Comparative Research

Comparative Market Research & Recommendation

A structured, bias-resistant process for evaluating options and producing a sourced, auditable recommendation.

Core failure mode this skill prevents: Recommending well-documented options over better-fit options because their spec sheets are easier to research. This skill forces primary criterion supremacy at every phase.

Phase 1: Requirements Extraction

Before any research, extract and confirm all of the following. Do NOT proceed until the primary criterion is unambiguous — everything else flows from it.

Phase 2: Parallel Research

Run up to 3 research threads in parallel, each with a distinct focus:

Research rules

• Every claim must have a source URL. No unsourced statistics, ever.
• Missing data is declared, not fabricated. Say: "No survey exists mapping X to Y — using Z as proxy. This is an assumption."
• Distinguish installed base from new sales. Installed base = what users have now. New sales = what they're buying next. Report both when available.
• Capture distribution channel data. For physical products, offline/online split and retail footprint are often more predictive than headline market share.

Phase 3: Candidate Selection

⚠️ This is where most recommendation processes fail. Follow the sequence exactly.

Step 3a — Forced-rank by primary criterion

Using ONLY the market/user-fit data from Phase 2, rank all potential candidates by the primary criterion. Write this ranking down before looking at spec completeness.

Step 3b — Filter by hard constraints

Remove candidates that fail pass/fail requirements. Document each elimination with a specific reason — this goes in the Eliminated Candidates section of the report.

Step 3c — Treat data gaps as risks, not disqualifiers

For high-ranking candidates with incomplete spec data:

• ✅ Include them in the shortlist
• ⚠️ Flag each gap with a specific verification action (e.g. "verify camera FOV post-purchase by recording a known-width object at 60cm")
• ❌ Never drop a high-ranking candidate because it's harder to research than a lower-ranking alternative

Anti-pattern to avoid:

"Option B has all specs on GSMArena, so I'll include it. Option A is missing FOV data, so I'll mention it as a backup."

This is wrong. If Option A ranks higher on the primary criterion, it goes in the top N with the FOV gap flagged. Option B's spec-sheet completeness is irrelevant to its ranking.

Step 3d — Diversity check

Before finalizing:

• Does any single brand/category occupy more than one slot while scoring poorly on the primary criterion? Remove the weaker entry; replace with the next-best from a different brand/category.
• Do the top N collectively cover the user's stated purposes (e.g. worst-case AND realistic floor)?

Phase 4: Evaluation Criteria Definition

Define 10 or fewer criteria. For each:

• What: One-sentence definition
• Why: Why this criterion matters for this specific decision, tied to a concrete user requirement
• Evidence: Source data supporting the inclusion of this criterion

Criteria ordering:

1. Primary criterion first
2. Hard pass/fail constraints next
3. Differentiating factors last

Anti-patterns to avoid:

• Criteria that every candidate passes equally — these add noise without discrimination
• "Data availability" or "spec confidence" as a criterion — this directly biases toward well-documented options

Phase 5: Per-Candidate Analysis

For each candidate in the top N:

1. Spec table — assess every criterion with a status icon:

• ✅ PASS / CONFIRMED
• ⚠️ LIKELY PASS — unconfirmed; state what's missing and how to verify
• ❓ UNKNOWN — state what's missing
• ❌ FAIL

2. Sources — URLs for every claim

3. Strengths — 2–4 bullets, evidence-grounded

4. Weaknesses — 2–4 bullets, evidence-grounded. Never omit.

Phase 6: Comparison Matrix

Single table, all candidates side by side, all criteria as rows, status icons throughout. Must be scannable in 30 seconds.

Phase 7: Recommendation

1. Winner(s) — one per role the user defined
2. Why each winner wins — 4–6 bullets, tied directly to criteria and evidence, leading with the primary criterion
3. Why runners-up lost — specific and evidence-based (e.g. "Samsung has minimal sub-₹10K market share per IDC, and vibration motor is unconfirmed")
4. Purchase summary — model / price / where to buy / purpose, one table
5. Verification checklist — post-purchase steps to confirm any unverified specs

Phase 8: Assumptions & Gaps

Surface every assumption and data gap as a scannable table. Never bury them in prose.

Phase 9: Sources Index

Table mapping every source URL to what it was used for. Every claim in the report must be auditable from this index.

Output Format

Save the report as a Markdown file with a descriptive filename (e.g. phone-recommendation-2024.md).

# [Title]
_Generated: YYYY-MM-DD_
---
## Context
## Evaluation Criteria (C1–C10)
## Eliminated Candidates
## Top N Candidates
## Comparison Matrix
## Recommendation
## Verification Checklist
## Assumptions & Gaps
## Sources Index

Guardrails