Deal Evaluate Skill v3.1.0

Automated dealflow evaluation pipeline with anti-bias MoE layer and True Potential Council for IASV investment decisions.

v3.1 changes:

• Two-tier output: Full Research Dossier + Executive Memo
• Research dossier (15-25 pages) serves as knowledge base for interactive Q&A
• Executive memo (5-7 pages) is the human-readable meeting document
• Q&A pulls from the dossier, not the memo — depth is preserved

v3.0 changes (Ed feedback 2026-04-30):

• Executive memo dramatically shorter — 5-7 pages max, not 15-20
• Summarize at source, eliminate redundancy between sections
• Hardware and consumer no longer auto-fail — evaluate case-by-case
• Interactive Q&A: powered by full research dossier
• Each memo section answers ONE question, no overlap

Invocation

cd /path/to/pipeline/Deals/CompanyName/
/deal-evaluate

Auto-discovers materials in the folder, extracts company name, uses web search to fill gaps.

/deal-evaluate --company "CompanyName" --website "https://company.com"
/deal-evaluate --quick-screen
/deal-evaluate --rerun-council

Minimum input: A folder with any company materials. Enhanced with: Pitch deck, meeting notes, data room materials, founder LinkedIn URLs, deal terms.

Two-Tier Output Architecture

Problem: Ed wants short reports for meetings, but interactive Q&A needs deep research to answer follow-up questions. A 5-page memo can't answer "what's their patent portfolio?" if that detail was never captured.

Solution: Generate two documents per evaluation:

The executive memo is a navigation layer into the research dossier. When Ed asks "what about competition?" in a meeting, the Q&A system surfaces the detailed competitive analysis from the dossier, not the compressed table from the memo.

Nothing is thrown away — the research is always preserved.

Executive Memo Rules (Ed's Feedback)

"It needs to be summarized more... I just don't like to read that much stuff anymore because I'll just say summarize it."

These rules apply to the executive memo only (not the research dossier):

1. No section longer than half a page. If it needs more, you're being redundant.
2. Never repeat a fact in two sections. State it once, in the most relevant section.
3. Lead with the verdict, then the evidence. Don't build up — say the conclusion first.
4. Flag unknowns explicitly rather than padding with hedging language.
5. Tables over prose for comparisons, metrics, and scoring.
6. Every section self-contained — structured for someone to ask "what about X?" and get a clean answer from that section alone.

Execution Steps

Step 1: Quick Screen

Check thesis fit. Note: hardware, consumer, and robotics are NOT automatic disqualifiers (updated per Ed 2026-04-30). Evaluate case-by-case based on:

• Is the core value prop software/AI-driven even if delivered via hardware?
• Does hardware create a moat (manufacturing relationships, supply chain)?
• Is the market opportunity large enough to justify hardware risk?

Output: PASS (stop) | INVESTIGATE | DEEP-DIVE.

Step 2: Parallel Deep Analysis (feeds Research Dossier)

Execute four research streams in parallel. Each stream writes its full findings to the research dossier — do not summarize at this stage. The executive memo will be distilled later in Step 4.

Market researcher:

1. WebSearch for TAM/SAM data, cross-reference with industry reports
2. Dossier output: TAM/SAM/SOM table with sources, market growth data, regulatory landscape, 5+ tailwinds/risks with evidence, adjacent market opportunities
3. Memo output (Step 4): TAM/SAM/SOM table (3 rows), 2-3 key tailwinds, 2-3 key risks

Biographical researcher:

1. WebSearch founder LinkedIn, prior companies, publications, patents, media appearances, verify key claims, check for red flags
2. Dossier output: Full founder profiles (education, career timeline, exits, board seats, domain expertise evidence, references found, social proof)
3. Memo output (Step 4): 3-5 sentence founder summary per founder. Verified claims vs unverified.

Competitive analyst:

1. WebSearch competitors, map positioning, analyze funding, product comparisons, customer overlap
2. Dossier output: Full competitive landscape (10+ competitors if available), feature comparison matrix, funding history per competitor, market share estimates, threat analysis
3. Memo output (Step 4): Competitive landscape table (3-5 competitors max) with funding, differentiation, threat level

VC analyst:

1. Extract unit economics, benchmark valuation, analyze cap table if available, comparable exits
2. Dossier output: Full financial analysis — unit economics breakdown, comparable company valuations, recent rounds in sector, bear/base/bull with assumptions, exit scenario modeling
3. Memo output (Step 4): Key metrics table + bear/base/bull scenario in ONE compact table

Step 2.5: True Potential Council (Ed's Framework)

Three weighted dimensions: Defensibility (0.35), Market Velocity (0.35), Expansion (0.30). Produces True Potential Score (1-5) with TAM calibration and wedge classification.

