Skip to content

AI agent skills for startup validation, competitive intelligence, and planning

feat: add verification agent and dynamic research scaling to all skills

Ferdinando Bonsegna1dbf3926043d
diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
index e89c3b0..760731e 100644
--- a/.claude-plugin/marketplace.json
+++ b/.claude-plugin/marketplace.json
@@ -6,7 +6,7 @@
},
"metadata": {
"description": "AI agent skills for startup design, validation, competitive intelligence, and planning",
- "version": "1.6.0"
+ "version": "1.7.0"
},
"plugins": [
{
diff --git a/AGENTS.md b/AGENTS.md
index c74c6f8..e9e9501 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -29,37 +29,45 @@ startup-skill/ # Plugin namespace (startup:*)
│ ├── SKILL.md # Main skill file (~522 lines, 8 phases)
│ └── references/ # Supporting documents loaded on demand
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-market.md
│ ├── research-wave-2-competitors.md
│ ├── research-wave-3-customers.md
│ ├── research-wave-4-distribution.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ └── frameworks.md
├── startup-competitors/ # Skill: startup:startup-competitors
│ ├── SKILL.md # Main skill file (~220 lines, 3 waves)
│ └── references/
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-profiles-pricing.md
│ ├── research-wave-2-sentiment-mining.md
│ ├── research-wave-3-gtm-signals.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ └── honesty-protocol.md
├── startup-positioning/ # Skill: startup:startup-positioning
│ ├── SKILL.md # Main skill file (~240 lines, 2 waves)
│ └── references/
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-alternatives.md
│ ├── research-wave-2-market-frame.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ ├── frameworks.md
│ └── honesty-protocol.md
├── startup-pitch/ # Skill: startup:startup-pitch
│ ├── SKILL.md # Main skill file (~370 lines, 2 waves)
│ └── references/
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-audience-narrative.md
│ ├── research-wave-2-competitive-framing.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ ├── pitch-frameworks.md
│ └── honesty-protocol.md
├── CLAUDE.md
@@ -110,6 +118,20 @@ startup-skill/ # Plugin namespace (startup:*)
Each wave must complete before the next starts. Agents use WebSearch for real data. All skills support Claude.ai (sequential fallback) and Knowledge-Based Mode when WebSearch is unavailable.
+### Research Scaling
+
+All skills use dynamic research scaling. After intake, a complexity score (3-9) is calculated from market breadth, known competitors, and geographic scope. This maps to three tiers:
+
+- **Light** (3-4): Merged agents, 2-3 search rounds. Quick scan for niche markets.
+- **Standard** (5-7): Default behavior, unchanged from current wave structure.
+- **Deep** (8-9): Extra agents per wave, 5-6 search rounds. Thorough due diligence.
+
+Users can override with "light", "deep", or "ok". See each skill's `references/research-scaling.md` for wave configurations per tier.
+
+### Verification Agent
+
+All skills run a **V1: Verification** agent after synthesis. It audits deliverables for: unlabeled claims, internal contradictions, confidence rating consistency, missing data gaps, missing flags, stale data, and duplicate-source false corroboration. Each skill adds domain-specific checks (e.g., cross-phase consistency for startup-design, battle card coherence for startup-competitors). Output: `verification-report.md`. Critical issues pause the flow for user review. See each skill's `references/verification-agent.md`.
+
### Output Structure
**startup-design** generates files in subdirectories:
diff --git a/CLAUDE.md b/CLAUDE.md
index 1176233..471d4e7 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -29,37 +29,45 @@ startup-skill/ # Plugin namespace (startup:*)
│ ├── SKILL.md # Main skill file (~522 lines, 8 phases)
│ └── references/ # Supporting documents loaded on demand
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-market.md
│ ├── research-wave-2-competitors.md
│ ├── research-wave-3-customers.md
│ ├── research-wave-4-distribution.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ └── frameworks.md
├── startup-competitors/ # Skill: startup:startup-competitors
│ ├── SKILL.md # Main skill file (~220 lines, 3 waves)
│ └── references/
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-profiles-pricing.md
│ ├── research-wave-2-sentiment-mining.md
│ ├── research-wave-3-gtm-signals.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ └── honesty-protocol.md
├── startup-positioning/ # Skill: startup:startup-positioning
│ ├── SKILL.md # Main skill file (~240 lines, 2 waves)
│ └── references/
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-alternatives.md
│ ├── research-wave-2-market-frame.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ ├── frameworks.md
│ └── honesty-protocol.md
├── startup-pitch/ # Skill: startup:startup-pitch
│ ├── SKILL.md # Main skill file (~370 lines, 2 waves)
│ └── references/
│ ├── research-principles.md
+│ ├── research-scaling.md
│ ├── research-wave-1-audience-narrative.md
│ ├── research-wave-2-competitive-framing.md
│ ├── research-synthesis.md
+│ ├── verification-agent.md
│ ├── pitch-frameworks.md
│ └── honesty-protocol.md
├── CLAUDE.md
@@ -109,6 +117,10 @@ startup-skill/ # Plugin namespace (startup:*)
Each wave must complete before the next starts. Agents use WebSearch for real data. All skills support Claude.ai (sequential fallback) and Knowledge-Based Mode when WebSearch is unavailable.
+**Research Scaling:** All skills dynamically adjust research depth. After intake, a complexity score (3-9) maps to Light/Standard/Deep tiers, changing agent count and search rounds per wave. Users can override. See each skill's `references/research-scaling.md`.
+
+**Verification Agent:** All skills run a V1: Verification agent after synthesis. It audits deliverables for consistency, unlabeled claims, and skill-specific coherence. Critical issues pause for user review. See each skill's `references/verification-agent.md`.
+
### Output Structure
**startup-design** generates files in subdirectories:
diff --git a/startup-competitors/SKILL.md b/startup-competitors/SKILL.md
index bf4638c..a4b3c8e 100644
--- a/startup-competitors/SKILL.md
+++ b/startup-competitors/SKILL.md
@@ -67,6 +67,24 @@ Create `{project-name}/PROGRESS.md` with: project name, skill name (`startup-com
---
+## Phase 1.5: Research Depth Assessment
+
+After intake, assess market complexity and present the Research Depth recommendation to the user.
+
+> **Reference:** Read `references/research-scaling.md` for the complexity scoring matrix, tier definitions, wave configurations, and the user communication template.
+
+### Process
+
+1. Score three factors from the intake: market breadth (1-3), known competitors (1-3), geographic scope (1-3)
+2. Sum the scores (range 3-9) and map to a tier: Light (3-4), Standard (5-7), Deep (8-9)
+3. Present the Research Depth table to the user (see `research-scaling.md` for the exact template)
+4. Wait for user response: **light**, **deep**, or **ok** to accept the recommendation
+5. Record the selected tier in PROGRESS.md
+
+The selected tier determines the number of agents per wave and search rounds per agent in Phase 2. See `research-scaling.md` for exact wave configurations per tier.
+
+---
+
## Phase 2: Research
Three parallel research waves, each attacking the competitive landscape from a different angle. Together they produce a 360-degree view.
@@ -188,6 +206,24 @@ Keep raw research files in `{project-name}/raw/` for reference:
---
+## Phase 3.5: Research Verification
+
+After synthesis completes and all deliverable files are written, run a verification pass.
+
+> **Reference:** Read `references/verification-agent.md` for the full verification protocol, universal checks, and skill-specific checks.
+
+### Process
+
+1. Spawn agent **V1: Verification** — it reads all deliverable files and checks for: unlabeled claims, internal contradictions, confidence rating consistency, missing data gaps, missing flags, stale data, and duplicate-source false corroboration
+2. V1 also runs startup-competitors-specific checks: battle card vs. report consistency, matrix vs. profiles alignment, pricing landscape vs. profiles consistency, cross-deliverable coherence
+3. V1 produces `{project-name}/verification-report.md`
+4. **If Critical issues found:** Pause and present issues to the user. Ask: fix first, or proceed as-is?
