal data on [X]?"
"Is there a document that covers [Y]?"

Then search externally:

Market data and trends
Competitor analysis
Industry reports
Academic research
News and recent developments

3. Find Reference Class (Kahneman)

Critical for countering overconfidence.

Ask:

What similar decisions have been made before?
What's the base rate of success for decisions like this?
What happened to others who made this choice?

Example: "M&A deals in this sector have a 60% failure rate in achieving projected synergies."

4. Evaluate Evidence Quality

For each piece of evidence, assess:

Source credibility: Who produced this? What's their bias?
Recency: Is this current or outdated?
Relevance: Does this directly apply to our situation?
Methodology: How was this data gathered?

Use deliberate-decisions:evidence-evaluation for detailed assessment.

Depth by Weight

Aspect	Light	Medium	Complete
Web research	None - user knowledge only	1-2 targeted searches	Thorough multi-source
Reference class	Ask user if known	Find basic reference	Deep reference class analysis
Document review	Skip unless provided	Review key docs	Comprehensive review
Evidence quality	Trust user input	Basic assessment	Full quality evaluation

Light: Rely on user's domain knowledge. Ask if they know the reference class. Skip external research unless user lacks knowledge.

Medium: 1-2 targeted web searches for key unknowns. Find basic reference class data. Standard evidence assessment.

Complete: Thorough external research. Multiple sources for key questions. Deep reference class analysis. Full evidence quality evaluation.

Upgrade Detection

Suggest upgrading if:

Reference class reveals higher risk than expected
Research uncovers surprising/contradictory information
Key unknowns remain after Light research
User's domain knowledge has significant gaps

Upgrade prompt:

⚠️ Research is revealing [unexpected findings]:
- [Finding 1]
- [Finding 2]

This suggests the decision may be higher-stakes than initially framed.

Current: [Weight]
Suggested: [Higher Weight] - would allow [deeper research / more sources]

Continue at current depth, or upgrade?

Output

Create research notes:

# Research Notes: [Decision]

## Research Questions

| Question | Priority | Status |
|----------|----------|--------|
| [Unknown-Knowable 1] | High | Researched |
| [Unknown-Knowable 2] | Medium | Partial |

## Findings

### [Topic 1]

**Finding:** [what we learned]
**Source:** [citation]
**Quality:** [high/medium/low]
**Relevance:** [direct/indirect]

### [Topic 2]
...

## Reference Class

**Similar decisions:** [examples]
**Base rate:** [success rate for this type of decision]
**Key lessons:** [what others learned]

## Remaining Gaps

- [Questions we couldn't answer]
- [Areas needing more research]

Save to: docs/decisions/YYYY-MM-DD-<decision-slug>/research-notes.md

Exit Criteria

High-priority Unknown-Knowables researched (depth per weight)
Reference class identified (depth per weight)
Evidence quality assessed (depth per weight)
Remaining gaps documented

Bias Watch

Watch for:

Confirmation bias - Only seeking evidence that supports preferred option
Availability bias - Over-weighting easily found information
Authority bias - Accepting claims because of source prestige

Actively search for disconfirming evidence.

Next Gate

Proceed to: deliberate-decisions:calibration