Full council prompts and scoring rubrics: See TRUE-POTENTIAL-COUNCIL.md

Step 3: Anti-Bias MoE Layer

Four council members challenge the thesis: Skeptical Judge, Conservative Analyst, Devil's Advocate, Value Investor. Each scores on /100 scale with a vote (INVEST/INVESTIGATE/PASS). Consensus = weighted average.

Full MoE prompts and thresholds: See ANTI-BIAS-MOE.md

Step 4: Synthesis & Output (Two Documents)

Generate two documents from the research collected in Steps 1-3:

4A: Research Dossier ({COMPANY}-RESEARCH.md)

The comprehensive knowledge base (15-25 pages). Contains ALL research findings with full detail, sources, and evidence. This document powers interactive Q&A during meetings.

HEADER BLOCK (signal, scores, confidence, metadata, generation date)
---
1. EXECUTIVE SUMMARY (full thesis paragraph)
2. COMPANY OVERVIEW (description, history, founding story, product details)
3. FOUNDER PROFILES (full career timelines, education, exits, verified/unverified)
4. MARKET ANALYSIS (TAM/SAM/SOM with sources, growth data, tailwinds, risks, regulatory)
5. COMPETITIVE LANDSCAPE (full competitor matrix, feature comparisons, funding, positioning)
6. PRODUCT DEEP DIVE (architecture, tech stack, IP/patents, demo notes, screenshots if available)
7. TRACTION & METRICS (revenue, customers, growth curves, engagement, retention, unit economics)
8. FINANCIAL ANALYSIS (valuation benchmarks, comparable deals, bear/base/bull with assumptions)
9. TRUE POTENTIAL COUNCIL (full deliberation, dimension scores, TAM calibration, wedge analysis)
10. ANTI-BIAS MOE COUNCIL (full arguments from all 4 members, individual scores, rebuttals)
11. IASV SCORECARD (detailed scoring rationale per dimension)
12. RISK REGISTER (all identified risks categorized: team, market, product, financial, regulatory)
13. DATA GAPS & UNKNOWNS (what we don't know, what to ask in a call)
14. SOURCE INDEX (all URLs, search queries used, data freshness dates)

4B: Executive Memo ({COMPANY}-EVALUATION.md)

The concise meeting document (5-7 pages). Distilled from the dossier. Every fact appears once.

HEADER BLOCK (signal, scores, confidence, metadata)
---
1. EXECUTIVE SUMMARY (3-5 sentences — the whole thesis in one paragraph)
2. COMPANY SNAPSHOT (table: sector, stage, location, team size, raised, ask, valuation)
3. TEAM (3-5 sentences per founder, verified/unverified flags)
4. MARKET & COMPETITION (TAM table + competitive landscape table, minimal prose)
5. PRODUCT & TRACTION (what it does, key metrics, growth — factual, no fluff)
6. TRUE POTENTIAL ASSESSMENT (scores table + wedge classification)
7. ANTI-BIAS COUNCIL (votes table + consensus)
8. SCORECARD (weighted scores table + total)
9. RECOMMENDATION (3-5 sentences: signal, reasons, key risks, next steps)
10. DATA GAPS (checklist of unknowns)

The memo cross-references the dossier — each section header links to the corresponding dossier section for drill-down during Q&A.

Step 5: Arithmetic Verification (MANDATORY)

Seven-point checklist: scorecard math, TPS arithmetic, header-body reconciliation, scale consistency, no ad-hoc adjustments, signal validation, base-score-only headers.

Full verification checklist: See VERIFICATION-CHECKLIST.md

Sector Evaluation Guidelines (Updated 2026-04-30)

No sector is an automatic disqualifier. IASV has invested across biotech, climate, space, defense, hardware, consumer, and more. The only hard filter is stage (Series B+ = too late).

Every deal is evaluated on its own merits. The key questions are:

1. Is the opportunity large enough?
2. Is the team executing?
3. Can IASV add value?

For hardware/consumer specifically: "Is the core business a software/AI problem delivered through a physical form factor?" If yes, this is often a strength — hardware creates moat, AI creates scale.

IASV Scorecard

Signal thresholds: 3.8-5.0 = INVEST | 2.8-3.7 = INVESTIGATE | Below 2.8 = PASS

Output Artifacts

The portal Q&A system reads from *-RESEARCH.md. The deal pipeline UI displays data from *-EVALUATION.md.

Completion Checklist

• Quick screen completed with clear signal
• All 4 parallel analyses completed with 3+ sources for market, 2+ for founders
• Bear/base/bull scenarios modeled
• True Potential Score calculated with TAM calibration
• Anti-bias council consensus generated with all 7 skeptical questions answered
• Research dossier generated — 15-25 pages, full depth, all sources indexed
• Executive memo generated — 5-7 pages, no redundancy, distilled from dossier
• Memo sections cross-reference dossier sections for Q&A drill-down
• Step 5 arithmetic verification passed — all 7 checks confirmed
• Both documents use consistent scores and signals (no header/body mismatch)

Failure Handling