analyze

name	analyze
description	Stage-aware due diligence analysis of a project across business, technical, and moat/strategy dimensions, weighted by lifecycle stage. Supports remote URLs, local directories, and local files.

Research-Forge — Analyze

Deep due diligence analysis across the three dimensions defined in the Research-Forge framework, with scoring weights determined by the project's lifecycle stage.

Command: /research-forge:analyze

Usage:

/research-forge:analyze <target> [--lang zh|en] [--focus business|technical|strategy]

• Target can be a URL, local directory path, or local file path.
• --focus: Optionally deep-dive into one specific dimension while keeping others at summary level.
• Default output language: Chinese (Simplified). Override with --lang en.

Analysis Workflow:

Phase 0: Stage Detection

If a scan has already classified the stage (auto-chain pipeline), reuse it. Otherwise classify the project into S0/S1/S2/S3 per the Stage Model, with evidence and confidence level. Then load the stage's weight profile from the framework — it decides which sub-dimensions are Critical (×2), Standard (×1), or Context (×0, reported but unscored).

State up front:

• Stage: [S0-S3] with evidence
• One Question this analysis must answer: [the stage's core question]
• Death cause to hunt for: [the stage's characteristic failure mode]

Phase 1: Data Collection

Remote targets:

1. WebFetch the project's README, docs site, and landing page.
2. WebSearch for:
   • Recent news, blog posts, or announcements about the project.
   • Funding rounds, acquisitions, or partnership announcements.
   • Community discussions (HN, Reddit, X.com) for sentiment.
   • Competitor landscape — search for "[project] vs" and "[project] alternatives".
3. If a GitHub URL, extract repo metrics via the GitHub web interface.

Local targets:

1. Read README, manifest files, key config files.
2. Glob to map project structure: source, tests, docs, CI config.
3. Read representative source files (entry points, core modules, tests) to assess code quality.
4. Grep for patterns: TODO/FIXME density, error handling, logging, security patterns.
5. Check git history for activity patterns, contributor stats, branch hygiene.
6. If git remote exists, WebSearch for external context about the project.
7. WebSearch for competitor landscape regardless of target type.

Prioritize collection toward the stage's Critical sub-dimensions — e.g., for S1 dig for retention proxies (repeat contributors, dependents growth, organic mentions); for S2 dig for pricing, margin structure, and hiring signals; for S3 dig for ecosystem breadth and emerging paradigm threats.

Phase 2: Business & Market Analysis

Rate each sub-dimension (1-5), applying the stage weight profile:

• Problem Validity (1-5): Is the pain real and acute? Key evidence: users already paying a cost today — money, time, or duct-taped workarounds. No current coping behavior → pseudo-demand.
• Why Now (1-5): Name the specific inflection — technology, cost, behavior, or regulation — that makes this viable now. No inflection = too early or too late.
• PMF Evidence (1-5): Retention over acquisition. Commercial: cohort retention flattening, organic growth share, Sean Ellis ≥ 40%. OSS proxies: repeat-contributor ratio, dependents growth, issue-reporter recurrence, production testimonials vs. promoted star spikes.
• Unit Economics (1-5): LTV/CAC (≥3 paper bar), gross margin direction with scale, NDR (>100% healthy, >120% excellent). OSS: credible open-core boundary and conversion path. Scale that worsens economics = wholesaling losses.
• Market Ceiling (1-5): Terminal market, not current TAM — total spend this could absorb at its limit.
• Growth Engine (1-5): Identifiable, repeatable engine (product-led / content / sales / paid-with-payback) vs. founder charisma and one-off wins.

Phase 3: Technical & Architecture Analysis

Rate each sub-dimension (1-5), applying the stage weight profile:

• Architecture Quality (1-5): Modularity, design patterns, API quality.
• Engineering Culture (1-5): Code quality, testing, CI/CD, PR review.
• Scalability (1-5): Can it handle 10x/100x growth without fundamental redesign?
• Security Posture (1-5): Dependency hygiene, vulnerability handling.
• Documentation (1-5): Quality, completeness, maintenance.
• Maintainability & Tech Debt (1-5): TODO density, deprecated APIs, stale branches, long-lived PRs.

For local targets, the Technical dimension benefits from direct code reading — provide specific evidence from the codebase (file paths, code patterns, architecture observations).

Phase 4: Moat & Strategy Analysis

Rate each sub-dimension (1-5), applying the stage weight profile:

• Moat Strength (1-5): Rate against the moat hierarchy — 5 network effects, 4 switching costs, 3 scale economies/brand-equals-category, 2 data/tech lead only (AI-era note: pure model lead lasts ~6-18 months; a data flywheel beats the model itself), 1 no moat.
• Founder-Market Fit (1-5): Unfair advantage — unique insight, resources, or technology. Include key-person risk.
• Org Capability (1-5): Can the org scale past the founders — delegation, senior hires, mid-level strength?
• Second Curve Readiness (1-5): Paradigm-shift exposure; willingness to self-disrupt; product → platform/ecosystem trajectory.

Then construct the qualitative theses:

• Bull Case (3-5 strongest arguments for success).
• Bear Case (3-5 most significant risks — MUST include an explicit check of the current stage's death cause, even if only to rule it out).
• Comparable Outcomes: similar projects and their trajectories.

Phase 5: Verdict & Kill Criteria

• Compute Overall Score: weighted average of non-Context sub-dimensions (Critical ×2, Standard ×1). List Context sub-dimensions separately as ctx — unscored.
• Produce Strategic Verdict: Strong Recommend | Watch Closely | Hold / Wait | Not Recommended.
  • Any Critical sub-dimension ≤ 2 caps the verdict at Hold / Wait.
  • Strong Recommend requires all Critical sub-dimensions ≥ 3.
• Produce Investment Signal (when applicable): Strong Buy / Watch / Hold / Pass.
• Write Kill Criteria (mandatory): 2-4 pre-registered, falsifiable, time-bound signals that would flip the verdict.
• Write a Final Evaluation paragraph: state the stage, answer the stage's One Question directly, then synthesize.

Phase 6: Output

Generate a structured report with:

1. Executive Summary (3-5 sentences, opening with stage + One Question answer)
2. Scorecard table (all sub-dimensions with ratings, weight class per stage, and confidence flags)
3. Detailed analysis per dimension
4. Bull Case / Bear Case (with stage death-cause check)
5. Verdict with confidence level (High/Medium/Low) + Kill Criteria
6. Recommended actions / next steps