+5. **If only Warnings/Info:** Show one-line summary
+
+In Claude.ai or when Agent tool is unavailable, run the verification checks yourself in the main conversation following the same protocol.
+
+---
+
## Honesty Protocol
> **Reference:** Read `references/honesty-protocol.md` for full protocol and anti-pattern details.
@@ -218,3 +254,5 @@ Read only what you need for the current phase.
| `research-wave-2-sentiment-mining.md` | When running Wave 2 | ~189 | Agent templates for review + forum mining |
| `research-wave-3-gtm-signals.md` | When running Wave 3 | ~192 | Agent templates for GTM + strategic signals |
| `research-synthesis.md` | After all waves complete | ~231 | How to synthesize + battle card template |
+| `research-scaling.md` | After intake, before Phase 2 | ~80 | Complexity scoring, tier definitions, wave configurations |
+| `verification-agent.md` | After synthesis | ~85 | Verification protocol, universal + skill-specific checks |
diff --git a/startup-competitors/references/research-scaling.md b/startup-competitors/references/research-scaling.md
new file mode 100644
index 0000000..64d257b
--- /dev/null
+++ b/startup-competitors/references/research-scaling.md
@@ -0,0 +1,106 @@
+# Research Scaling Protocol
+
+Dynamic scaling adjusts research depth based on market complexity and user preference. Evaluated after intake, before research begins.
+
+## Complexity Score
+
+Assess three factors from the intake data:
+
+| Factor | Low (1) | Medium (2) | High (3) |
+|--------|---------|------------|----------|
+| **Market breadth** | Ultra-niche, few players, well-defined segment | Defined market, moderate competition | Broad market, many segments, diverse players |
+| **Known competitors** | 0-2 identified | 3-5 identified | 6+ identified |
+| **Geographic scope** | Single country | Regional (e.g., Europe, North America) | Global or multi-region |
+
+**Complexity score** = sum of the three factors (range: 3-9)
+
+## Research Depth Tiers
+
+| Tier | Score Range | Manual Trigger | Description |
+|------|-----------|----------------|-------------|
+| **Light** | 3-4 | User says "light", "quick", or "fast research" | Quick scan, fewer agents, 2-3 search rounds |
+| **Standard** | 5-7 | Default (no override needed) | Current behavior, balanced depth |
+| **Deep** | 8-9 | User says "deep", "thorough", or "deep research" | More agents, 5-6 search rounds, extra coverage |
+
+**Manual override always wins.** If the user requests "light" on a score-9 market, use Light. If they request "deep" on a score-3 market, use Deep.
+
+## User Communication
+
+After calculating the score, show this to the user:
+
+```
+## Research Depth
+
+Based on your intake, I've assessed the research complexity:
+
+| Factor | Assessment | Score |
+|------------------|---------------------|-------|
+| Market breadth | {description} | {1-3} |
+| Known competitors| {N} identified | {1-3} |
+| Geographic scope | {description} | {1-3} |
+
+**Complexity score: {X}/9 — recommended depth: {Light/Standard/Deep}**
+
+You can override this. Here's what each depth means:
+
+| Depth | Agents | Searches per agent | Best for |
+|--------------|--------|--------------------|-----------------------------------------------|
+| **Light** | {N} | 2-3 rounds | Quick scan, niche markets, time-sensitive decisions |
+| **Standard** | {N} | 3-4 rounds | Most cases, balanced depth vs. speed |
+| **Deep** | {N} | 5-6 rounds | Crowded markets, high-stakes decisions, thorough due diligence |
+
+→ Type **light**, **deep**, or **ok** to accept the recommendation.
+```
+
+The agent counts shown should reflect the actual numbers for this skill (see Wave Configuration below).
+
+## Wave Configuration: startup-competitors
+
+### Light (3-4 score or user override)
+
+**Wave 1: Competitor Profiles + Pricing** (1 agent)
+- A1: Competitor Profiles & Pricing (merge A1+A2 into one agent, cover profiles and pricing together)
+
+**Wave 2: Customer Sentiment** (1 agent)
+- B1: Review & Community Mining (merge B1+B2 into one agent, cover reviews and forums together)
+
+**Wave 3: GTM & Strategic Signals** (1 agent)
+- C1: GTM & Growth Signals (merge C1+C2 into one agent, cover GTM and strategic signals together)
+
+**Total: 3 agents** (vs. 6 Standard), 2-3 search rounds per agent
+
+### Standard (5-7 score, default)
+
+No changes to current wave structure:
+- Wave 1: 2 agents (A1, A2)
+- Wave 2: 2 agents (B1, B2)
+- Wave 3: 2 agents (C1, C2)
+
+**Total: 6 agents**, 3-4 search rounds per agent
+
+### Deep (8-9 score or user override)
+
+**Wave 1: Competitor Profiles + Pricing** (3 agents)
+- A1: Competitor Deep-Dives (unchanged)
+- A2: Pricing Intelligence (unchanged)
+- A3: Adjacent Competitor Profiles (NEW: profile 3-5 adjacent/emerging competitors not in the direct set, including recent launches and stealth startups)
+
+**Wave 2: Customer Sentiment** (3 agents)
+- B1: Review Mining (unchanged)
+- B2: Forum & Community Mining (unchanged)
+- B3: Social Media Sentiment (NEW: mine Twitter/X, LinkedIn, and YouTube for competitor mentions, sentiment patterns, and influencer opinions)
+
+**Wave 3: GTM & Strategic Signals** (3 agents)
+- C1: Go-to-Market Analysis (unchanged)
+- C2: Strategic & Growth Signals (unchanged)
+- C3: Tech Stack & Product Analysis (NEW: analyze competitors' technology choices, API ecosystems, integration depth, and technical moats)
+
+**Total: 9 agents**, 5-6 search rounds per agent
+
+## PROGRESS.md
+
+Record the selected tier in PROGRESS.md:
+
+```markdown
+- **Research Depth:** {Light/Standard/Deep} (score: {X}/9, {override: user request / auto})
+```
diff --git a/startup-competitors/references/research-synthesis.md b/startup-competitors/references/research-synthesis.md
index 0838901..a5640d1 100644
--- a/startup-competitors/references/research-synthesis.md
+++ b/startup-competitors/references/research-synthesis.md
@@ -179,6 +179,12 @@ Based on the matrix, the clearest paths to differentiation:
---
+## Post-Synthesis Verification
+
+After writing all deliverables and battle cards, run the Verification Agent protocol. See `references/verification-agent.md` for the full process. The verification step checks all deliverables for unlabeled claims, internal contradictions, confidence rating consistency, and startup-competitors-specific coherence (battle card vs. report consistency, matrix vs. profiles alignment, pricing landscape vs. profiles consistency, cross-deliverable opportunity/risk traceability).
+
+---
+
## Output File: battle-cards/{competitor-name}.md
One battle card per competitor. Keep each to ONE page — these are reference tools for quick use, not deep research docs.
diff --git a/startup-competitors/references/verification-agent.md b/startup-competitors/references/verification-agent.md
new file mode 100644
index 0000000..2e03107
--- /dev/null
+++ b/startup-competitors/references/verification-agent.md
@@ -0,0 +1,126 @@
+# Verification Agent Protocol
+
+After synthesis completes, spawn a **Verification Agent (V1)** that audits all deliverables for consistency, accuracy, and completeness. This step catches issues that individual agents and synthesis can miss.
+
+## When to Run
+
+- After synthesis is complete and all deliverable files are written
+- Before presenting the final output to the user
+- Uses one agent: **V1: Verification**
+
+## Agent Task
+
+The V1 agent reads ALL deliverable files (not raw files) and checks them against the rules below. It produces a `verification-report.md` in the project directory.
