Rory-X/ppt-research

PPT Research Analyst

When to Use

Before designing any PPT, conduct thorough research on the topic to gather facts, trends, risks, and opposing viewpoints.

Research Dimensions

You MUST search across at least 3 of these dimensions:

1. Market/Industry Status -- current state, trends, growth trajectory
2. Competitors/Alternatives -- how others present or solve similar topics
3. Audience Focus -- what the target audience cares about, their pain points and expectations
4. Counter-Arguments -- opposing views, risks, limitations, trade-offs
5. Technology Reality -- actual technical feasibility, maturity level, adoption status

Process

1. Read the brief (topic, audience, scenario, goal).
2. Formulate at least 3 distinct WebSearch queries covering different dimensions.
3. Execute searches and collect findings.
4. If user provided materials (URLs, notes), integrate them as primary sources.
5. Synthesize findings into a structured report.

Output: research-report.md

Write the report to the artifacts directory. It MUST include:

• Topic and generation timestamp
• Sources with type (web/user-material) and reference URL/text
• Findings grouped by dimension, each with:
  • Insight (one sentence)
  • Evidence (specific fact or quote)
  • Confidence level (high/medium/low)
  • Risk or caveat (if any)
• Recommendation section:
  • Suggested narrative angle
  • Suggested style direction (technical-share / pitch-deck / other)
  • Rationale (2-3 bullet points)

Quality Gates

• At least 3 search dimensions covered
• At least one counter-argument or risk included
• All claims backed by cited sources
• No fabricated statistics or unsourced claims

Source Quality Ladder

Prioritize higher-tier sources. Every research report MUST include at least one Tier 1 or Tier 2 source.

| Tier | Examples | Trust Level | |------|----------|-------------| | Tier 1 (highest) | Academic papers, official documentation, government statistics | Cite directly; no hedging needed | | Tier 2 | Industry reports (Gartner, McKinsey, Forrester), established media (Reuters, Bloomberg, WSJ) | Cite directly; note report date | | Tier 3 | Tech blogs from recognized authors, company engineering blogs (Netflix Tech Blog, Uber Engineering) | Cite with author attribution; cross-reference when possible | | Tier 4 (lowest) | Forums, social media, anonymous posts, AI-generated summaries | Use only as supplementary signal; never as sole evidence |

Rule: If a finding relies solely on Tier 3-4 sources, mark its confidence as low regardless of other factors.

Search Strategy Templates

For each dimension, use these query patterns. Replace [topic], [audience], and [competitor] with actual values. Always add time and locale qualifiers to avoid stale results.

Market / Industry

• "[topic] market size 2025-2026"
• "[topic] industry trend site:mckinsey.com OR site:gartner.com"
• "[topic] adoption rate enterprise 2026"

Competitor / Alternatives

• "[topic] alternatives comparison 2026"
• "[topic] vs [competitor] benchmark"
• "[topic] competitive landscape analysis"

Audience Focus

• "[audience role] challenges [topic] 2026"
• "what [audience] need to know about [topic]"
• "[audience role] survey [topic] pain points"

Counter-Argument

• "[topic] criticism"
• "[topic] limitations risks"
• "why [topic] fails" OR "[topic] drawbacks"

Technology Reality

• "[topic] production experience"
• "[topic] case study real-world"
• "[topic] post-mortem lessons learned"

Confidence Levels

Assign one of these levels to every finding:

• high: Multiple independent sources agree, quantitative data available, sources are Tier 1-2.
• medium: Single reliable source (Tier 1-2), or multiple Tier 3 sources with minor discrepancies.
• low: Inference/analogy from adjacent domains, single blog post, anecdotal evidence, or sole Tier 4 source.

When aggregating findings into the recommendation section, weight high-confidence findings most heavily. Flag any recommendation that rests primarily on low-confidence findings with [WEAK_EVIDENCE].

Contradiction Handling

When user-provided materials contradict search results:

1. Tag the finding with [CONFLICT] in the report.
2. Present both sides with full source attribution:
   • User material says: (quote or paraphrase) — source: user-provided
   • Search results say: (quote or paraphrase) — source: URL
3. Do NOT silently favor either side; let the outline architect decide which angle to adopt.
4. If two search results contradict each other (no user material involved), tag with [DISPUTED] and list both sources.

Finding ID System

Every finding MUST have a unique ID following this format:

- **F1** [dimension:market] Insight text here (source: URL, confidence: high)
- **F2** [dimension:competitor] Insight text here (source: URL, confidence: medium)
- **F3** [dimension:audience] Insight text here (source: user-material, confidence: high)

Rules:

• IDs are sequential integers: F1, F2, F3, ...
• Each ID includes its dimension tag in brackets.
• These IDs are referenced downstream by outline.json (evidence_refs) and by the reviewer (fact-checking).
• Never reuse an ID within a single report. If a finding is updated, keep its original ID.

Schema Alignment

The final research-report.md output MUST conform to the structure defined in schemas/research-report.schema.json.

Before finalizing the report, cross-check:

1. All required top-level fields are present (topic, timestamp, sources, findings, recommendation).
2. Every finding has: id, dimension, insight, evidence, confidence, and source.
3. The sources array includes every URL/material referenced in findings.
4. Finding IDs are unique and sequential.
5. At least one source is Tier 1 or Tier 2 (see Source Quality Ladder).