Horizontal-Vertical Analysis Method

Deep Research Framework — v2.0 (2026 Q2): Reusable prompt for anyone who regularly needs to evaluate technologies, companies, tools, or people — analysts, founders, investors, product leads.

Variable Definition

Research Subject    = 
Execution Model     = [ Claude Opus 4.6 | Gemini 3.1 Pro | DeepSeek V4 Pro | GPT-5.5 ]
Search Augmented    = [ Yes | No ]
Output Mode         = [ Single-pass | Pipeline ]

To reuse this framework: change only the values to the right of each equals sign. All downstream instructions self-adapt.

Pre-Execution Checklist

Before generating the report, state explicitly:

Knowledge confidence level for this Research Subject: High / Medium / Low / Speculative
- High = well-documented, dense public record
- Medium = partial coverage, some gaps expected
- Low = thin coverage, significant inference required
- Speculative = minimal verified data, output is primarily structured reasoning
Search augmentation status: Is real-time retrieval active for this execution?
- If No and confidence level is Medium or below: flag this prominently before the report body. The horizontal analysis section will reflect data as of your training cutoff, not today.
Subject type classification: Product / Tool / Company / Technical Concept / Person
- This determines which analytic dimensions are prioritized in each section.

I. Vertical Analysis — Longitudinal Depth

Trace the full arc of the Research Subject along the time axis, from origin to present.

Required Coverage

1. Origin Context What was the environment that made this possible or necessary? What technology, need, or idea does it build on? Who drove it into existence, and what were their motivations? What did the competitive or intellectual landscape look like at the moment of creation?

2. Birth Milestone The precise first public moment — release, founding, publication, or proposal — and what the initial form and positioning actually were. Not what it became. What it was.

3. Evolution Map Every significant node from origin to present, in chronological order. Include:

Major version releases or strategic pivots
Funding events and capital structure changes
Team changes that affected direction
Architectural or methodological shifts
User or revenue scale milestones
Key partnerships, acquisitions, or integrations
Public controversies, failures, or crisis moments

4. Decision Logic Reconstruction At each critical node: why this choice and not another? What constraints existed? What options were available but rejected? What was the logic — stated or inferable?

Label clearly when this is reconstruction from public evidence vs. directly documented reasoning.

5. Narrative Requirement Write this as a story, not a timeline. The reader should feel the causality — why one moment made the next one inevitable or surprising. Bring in the people, the context, the pressures. If a product suddenly exploded in adoption or collapsed in relevance, the narrative should make that feel earned, not arbitrary.

II. Horizontal Analysis — Synchronic Breadth

Cross-section at the current moment. Compare the Research Subject against its competitive field.

Step 1 — Competitive Landscape Assessment

Before analysis, classify the competitive situation:

Scenario A — No direct competitors The Research Subject is a genuinely new category or holds a near-monopoly position. Skip competitor-by-competitor analysis. Instead analyze:

Why does no direct competitor exist? (Category too new / moat too high / market too small / regulatory barrier)
What indirect substitutes or legacy solutions exist as reference points?
From which direction — and on what timeline — is a competitor most likely to emerge?

Scenario B — Few competitors (1–2) Go deep on each. No surface-level comparisons.

Scenario C — Sufficient competitors (3+) Select the 3–5 most representative. Others may be acknowledged briefly. Each selected competitor warrants standalone analysis, not a row in a table.

Step 2 — Comparison Dimensions

Adjust these dimensions based on subject type. Apply all that are relevant:

Technical / Methodological

Core approach, architecture, or methodology
Key technical differentiators or constraints
What the underlying bet is — what has to be true for this to work

Product / Commercial

Product form and user experience
Business model and revenue structure
Pricing strategy and cost structure
Scale: users, revenue, team, infrastructure

Positioning

Target user and actual user (these often diverge — note it when they do)
Use cases where this wins vs. where it loses
What problem it solves that others don't, or solves better

User Reality

Community and review sentiment: what do users actually say?
Most cited strengths and most cited frustrations
Gap between official positioning and actual usage patterns

Ecosystem

Integrations, partnerships, platform dependencies
Community, developer ecosystem, extension surface
Lock-in mechanisms or switching costs

Step 3 — Niche and Trajectory

Niche Analysis: In the full map of this space, what position does the Research Subject occupy? Does it fill a gap, compete head-on, or create a new category?