+
+## Universal Checks
+
+These apply to every skill in the startup plugin:
+
+### 1. Claims Without Source
+Every quantitative claim must have a data label: **[Data]**, **[Estimate]**, **[Assumption]**, or **[Opinion]**. Flag any number, percentage, or factual assertion without a label.
+
+### 2. Internal Contradictions
+Cross-check numbers and statements across deliverable files. Flag when:
+- The same metric appears with different values in two files
+- A claim in one file contradicts a claim in another
+- Confidence ratings disagree (e.g., "High confidence" in one file, different data in another suggests Medium)
+
+### 3. Confidence Rating Consistency
+Verify that confidence ratings match the evidence:
+- A claim with only one Tier 3 source cannot be rated **High**
+- A claim with multiple Tier 1 sources should not be rated **Low**
+- Every major section must have a confidence rating
+
+### 4. Data Gaps Declared
+Every deliverable must have a Data Gaps section. Flag:
+- Files missing the Data Gaps section entirely
+- Sections where data is clearly thin but no gap is declared
+- Gaps mentioned in raw files that didn't make it into the synthesized deliverables
+
+### 5. Flags Present
+Every deliverable must end with Red Flags and Yellow Flags sections. Flag:
+- Files missing these sections
+- Files with "No flags identified" where the content clearly contains risks
+
+### 6. Stale Data
+Flag any data point older than 18 months that isn't marked as potentially outdated.
+
+### 7. Duplicate Sources
+Flag when the same source is used as "independent corroboration" in multiple places. Two claims both citing the same blog post don't have independent verification.
+
+## Skill-Specific Checks: startup-competitors
+
+In addition to the universal checks above, verify:
+
+### Battle Card vs. Report Consistency
+- Every competitor in the battle cards must appear in the `competitors-report.md` Key Players table
+- Strengths/weaknesses in battle cards must not contradict the competitor profiles in the report
+- Threat levels must be consistent between battle cards and the report
+
+### Matrix vs. Profiles Alignment
+- Every competitor in `competitive-matrix.md` must have a profile in `competitors-report.md`
+- Feature ratings (Strong/Adequate/Weak/Missing) in the matrix must be supported by evidence in the report or battle cards
+- Gap analysis items must connect to actual findings, not assumptions
+
+### Pricing Landscape vs. Profiles Consistency
+- Pricing data in `pricing-landscape.md` must match pricing mentioned in battle cards
+- Value metrics must be consistent across the pricing landscape and competitor profiles
+- Switching cost assessments must align between pricing landscape and battle cards
+
+### Cross-Deliverable Coherence
+- Strategic opportunities in the report must be supported by evidence from at least two deliverables (e.g., pricing gap + customer complaint)
+- Strategic risks must be traceable to specific competitor data
+- Moat assessment must reference specific competitors and evidence
+
+## Output: verification-report.md
+
+```markdown
+# Verification Report: {project-name}
+*Generated: {date}*
+
+## Summary
+- **Critical issues:** {count}
+- **Warnings:** {count}
+- **Info:** {count}
+
+## Critical Issues
+Issues that could mislead decision-making. The process pauses here for user review.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Section:** {section name}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Warnings
+Issues that reduce quality but don't block decisions.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Info
+Minor improvements and observations.
+
+- {observation}
+- {observation}
+
+## Verification Checklist
+- [ ] All quantitative claims labeled
+- [ ] No internal contradictions found
+- [ ] Confidence ratings consistent with evidence
+- [ ] Data gaps declared in all deliverables
+- [ ] Red/Yellow flags present in all deliverables
+- [ ] No stale data unmarked
+- [ ] No duplicate-source false corroboration
+- [ ] Battle cards consistent with report (skill-specific)
+- [ ] Matrix aligned with profiles (skill-specific)
+- [ ] Pricing landscape consistent with profiles (skill-specific)
+- [ ] Opportunities backed by multi-source evidence (skill-specific)
+```
+
+## Flow Control
+
+- **If Critical issues > 0:** Pause. Show the user: "Verification found {N} critical issues that could affect decision-making." List them. Ask: "Should I fix these before continuing, or proceed as-is?"
+- **If only Warnings/Info:** Show a one-line summary: "Verification complete: {N} warnings, {N} info items. See `verification-report.md` for details." Continue.
diff --git a/startup-design/SKILL.md b/startup-design/SKILL.md
index f374118..c891b41 100644
--- a/startup-design/SKILL.md
+++ b/startup-design/SKILL.md
@@ -137,6 +137,24 @@ Save to `{project-name}/00-intake/brainstorm.md`. Update PROGRESS.md.
---
+## Phase 2.5: Research Depth Assessment
+
+After intake (and brainstorm if applicable), assess market complexity and present the Research Depth recommendation to the user.
+
+> **Reference:** Read `references/research-scaling.md` for the complexity scoring matrix, tier definitions, wave configurations, and the user communication template.
+
+### Process
+
+1. Score three factors from the intake: market breadth (1-3), known competitors (1-3), geographic scope (1-3)
+2. Sum the scores (range 3-9) and map to a tier: Light (3-4), Standard (5-7), Deep (8-9)
+3. Present the Research Depth table to the user (see `research-scaling.md` for the exact template)
+4. Wait for user response: **light**, **deep**, or **ok** to accept the recommendation
+5. Record the selected tier in PROGRESS.md
+
+The selected tier determines the number of agents per wave and search rounds per agent in Phase 3. See `research-scaling.md` for exact wave configurations per tier.
+
+---
+
## Phase 3: Market Research
This is the most resource-intensive phase. It uses 4 sequential waves of web research, each building on the previous one's findings.
@@ -219,6 +237,24 @@ Update PROGRESS.md.
---
+## Phase 3.5a: Research Verification
+
+After synthesis completes and all deliverable files are written, run a verification pass to catch inconsistencies.
+
+> **Reference:** Read `references/verification-agent.md` for the full verification protocol, universal checks, and skill-specific checks.
+
+### Process
+
+1. Spawn agent **V1: Verification** — it reads all deliverable files in `01-discovery/` and checks for: unlabeled claims, internal contradictions, confidence rating consistency, missing data gaps, missing flags, stale data, and duplicate-source false corroboration
+2. V1 also runs startup-design-specific checks: cross-phase consistency (strategy reflects market data, product reflects customer pains, financial reflects business model, validation covers identified risks)
+3. V1 produces `{project-name}/01-discovery/verification-report.md`
+4. **If Critical issues found:** Pause and present issues to the user. Ask: fix first, or proceed as-is?
+5. **If only Warnings/Info:** Show one-line summary and continue to Research Gate
+
+In Claude.ai or when Agent tool is unavailable, run the verification checks yourself in the main conversation following the same protocol.
+
+---
+
## Phase 3.5: Research Gate (Go/No-Go Checkpoint)
Before investing time in Strategy through Validation, pause and present the founder with an honest assessment based on research findings. This is a decision point, not a formality.
@@ -520,5 +556,7 @@ The `references/` directory contains supporting documentation. Read only what yo
| `research-wave-3-customers.md` | When spawning Wave 3 agents | ~233 |
| `research-wave-4-distribution.md` | When spawning Wave 4 agents | ~132 |
| `research-synthesis.md` | After all waves complete, before writing final files | ~104 |
+| `research-scaling.md` | After intake, before Phase 3 | ~95 | Complexity scoring, tier definitions, wave configurations |
+| `verification-agent.md` | After synthesis, before Phase 3.5 | ~100 | Verification protocol, universal + skill-specific checks |
| `frameworks.md` | During Phase 4 (Strategy), Phase 6 (Product), and Phase 8 (Validation) | ~110 |
| `industry-benchmarks.md` | During Phase 7 (Financial) | ~80 |
diff --git a/startup-design/references/research-scaling.md b/startup-design/references/research-scaling.md
new file mode 100644
index 0000000..0a495ad
--- /dev/null
+++ b/startup-design/references/research-scaling.md
@@ -0,0 +1,122 @@
+# Research Scaling Protocol
+
+Dynamic scaling adjusts research depth based on market complexity and user preference. Evaluated after intake, before research begins.
