research-analysis

Research & Analysis

Structured approaches to research, comparison, and synthesis across any domain.

Rigor Scaling

Before starting, assess the task. Not all research needs the same discipline.

Signal	Level	What Changes
Quick factual lookup, single source sufficient, low stakes	Lookup	Just answer. No framework needed.
Multi-source research, comparison, synthesis, or anything informing a decision	Standard	Apply routing table below. Resolve entities before searching. Tag key claims with source quality. Note gaps.
High-stakes research informing strategy, investment, hiring, or market entry — OR user says "be rigorous", "high confidence", "I need to be sure"	Rigorous	Full analytical discipline (see below).

Rigorous Mode Protocol

When rigorous research is warranted, apply these before presenting findings:

1. Decompose first. Break the question into 3-5 independent sub-questions. Don't research a blob — research specific, answerable pieces.

2. Entity resolution. Before any searches fire, resolve ambiguous names and entities bidirectionally:

• Person → company, role, handles, related entities ("Dario Amodei" → Anthropic CEO, @DarioAmodei)
• Company → key people, products, subsidiaries, common abbreviations ("SRSA" → SRS Acquiom)
• Product → maker, competitors, version history ("Loan IQ" → Finastra, formerly Misys)
• Acronym → full name AND full name → acronym ("SOFR" → Secured Overnight Financing Rate)
• If the entity is ambiguous (multiple possible matches), resolve it BEFORE searching. Ask the user if needed. This prevents searching for the wrong entity, missing relevant results, or conflating similarly-named things.

3. Source tier every claim.

• T1: Peer-reviewed, regulatory filing, official data, primary source code
• T2: Industry report, corroborated company data, established journalism
• T3: Blog post, PR, advocacy, single-source claim
• Rule: Load-bearing conclusions built on T3 alone get flagged explicitly

4. Assume-wrong check on top conclusions. For your 2-3 highest-conviction findings:

• State the finding
• Assume it's wrong — what would that mean?
• Name the strongest counter-argument
• Reconcile: does the finding hold, weaken, or change?
• Present the reconciled position, not just the initial one

5. Confidence and gaps. For each key conclusion:

• State your confidence (low / medium / high) with what drives it
• Name what evidence would change your mind
• Name what you couldn't find or verify

6. Anti-findings. Explicitly state what you looked for and did NOT find. Absence of evidence is itself a finding when documented.

The goal: research you'd feel confident presenting to someone who will make a real decision based on it. Not sigma-review depth, but more than "I searched and here's what I found."

---

Routing

If the user wants to...	Read
Compare competitors, analyze market positioning, feature matrices	references/competitive-analysis.md
Synthesize user research — interviews, surveys, feedback — into insights	references/user-research-synthesis.md
Plan or conduct user research — interview guides, usability tests	references/user-research.md
Profile and explore a dataset before analysis	references/data-exploration.md
Synthesize information from multiple internal or external sources	references/knowledge-synthesis.md
Design a research strategy — what sources, what approach, what sequence	references/search-strategy.md
Gather competitive intelligence for sales positioning	references/competitive-intelligence.md
Research an account, company, or contact	references/account-research.md
Benchmark compensation, analyze pay equity	references/comp-analysis.md

Gotchas

• Rigor should match stakes, not effort. A quick competitive scan doesn't need dialectical bootstrapping. A market entry analysis does. If unsure, ask: "Is someone making a decision based on this?"
• Research and analysis are different activities. Research = gathering information. Analysis = making sense of it. Do both.
• Competitive analysis should focus on positioning implications, not just feature checklists.
• User research synthesis should preserve participant voice — don't sanitize findings into what stakeholders want to hear.
• Account research should be actionable — "here's what to say in the meeting" not just "here's what I found."
• Always identify what you DON'T know. Gaps in research are as important as findings.
• T3 sources aren't invalid — they're just not load-bearing. A blog post can provide a lead. It can't be the foundation of a strategic conclusion.
• "I found 5 sources that agree" ≠ high confidence if they're all citing the same original source. Trace the citation chain.
• The assume-wrong check is not theater. If you can't articulate a genuine counter-argument, either the finding is trivially true (doesn't need rigorous treatment) or you haven't looked hard enough.

Pre-Step

If the user hasn't fully articulated what they need researched — vague scope, unclear questions, unstated assumptions — suggest socratic-grill Extract mode first: "Want me to ask you some questions to scope this before I start researching? It usually produces more targeted results." The socratic-grill session produces a Q/H/C decomposition that feeds directly into rigorous-mode research.

When the Skill Doesn't Cover It

If you load this skill and the reference material does not answer the question — do NOT guess, generalize, or answer from general knowledge alone. Instead:

1. Say what the skill covers and where the gap is. "The loan-agency references cover payment waterfalls but not CLO compliance testing mechanics."
2. Trigger a rigorous web search. Search for current, authoritative sources. Apply source tiers: T1 (official/peer-reviewed) > T2 (industry reports) > T3 (blogs/PR).
3. Flag the provenance. "This answer comes from web research, not the skill references — [source, tier]."
4. Suggest a skill update if the gap is recurring. "This came up before — worth adding to the skill?"

The worst outcome is a confident wrong answer built on stale knowledge. A searched answer with source attribution is always better than an unsourced guess.