Trend Judgment: Based on the cross-sectional picture, what is the most likely competitive trajectory? What would have to change for the Research Subject to win decisively, lose ground, or get acquired?

III. Writing Constraints

These are not stylistic preferences. They are quality controls targeting known LLM failure modes.

Failure Mode	Rule
Bullet-point enumeration substituting for synthesis	Write in prose. Use lists only for genuine enumerations, never for analysis.
Consulting-speak and hollow abstractions	No "leverage," "synergy," "empower," "ecosystem flywheel," or equivalent. Replace with concrete specifics.
Ungrounded opinion	Facts first. Judgment follows. Never invert this order.
Confident hallucination	If you cannot verify it, say so. Label speculation explicitly with `[Inferred]` or `[Unverified]`.
Feature matrix dressed as comparison	Each competitor analysis must explain what it became and why users actually chose it — not just what features it has.
Narrative flatness	The vertical section must have tension, turning points, and causality. A chronology without narrative logic is a failure mode.
False completeness	If knowledge confidence is Low or Speculative, the report must communicate this clearly and structurally — not just in a one-line caveat.

IV. Length Calibration

Self-calibrate based on subject complexity and knowledge confidence level. Do not pad. Do not compress at the cost of key detail.

Section	Range	Principle
Vertical Analysis	6,000–15,000 words	Proportional to history depth and node count. New subjects toward lower bound, long-lived complex subjects toward upper.
Horizontal Analysis	3,000–10,000 words	Proportional to competitive density. Scenario A toward lower bound, Scenario C toward upper. Each major competitor: minimum 1,500 words.
Synthesis	1,500–3,000 words	Must deliver new integrated judgment — not a summary of what came before.
Total	10,000–30,000 words	Stop when the story is complete. Not before. Not after.

V. Output Structure

Execute in this order:

Pre-Execution Statement — confidence level, search status, subject type classification
Vertical Analysis — chronological narrative
Horizontal Analysis — competitive cross-section
Horizontal-Vertical Synthesis — integrated forward judgment

Synthesis Requirements

This section must not recap the preceding content. It must answer:

Given the full history, where does the Research Subject actually stand right now?
What does the competitive picture reveal that the history alone does not?
What is the most credible trajectory from here — and what would have to be true for it to play out?
What is the single highest-leverage question an analyst, investor, or builder should be asking about this subject right now?

Attribution Standard

Cite sources or timestamp claims wherever possible
Label inferences explicitly: [Inferred from X]
Label unverified information: [Unverified — requires confirmation]
Never fabricate. A gap acknowledged is always better than a gap filled with plausible fiction.

VI. Recommended Execution Configuration (2026 Q2)

Model	Strength in This Framework	Known Weakness	Search Required?
Claude Opus 4.6	Narrative quality, synthesis, uncertainty flagging, style compliance	Knowledge cutoff; thinner on niche subjects	Yes — horizontal section
Gemini 3.1 Pro	Real-time retrieval, long context (1M tokens), agentic workflows	Defaults to lists; needs explicit prose enforcement	Built-in (grounding)
DeepSeek V4 Pro	Technical depth, decision logic reconstruction, cost efficiency ($1.74/M input)	English prose register; less Western AI ecosystem coverage	Yes — strongly recommended
GPT-5.5	Breadth, polish, confident synthesis	Highest hallucination risk on niche subjects; least reliable uncertainty flagging	Yes — critical for this subject

Recommended pipeline for maximum output quality:

Stage 1 — Retrieval     : Gemini 3.1 Pro (grounded search) or any model + web tool
Stage 2 — Vertical      : Claude Opus 4.6 (narrative + causality)
Stage 3 — Horizontal    : DeepSeek V4 Pro (technical comparison) + Gemini (current data)
Stage 4 — Synthesis     : Claude Opus 4.6
Stage 5 — Audit         : Cross-check factual claims across ≥2 model outputs

VII. Applicable Subject Types

This framework applies without modification to:

Products / Tools: e.g., Hermes Agent, Cursor, Claude Code
Companies / Organizations: e.g., Anthropic, DeepSeek, OpenAI
Technical Concepts: e.g., MCP Protocol, RAG, Agent frameworks
People: career trajectory and peer comparison of key figures

Core principle remains constant across all types:

Vertical analysis pursues depth through time. Horizontal analysis pursues breadth across the contemporaneous field. Their intersection yields judgment.