+
+## Complexity Score
+
+Assess three factors from the intake data:
+
+| Factor | Low (1) | Medium (2) | High (3) |
+|--------|---------|------------|----------|
+| **Market breadth** | Ultra-niche, few players, well-defined segment | Defined market, moderate competition | Broad market, many segments, diverse players |
+| **Known competitors** | 0-2 identified | 3-5 identified | 6+ identified |
+| **Geographic scope** | Single country | Regional (e.g., Europe, North America) | Global or multi-region |
+
+**Complexity score** = sum of the three factors (range: 3-9)
+
+## Research Depth Tiers
+
+| Tier | Score Range | Manual Trigger | Description |
+|------|-----------|----------------|-------------|
+| **Light** | 3-4 | User says "light", "quick", or "fast research" | Quick scan, fewer agents, 2-3 search rounds |
+| **Standard** | 5-7 | Default (no override needed) | Current behavior, balanced depth |
+| **Deep** | 8-9 | User says "deep", "thorough", or "deep research" | More agents, 5-6 search rounds, extra coverage |
+
+**Manual override always wins.** If the user requests "light" on a score-9 market, use Light. If they request "deep" on a score-3 market, use Deep.
+
+## User Communication
+
+After calculating the score, show this to the user:
+
+```
+## Research Depth
+
+Based on your intake, I've assessed the research complexity:
+
+| Factor | Assessment | Score |
+|------------------|---------------------|-------|
+| Market breadth | {description} | {1-3} |
+| Known competitors| {N} identified | {1-3} |
+| Geographic scope | {description} | {1-3} |
+
+**Complexity score: {X}/9 — recommended depth: {Light/Standard/Deep}**
+
+You can override this. Here's what each depth means:
+
+| Depth | Agents | Searches per agent | Best for |
+|--------------|--------|--------------------|-----------------------------------------------|
+| **Light** | {N} | 2-3 rounds | Quick scan, niche markets, time-sensitive decisions |
+| **Standard** | {N} | 3-4 rounds | Most cases, balanced depth vs. speed |
+| **Deep** | {N} | 5-6 rounds | Crowded markets, high-stakes decisions, thorough due diligence |
+
+→ Type **light**, **deep**, or **ok** to accept the recommendation.
+```
+
+The agent counts shown should reflect the actual numbers for this skill (see Wave Configuration below).
+
+## Wave Configuration: startup-design
+
+### Light (3-4 score or user override)
+
+**Wave 1: Market Landscape** (2 agents)
+- A1: Market Sizing & Economics (unchanged)
+- A2: Industry Trends, Timing & Regulatory (merge A2+A3 into one agent)
+
+**Wave 2: Competitive Analysis** (2 agents)
+- B1: Direct Competitor Deep-Dives (unchanged)
+- B2: Indirect Competitors & GTM (merge B2+B3 into one agent)
+
+**Wave 3: Customer & Demand** (2 agents)
+- C1: Customer Voice & Pain Points (unchanged)
+- C2: Demand Signals & Audience Profiling (merge C2+C3 into one agent)
+
+**Wave 4: Distribution** (1 agent)
+- D1: Distribution & Geographic Entry (merge D1+D2 into one agent)
+- Skip Wave 4 entirely if market is local/single-country
+
+**Total: 6-7 agents** (vs. 11 Standard), 2-3 search rounds per agent
+
+### Standard (5-7 score, default)
+
+No changes to current wave structure:
+- Wave 1: 3 agents (A1, A2, A3)
+- Wave 2: 3 agents (B1, B2, B3)
+- Wave 3: 3 agents (C1, C2, C3)
+- Wave 4: 2 agents (D1, D2)
+
+**Total: 11 agents**, 3-4 search rounds per agent
+
+### Deep (8-9 score or user override)
+
+**Wave 1: Market Landscape** (4 agents)
+- A1: Market Sizing & Economics (unchanged)
+- A2: Industry Trends & Timing (unchanged)
+- A3: Regulatory & Compliance (unchanged)
+- A4: Adjacent Markets & Expansion (NEW: research adjacent verticals, whitespace, expansion paths)
+
+**Wave 2: Competitive Analysis** (4 agents)
+- B1: Direct Competitor Deep-Dives (unchanged)
+- B2: Indirect Competitors & Substitutes (unchanged)
+- B3: Competitor Go-to-Market (unchanged)
+- B4: Emerging & Stealth Competitors (NEW: search for stealth startups, recent launches, pre-launch products in the space)
+
+**Wave 3: Customer & Demand** (4 agents)
+- C1: Customer Voice & Pain Points (unchanged)
+- C2: Demand Signals & Market Validation (unchanged)
+- C3: Target Audience Profiling (unchanged)
+- C4: Pricing Deep-Dive (NEW: detailed pricing intelligence, WTP analysis, price sensitivity research)
+
+**Wave 4: Distribution & Partnerships** (3 agents)
+- D1: Distribution Channel Deep-Dive (unchanged)
+- D2: Geographic & Market Entry (unchanged)
+- D3: Strategic Partnership Mapping (NEW: potential partners, integration opportunities, channel partnerships)
+
+**Total: 15 agents**, 5-6 search rounds per agent
+
+## PROGRESS.md
+
+Record the selected tier in PROGRESS.md:
+
+```markdown
+- **Research Depth:** {Light/Standard/Deep} (score: {X}/9, {override: user request / auto})
+```
diff --git a/startup-design/references/research-synthesis.md b/startup-design/references/research-synthesis.md
index 601ae5a..3adec83 100644
--- a/startup-design/references/research-synthesis.md
+++ b/startup-design/references/research-synthesis.md
@@ -99,6 +99,10 @@ After writing all 4 files, add a "Strategic Connections" section at the end of e
These connections help Phase 4 (Strategy) build on concrete research rather than abstract claims.
+## Post-Synthesis Verification
+
+After writing all deliverables, run the Verification Agent protocol. See `references/verification-agent.md` for the full process. The verification step checks all deliverables for unlabeled claims, internal contradictions, confidence rating consistency, and startup-design-specific cross-phase coherence (strategy reflects market data, product reflects customer pains, financial reflects business model, validation covers identified risks).
+
## Cleanup
Keep the `raw/` directory for reference. Note in PROGRESS.md that raw research files are available at `01-discovery/raw/`.
diff --git a/startup-design/references/verification-agent.md b/startup-design/references/verification-agent.md
new file mode 100644
index 0000000..d58da12
--- /dev/null
+++ b/startup-design/references/verification-agent.md
@@ -0,0 +1,117 @@
+# Verification Agent Protocol
+
+After synthesis completes, spawn a **Verification Agent (V1)** that audits all deliverables for consistency, accuracy, and completeness. This step catches issues that individual agents and synthesis can miss.
+
+## When to Run
+
+- After synthesis is complete and all deliverable files are written
+- Before the next phase begins (e.g., before Phase 3.5 Research Gate in startup-design)
+- Uses one agent: **V1: Verification**
+
+## Agent Task
+
+The V1 agent reads ALL deliverable files (not raw files) and checks them against the rules below. It produces a `verification-report.md` in the project directory.
+
+## Universal Checks
+
+These apply to every skill in the startup plugin:
+
+### 1. Claims Without Source
+Every quantitative claim must have a data label: **[Data]**, **[Estimate]**, **[Assumption]**, or **[Opinion]**. Flag any number, percentage, or factual assertion without a label.
+
+### 2. Internal Contradictions
+Cross-check numbers and statements across deliverable files. Flag when:
+- The same metric appears with different values in two files
+- A claim in one file contradicts a claim in another
+- Confidence ratings disagree (e.g., "High confidence" in one file, different data in another suggests Medium)
+
+### 3. Confidence Rating Consistency
+Verify that confidence ratings match the evidence:
+- A claim with only one Tier 3 source cannot be rated **High**
+- A claim with multiple Tier 1 sources should not be rated **Low**
+- Every major section must have a confidence rating
+
+### 4. Data Gaps Declared
+Every deliverable must have a Data Gaps section. Flag:
+- Files missing the Data Gaps section entirely
+- Sections where data is clearly thin but no gap is declared
+- Gaps mentioned in raw files that didn't make it into the synthesized deliverables
+
+### 5. Flags Present
+Every deliverable must end with Red Flags and Yellow Flags sections. Flag:
+- Files missing these sections
+- Files with "No flags identified" where the content clearly contains risks
+
+### 6. Stale Data
+Flag any data point older than 18 months that isn't marked as potentially outdated.
+
+### 7. Duplicate Sources
+Flag when the same source is used as "independent corroboration" in multiple places. Two claims both citing the same blog post don't have independent verification.
+
+## Skill-Specific Checks: startup-design
+
+In addition to the universal checks above, verify:
+
+### Cross-Phase Consistency
+- **Strategy reflects market data:** Do the Lean Canvas, positioning, and GTM in Phase 4 reference specific findings from Phase 3? Flag strategy claims that have no research backing.
+- **Product reflects customer pains:** Does the MVP definition (Phase 6) address the top pains identified in `target-audience.md`? Flag features that don't map to any validated customer need.
+- **Financial reflects business model:** Do the revenue projections (Phase 7) use the pricing and unit economics from `market-analysis.md` and `business-model.md`? Flag projection assumptions that contradict research data.
+- **Validation covers key risks:** Does the risk analysis (Phase 8) address the Red Flags and data gaps surfaced in earlier phases? Flag risks that were identified in research but missing from validation.
+
+### Numerical Consistency
+- TAM/SAM/SOM numbers must be consistent across `market-analysis.md`, `lean-canvas.md`, and `revenue-model.md`
+- Pricing references must match between `competitor-landscape.md` and `business-model.md`
+- Customer segment descriptions must be consistent between `target-audience.md`, `lean-canvas.md`, and `mvp-definition.md`
+
+## Output: verification-report.md
+
+```markdown
+# Verification Report: {project-name}
+*Generated: {date}*
+
+## Summary
+- **Critical issues:** {count}
+- **Warnings:** {count}
+- **Info:** {count}
+
+## Critical Issues
+Issues that could mislead decision-making. The process pauses here for user review.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Section:** {section name}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Warnings
+Issues that reduce quality but don't block decisions.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Info
+Minor improvements and observations.
+
+- {observation}
+- {observation}
+
+## Verification Checklist
+- [ ] All quantitative claims labeled
+- [ ] No internal contradictions found
+- [ ] Confidence ratings consistent with evidence
+- [ ] Data gaps declared in all deliverables
+- [ ] Red/Yellow flags present in all deliverables
+- [ ] No stale data unmarked
+- [ ] No duplicate-source false corroboration
+- [ ] Strategy reflects market data (cross-phase)
+- [ ] Product reflects customer pains (cross-phase)
+- [ ] Financial reflects business model (cross-phase)
+- [ ] Validation covers identified risks (cross-phase)
+```
+
+## Flow Control
+
+- **If Critical issues > 0:** Pause. Show the user: "Verification found {N} critical issues that could affect decision-making." List them. Ask: "Should I fix these before continuing, or proceed as-is?"
+- **If only Warnings/Info:** Show a one-line summary: "Verification complete: {N} warnings, {N} info items. See `verification-report.md` for details." Continue to next phase.
diff --git a/startup-pitch/SKILL.md b/startup-pitch/SKILL.md
index 16e8ea4..3ed4d1d 100644
--- a/startup-pitch/SKILL.md
+++ b/startup-pitch/SKILL.md
@@ -107,6 +107,24 @@ Create `{project-name}/PROGRESS.md` with: project name, skill name (`startup-pit
---
+## Phase 1.5: Research Depth Assessment
+
+After intake, assess market complexity and present the Research Depth recommendation to the user.
+
+> **Reference:** Read `references/research-scaling.md` for the complexity scoring matrix, tier definitions, wave configurations, and the user communication template.
+
+### Process
+
+1. Score three factors from the intake: market breadth (1-3), known competitors (1-3), geographic scope (1-3)
+2. Sum the scores (range 3-9) and map to a tier: Light (3-4), Standard (5-7), Deep (8-9)
+3. Present the Research Depth table to the user (see `research-scaling.md` for the exact template)
+4. Wait for user response: **light**, **deep**, or **ok** to accept the recommendation
+5. Record the selected tier in PROGRESS.md
+
+The selected tier determines the number of agents per wave and search rounds per agent in Phase 2. See `research-scaling.md` for exact wave configurations per tier.
+
+---
+
## Phase 2: Research
Two parallel research waves exploring investor audience and competitive/market framing. Together they provide the raw material for a pitch that resonates with the target audience.
@@ -256,6 +274,24 @@ Raw research files:
---
+## Phase 3.5: Pitch Verification
+
+After all pitch deliverables are written, before scoring and review, run a verification pass.
+
+> **Reference:** Read `references/verification-agent.md` for the full verification protocol, universal checks, and skill-specific checks.
+
+### Process
+
+1. Spawn agent **V1: Verification** — it reads all deliverable files and checks for: unlabeled claims, internal contradictions, confidence rating consistency, missing data gaps, missing flags, stale data, and duplicate-source false corroboration
+2. V1 also runs startup-pitch-specific checks: pitch claims vs. source data, cross-format consistency, pitch vs. appendix alignment, honesty checks (traction without timeframes, top-down TAM, titles without accomplishments)
+3. V1 produces `{project-name}/verification-report.md`
+4. **If Critical issues found:** Pause and present issues to the user. Ask: fix first, or proceed as-is?
+5. **If only Warnings/Info:** Show one-line summary and continue to review phase
+
+In Claude.ai or when Agent tool is unavailable, run the verification checks yourself in the main conversation following the same protocol.
+
+---
+
## Phase 4: Review & Practice
After generating the pitch deliverables, review quality and optionally practice with investor roleplay.
@@ -368,3 +404,5 @@ Read only what you need for the current phase.
| `research-wave-2-competitive-framing.md` | When running Wave 2 | ~159 | Agent templates for competitive framing + why now |
| `pitch-frameworks.md` | During Phase 3 | ~261 | Complete pitch framework with integration matrix |
| `research-synthesis.md` | After waves complete | ~417 | Synthesis protocol and output templates |
+| `research-scaling.md` | After intake, before Phase 2 | ~75 | Complexity scoring, tier definitions, wave configurations |
+| `verification-agent.md` | After pitch construction | ~80 | Verification protocol, universal + skill-specific checks |
diff --git a/startup-pitch/references/research-scaling.md b/startup-pitch/references/research-scaling.md
new file mode 100644
index 0000000..88c64a4
--- /dev/null
+++ b/startup-pitch/references/research-scaling.md
@@ -0,0 +1,96 @@
+# Research Scaling Protocol
+
+Dynamic scaling adjusts research depth based on market complexity and user preference. Evaluated after intake, before research begins.
+
+## Complexity Score
+
+Assess three factors from the intake data:
+
+| Factor | Low (1) | Medium (2) | High (3) |
+|--------|---------|------------|----------|
+| **Market breadth** | Ultra-niche, few players, well-defined segment | Defined market, moderate competition | Broad market, many segments, diverse players |
+| **Known competitors** | 0-2 identified | 3-5 identified | 6+ identified |
+| **Geographic scope** | Single country | Regional (e.g., Europe, North America) | Global or multi-region |
+
+**Complexity score** = sum of the three factors (range: 3-9)
+
+## Research Depth Tiers
+
+| Tier | Score Range | Manual Trigger | Description |
+|------|-----------|----------------|-------------|
+| **Light** | 3-4 | User says "light", "quick", or "fast research" | Quick scan, fewer agents, 2-3 search rounds |
+| **Standard** | 5-7 | Default (no override needed) | Current behavior, balanced depth |
+| **Deep** | 8-9 | User says "deep", "thorough", or "deep research" | More agents, 5-6 search rounds, extra coverage |
+
+**Manual override always wins.** If the user requests "light" on a score-9 market, use Light. If they request "deep" on a score-3 market, use Deep.
+
+## User Communication
+
+After calculating the score, show this to the user:
+
+```
+## Research Depth
+
+Based on your intake, I've assessed the research complexity:
+
+| Factor | Assessment | Score |
+|------------------|---------------------|-------|
+| Market breadth | {description} | {1-3} |
+| Known competitors| {N} identified | {1-3} |
+| Geographic scope | {description} | {1-3} |
+
+**Complexity score: {X}/9 — recommended depth: {Light/Standard/Deep}**
+
+You can override this. Here's what each depth means:
+
+| Depth | Agents | Searches per agent | Best for |
+|--------------|--------|--------------------|-----------------------------------------------|
+| **Light** | {N} | 2-3 rounds | Quick scan, niche markets, time-sensitive decisions |
+| **Standard** | {N} | 3-4 rounds | Most cases, balanced depth vs. speed |
+| **Deep** | {N} | 5-6 rounds | Crowded markets, high-stakes decisions, thorough due diligence |
+
+→ Type **light**, **deep**, or **ok** to accept the recommendation.
+```
+
+The agent counts shown should reflect the actual numbers for this skill (see Wave Configuration below).
+
+## Wave Configuration: startup-pitch
+
+### Light (3-4 score or user override)
+
+**Wave 1** (1 wave, 2 agents)
+- A1: Audience, Narrative & Competitive Framing (merge A1+A2+B1 into one agent)
+- A2: Why Now & Timing (B2, unchanged scope)
+
+Merge both waves into a single wave since there's no dependency between them at this scale.
+
+**Total: 2 agents** (vs. 4 Standard), 2-3 search rounds per agent
+
+### Standard (5-7 score, default)
+
+No changes to current wave structure:
+- Wave 1: 2 agents (A1, A2)
+- Wave 2: 2 agents (B1, B2)
+
+**Total: 4 agents**, 3-4 search rounds per agent
+
+### Deep (8-9 score or user override)
+
+**Wave 1: Audience & Narrative** (3 agents)
+- A1: Investor & Audience Intelligence (unchanged)
+- A2: Comparable & Narrative Research (unchanged)
+- A3: Pitch Benchmark Analysis (NEW: research how similar startups in this space pitched, what decks worked, what narratives raised money, typical round sizes and valuations)
+
+**Wave 2: Competitive Framing & Why Now** (2 agents)
+- B1: Competitive Framing for Pitch (unchanged)
+- B2: Why Now & Market Timing (unchanged)
+
+**Total: 5 agents**, 5-6 search rounds per agent
+
+## PROGRESS.md
+
+Record the selected tier in PROGRESS.md:
+
+```markdown
+- **Research Depth:** {Light/Standard/Deep} (score: {X}/9, {override: user request / auto})
+```
diff --git a/startup-pitch/references/research-synthesis.md b/startup-pitch/references/research-synthesis.md
index d4e6ae8..1e80953 100644
--- a/startup-pitch/references/research-synthesis.md
+++ b/startup-pitch/references/research-synthesis.md
@@ -365,6 +365,12 @@ From the full narrative, derive each format by compression — not by rewriting
- {all data sources used across all deliverables, with tier ratings}
```
+## Post-Synthesis Verification
+
+After writing all pitch deliverables (before the scorecard/review phase), run the Verification Agent protocol. See `references/verification-agent.md` for the full process. The verification step checks all deliverables for unlabeled claims, internal contradictions, confidence rating consistency, and startup-pitch-specific coherence (pitch claims vs. source data, cross-format consistency, pitch vs. appendix alignment, honesty checks).
+
+---
+
### pitch-scorecard.md
```markdown
diff --git a/startup-pitch/references/verification-agent.md b/startup-pitch/references/verification-agent.md
new file mode 100644
index 0000000..73e63e0
--- /dev/null
+++ b/startup-pitch/references/verification-agent.md
@@ -0,0 +1,129 @@
+# Verification Agent Protocol
+
+After pitch construction completes, spawn a **Verification Agent (V1)** that audits all deliverables for consistency, accuracy, and completeness. This step catches issues that individual agents and synthesis can miss.
+
+## When to Run
+
+- After all pitch deliverables are written (before the scorecard/review phase)
+- Uses one agent: **V1: Verification**
+
+## Agent Task
+
+The V1 agent reads ALL deliverable files (not raw files) and checks them against the rules below. It produces a `verification-report.md` in the project directory.
+
+## Universal Checks
+
+These apply to every skill in the startup plugin:
+
+### 1. Claims Without Source
+Every quantitative claim must have a data label: **[Data]**, **[Estimate]**, **[Assumption]**, or **[Opinion]**. Flag any number, percentage, or factual assertion without a label.
+
+### 2. Internal Contradictions
+Cross-check numbers and statements across deliverable files. Flag when:
+- The same metric appears with different values in two files
+- A claim in one file contradicts a claim in another
+- Confidence ratings disagree (e.g., "High confidence" in one file, different data in another suggests Medium)
+
+### 3. Confidence Rating Consistency
+Verify that confidence ratings match the evidence:
+- A claim with only one Tier 3 source cannot be rated **High**
+- A claim with multiple Tier 1 sources should not be rated **Low**
+- Every major section must have a confidence rating
+
+### 4. Data Gaps Declared
+Every deliverable must have a Data Gaps section. Flag:
+- Files missing the Data Gaps section entirely
+- Sections where data is clearly thin but no gap is declared
+- Gaps mentioned in raw files that didn't make it into the synthesized deliverables
+
+### 5. Flags Present
+Every deliverable must end with Red Flags and Yellow Flags sections. Flag:
+- Files missing these sections
+- Files with "No flags identified" where the content clearly contains risks
+
+### 6. Stale Data
+Flag any data point older than 18 months that isn't marked as potentially outdated.
+
+### 7. Duplicate Sources
+Flag when the same source is used as "independent corroboration" in multiple places. Two claims both citing the same blog post don't have independent verification.
+
+## Skill-Specific Checks: startup-pitch
+
+In addition to the universal checks above, verify:
+
+### Pitch Claims vs. Source Data
+- Every numerical claim in the pitch narratives (traction, market size, revenue, growth rate) must be traceable to either intake data or prior session files
+- If the pitch was built on startup-design data, verify that market size numbers match `market-analysis.md`
+- If built on startup-competitors data, verify competitive claims match `competitors-report.md`
+- Flag any number in the pitch that doesn't appear in any source file
+
+### Cross-Format Consistency
+- The 2-sentence opener must be identical (or a faithful compression) across all pitch formats
+- Market size, traction numbers, and the ask must be consistent across pitch-full, pitch-5min, pitch-2min, pitch-1min, and pitch-email
+- The unique insight must be the same concept across all formats (phrased differently but same core idea)
+- Team credentials must not vary between formats (no inflated version in one, modest in another)
+
+### Pitch vs. Appendix Alignment
+- Every objection in `pitch-appendix.md` must have a corresponding answer
+- Known weaknesses in the appendix must match the Red/Yellow Flags in pitch files
+- Q&A answers must not contradict claims made in the pitch narratives
+- If competitive backup references battle cards, verify the competitor names and claims match
+
+### Honesty Checks
+- Flag any traction claim without a timeframe ("10K users" without "in X months")
+- Flag top-down TAM without bottom-up math
+- Flag team credentials that are titles without accomplishments
+- Flag "no competition" or equivalent claims
+
+## Output: verification-report.md
+
+```markdown
+# Verification Report: {project-name}
+*Generated: {date}*
+
+## Summary
+- **Critical issues:** {count}
+- **Warnings:** {count}
+- **Info:** {count}
+
+## Critical Issues
+Issues that could mislead decision-making. The process pauses here for user review.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Section:** {section name}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Warnings
+Issues that reduce quality but don't block decisions.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Info
+Minor improvements and observations.
+
+- {observation}
+- {observation}
+
+## Verification Checklist
+- [ ] All quantitative claims labeled
+- [ ] No internal contradictions found
+- [ ] Confidence ratings consistent with evidence
+- [ ] Data gaps declared in all deliverables
+- [ ] Red/Yellow flags present in all deliverables
+- [ ] No stale data unmarked
+- [ ] No duplicate-source false corroboration
+- [ ] Pitch claims traceable to source data (skill-specific)
+- [ ] Cross-format consistency verified (skill-specific)
+- [ ] Pitch and appendix aligned (skill-specific)
+- [ ] Honesty checks passed (skill-specific)
+```
+
+## Flow Control
+
+- **If Critical issues > 0:** Pause. Show the user: "Verification found {N} critical issues that could affect decision-making." List them. Ask: "Should I fix these before continuing, or proceed as-is?"
+- **If only Warnings/Info:** Show a one-line summary: "Verification complete: {N} warnings, {N} info items. See `verification-report.md` for details." Continue to scorecard/review phase.
diff --git a/startup-positioning/SKILL.md b/startup-positioning/SKILL.md
index a3a39cc..3402753 100644
--- a/startup-positioning/SKILL.md
+++ b/startup-positioning/SKILL.md
@@ -74,6 +74,24 @@ Create `{project-name}/PROGRESS.md` with: project name, skill name (`startup-pos
---
+## Phase 1.5: Research Depth Assessment
+
+After intake, assess market complexity and present the Research Depth recommendation to the user.
+
+> **Reference:** Read `references/research-scaling.md` for the complexity scoring matrix, tier definitions, wave configurations, and the user communication template.
+
+### Process
+
+1. Score three factors from the intake: market breadth (1-3), known competitors (1-3), geographic scope (1-3)
+2. Sum the scores (range 3-9) and map to a tier: Light (3-4), Standard (5-7), Deep (8-9)
+3. Present the Research Depth table to the user (see `research-scaling.md` for the exact template)
+4. Wait for user response: **light**, **deep**, or **ok** to accept the recommendation
+5. Record the selected tier in PROGRESS.md
+
+The selected tier determines the number of agents per wave and search rounds per agent in Phase 2. See `research-scaling.md` for exact wave configurations per tier.
+
+---
+
## Phase 2: Research
Two parallel research waves exploring competitive alternatives and market context. Together they provide the raw material for Dunford's 5+1 positioning components.
@@ -217,6 +235,24 @@ Raw research files:
---
+## Phase 3.5: Research Verification
+
+After all positioning deliverables are written, run a verification pass.
+
+> **Reference:** Read `references/verification-agent.md` for the full verification protocol, universal checks, and skill-specific checks.
+
+### Process
+
+1. Spawn agent **V1: Verification** — it reads all deliverable files and checks for: unlabeled claims, internal contradictions, confidence rating consistency, missing data gaps, missing flags, stale data, and duplicate-source false corroboration
+2. V1 also runs startup-positioning-specific checks: positioning statement vs. research data, JTBD vs. customer intelligence, cross-deliverable coherence, validation test integrity
+3. V1 produces `{project-name}/verification-report.md`
+4. **If Critical issues found:** Pause and present issues to the user. Ask: fix first, or proceed as-is?
+5. **If only Warnings/Info:** Show one-line summary
+
+In Claude.ai or when Agent tool is unavailable, run the verification checks yourself in the main conversation following the same protocol.
+
+---
+
## Honesty Protocol
> **Reference:** Read `references/honesty-protocol.md` for full protocol and anti-pattern details.
@@ -252,3 +288,5 @@ Read only what you need for the current phase.
| `research-wave-2-market-frame.md` | When running Wave 2 | ~210 | Agent templates for categories + trends |
| `research-synthesis.md` | After both waves complete | ~380 | Synthesis protocol, Dunford process, validation tests, messaging implications |
| `frameworks.md` | During Phase 3 | ~133 | Dunford/Moore/Neumeier/JTBD/Ries reference |
+| `research-scaling.md` | After intake, before Phase 2 | ~75 | Complexity scoring, tier definitions, wave configurations |
+| `verification-agent.md` | After synthesis | ~80 | Verification protocol, universal + skill-specific checks |
diff --git a/startup-positioning/references/research-scaling.md b/startup-positioning/references/research-scaling.md
new file mode 100644
index 0000000..a6bed15
--- /dev/null
+++ b/startup-positioning/references/research-scaling.md
@@ -0,0 +1,96 @@
+# Research Scaling Protocol
+
+Dynamic scaling adjusts research depth based on market complexity and user preference. Evaluated after intake, before research begins.
+
+## Complexity Score
+
+Assess three factors from the intake data:
+
+| Factor | Low (1) | Medium (2) | High (3) |
+|--------|---------|------------|----------|
+| **Market breadth** | Ultra-niche, few players, well-defined segment | Defined market, moderate competition | Broad market, many segments, diverse players |
+| **Known competitors** | 0-2 identified | 3-5 identified | 6+ identified |
+| **Geographic scope** | Single country | Regional (e.g., Europe, North America) | Global or multi-region |
+
+**Complexity score** = sum of the three factors (range: 3-9)
+
+## Research Depth Tiers
+
+| Tier | Score Range | Manual Trigger | Description |
+|------|-----------|----------------|-------------|
+| **Light** | 3-4 | User says "light", "quick", or "fast research" | Quick scan, fewer agents, 2-3 search rounds |
+| **Standard** | 5-7 | Default (no override needed) | Current behavior, balanced depth |
+| **Deep** | 8-9 | User says "deep", "thorough", or "deep research" | More agents, 5-6 search rounds, extra coverage |
+
+**Manual override always wins.** If the user requests "light" on a score-9 market, use Light. If they request "deep" on a score-3 market, use Deep.
+
+## User Communication
+
+After calculating the score, show this to the user:
+
+```
+## Research Depth
+
+Based on your intake, I've assessed the research complexity:
+
+| Factor | Assessment | Score |
+|------------------|---------------------|-------|
+| Market breadth | {description} | {1-3} |
+| Known competitors| {N} identified | {1-3} |
+| Geographic scope | {description} | {1-3} |
+
+**Complexity score: {X}/9 — recommended depth: {Light/Standard/Deep}**
+
+You can override this. Here's what each depth means:
+
+| Depth | Agents | Searches per agent | Best for |
+|--------------|--------|--------------------|-----------------------------------------------|
+| **Light** | {N} | 2-3 rounds | Quick scan, niche markets, time-sensitive decisions |
+| **Standard** | {N} | 3-4 rounds | Most cases, balanced depth vs. speed |
+| **Deep** | {N} | 5-6 rounds | Crowded markets, high-stakes decisions, thorough due diligence |
+
+→ Type **light**, **deep**, or **ok** to accept the recommendation.
+```
+
+The agent counts shown should reflect the actual numbers for this skill (see Wave Configuration below).
+
+## Wave Configuration: startup-positioning
+
+### Light (3-4 score or user override)
+
+**Wave 1** (1 wave, 2 agents)
+- A1: Alternative Mapping & Customer Intelligence (merge A1+A2 into one agent)
+- B1: Market Category & Trends (merge B1+B2 into one agent)
+
+Merge both waves into a single wave since there's no dependency between them at this scale.
+
+**Total: 2 agents** (vs. 4 Standard), 2-3 search rounds per agent
+
+### Standard (5-7 score, default)
+
+No changes to current wave structure:
+- Wave 1: 2 agents (A1, A2)
+- Wave 2: 2 agents (B1, B2)
+
+**Total: 4 agents**, 3-4 search rounds per agent
+
+### Deep (8-9 score or user override)
+
+**Wave 1: Competitive Alternatives & Customer** (3 agents)
+- A1: Alternative Mapping (unchanged)
+- A2: Customer Intelligence (unchanged)
+- A3: Positioning Perception Audit (NEW: research how competitors position themselves, what messaging they use, what category they claim, where buyer perception differs from positioning intent)
+
+**Wave 2: Market Frame & Trends** (2 agents)
+- B1: Market Category Analysis (unchanged)
+- B2: Trend & Timing Analysis (unchanged)
+
+**Total: 5 agents**, 5-6 search rounds per agent
+
+## PROGRESS.md
+
+Record the selected tier in PROGRESS.md:
+
+```markdown
+- **Research Depth:** {Light/Standard/Deep} (score: {X}/9, {override: user request / auto})
+```
diff --git a/startup-positioning/references/research-synthesis.md b/startup-positioning/references/research-synthesis.md
index e3d916a..7dd9eea 100644
--- a/startup-positioning/references/research-synthesis.md
+++ b/startup-positioning/references/research-synthesis.md
@@ -347,6 +347,12 @@ Anti-pattern: never make the free tier feel "incomplete" to push premium. The fr
- {key sources with tier ratings}
```
+## Post-Synthesis Verification
+
+After writing all deliverables, run the Verification Agent protocol. See `references/verification-agent.md` for the full process. The verification step checks all deliverables for unlabeled claims, internal contradictions, confidence rating consistency, and startup-positioning-specific coherence (positioning statement vs. research data, JTBD vs. customer intelligence, cross-deliverable messaging consistency, validation test integrity).
+
+---
+
### messaging-implications.md
```markdown
diff --git a/startup-positioning/references/verification-agent.md b/startup-positioning/references/verification-agent.md
new file mode 100644
index 0000000..a9595f0
--- /dev/null
+++ b/startup-positioning/references/verification-agent.md
@@ -0,0 +1,128 @@
+# Verification Agent Protocol
+
+After synthesis completes, spawn a **Verification Agent (V1)** that audits all deliverables for consistency, accuracy, and completeness. This step catches issues that individual agents and synthesis can miss.
+
+## When to Run
+
+- After all positioning deliverables are written
+- Before presenting the final output to the user
+- Uses one agent: **V1: Verification**
+
+## Agent Task
+
+The V1 agent reads ALL deliverable files (not raw files) and checks them against the rules below. It produces a `verification-report.md` in the project directory.
+
+## Universal Checks
+
+These apply to every skill in the startup plugin:
+
+### 1. Claims Without Source
+Every quantitative claim must have a data label: **[Data]**, **[Estimate]**, **[Assumption]**, or **[Opinion]**. Flag any number, percentage, or factual assertion without a label.
+
+### 2. Internal Contradictions
+Cross-check numbers and statements across deliverable files. Flag when:
+- The same metric appears with different values in two files
+- A claim in one file contradicts a claim in another
+- Confidence ratings disagree (e.g., "High confidence" in one file, different data in another suggests Medium)
+
+### 3. Confidence Rating Consistency
+Verify that confidence ratings match the evidence:
+- A claim with only one Tier 3 source cannot be rated **High**
+- A claim with multiple Tier 1 sources should not be rated **Low**
+- Every major section must have a confidence rating
+
+### 4. Data Gaps Declared
+Every deliverable must have a Data Gaps section. Flag:
+- Files missing the Data Gaps section entirely
+- Sections where data is clearly thin but no gap is declared
+- Gaps mentioned in raw files that didn't make it into the synthesized deliverables
+
+### 5. Flags Present
+Every deliverable must end with Red Flags and Yellow Flags sections. Flag:
+- Files missing these sections
+- Files with "No flags identified" where the content clearly contains risks
+
+### 6. Stale Data
+Flag any data point older than 18 months that isn't marked as potentially outdated.
+
+### 7. Duplicate Sources
+Flag when the same source is used as "independent corroboration" in multiple places. Two claims both citing the same blog post don't have independent verification.
+
+## Skill-Specific Checks: startup-positioning
+
+In addition to the universal checks above, verify:
+
+### Positioning Statement vs. Research
+- The Moore positioning statement must use the exact best-fit customer profile from `positioning-doc.md` Component 4
+- The category in the statement must match the recommended category from `market-category-analysis.md`
+- The differentiator must map to a verified unique attribute from Component 2
+- The Neumeier Onliness Statement must pass: does "only" genuinely hold up given the competitive alternatives in `competitive-alternatives.md`?
+
+### JTBD vs. Customer Intelligence
+- Jobs-to-be-done referenced in the positioning must have evidence in `competitive-alternatives.md`
+- Best-fit customer characteristics must connect to actual pain points from customer intelligence research
+- Value themes must trace back to real customer language from the language map
+
+### Cross-Deliverable Coherence
+- The messaging hierarchy in `messaging-implications.md` must reflect the positioning priority from `positioning-doc.md`
+- Category label must be consistent across all deliverables
+- Words to use/avoid must not contradict the language used in positioning statements
+- Freemium positioning (if present) must be consistent across `positioning-statement.md` and `messaging-implications.md`
+
+### Validation Test Results
+- If Neumeier Onliness Test is marked "Pass" but the basic statement contains qualifiers or hedging, flag it
+- If Moore Template is marked "Pass" but any field is vague or generic, flag it
+- If Mental Ladder test is marked "Pass" but the position requires more than 10 words to state, flag it
+
+## Output: verification-report.md
+
+```markdown
+# Verification Report: {project-name}
+*Generated: {date}*
+
+## Summary
+- **Critical issues:** {count}
+- **Warnings:** {count}
+- **Info:** {count}
+
+## Critical Issues
+Issues that could mislead decision-making. The process pauses here for user review.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Section:** {section name}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Warnings
+Issues that reduce quality but don't block decisions.
+
+### {Issue title}
+- **File(s):** {affected files}
+- **Problem:** {description}
+- **Suggested fix:** {how to resolve}
+
+## Info
+Minor improvements and observations.
+
+- {observation}
+- {observation}
+
+## Verification Checklist
+- [ ] All quantitative claims labeled
+- [ ] No internal contradictions found
+- [ ] Confidence ratings consistent with evidence
+- [ ] Data gaps declared in all deliverables
+- [ ] Red/Yellow flags present in all deliverables
+- [ ] No stale data unmarked
+- [ ] No duplicate-source false corroboration
+- [ ] Positioning statements match research data (skill-specific)
+- [ ] JTBD consistent with customer intelligence (skill-specific)
+- [ ] Cross-deliverable coherence verified (skill-specific)
+- [ ] Validation tests genuinely pass (skill-specific)
+```
+
+## Flow Control
+
+- **If Critical issues > 0:** Pause. Show the user: "Verification found {N} critical issues that could affect decision-making." List them. Ask: "Should I fix these before continuing, or proceed as-is?"
+- **If only Warnings/Info:** Show a one-line summary: "Verification complete: {N} warnings, {N} info items. See `verification-report.md` for details." Continue.