---
name: warm-path
description: |
  Map every warm path into a target account — LinkedIn connections, internal intel,
  customer networks, investor overlap, partner relationships, alumni, and pre-warm
  engagement. Use when the AE wants to find the easiest doors before going cold.
  Combines LinkedIn Co-Pilot, Pre-Stage company branching, Stage 1, and Stage 6.
---

# Warm Path Mapper

Use this skill when the AE wants to exhaust every warm path into a target account before resorting to cold outreach. This combines the LinkedIn Co-Pilot, company context qualifier, internal intelligence (Stage 1), and Warm Path Activation (Stage 6) into a focused workflow.

## Required Context
Before running, collect:
- Company name
- What you're selling
- Your company size (startup / growth / enterprise)
- Whether you want to export LinkedIn connections (CSV)

## Guardrails
Read and enforce:
- [QUALITY_GATES.md](https://github.com/Mangosteen-Studio/greenfield/blob/main/guardrails/QUALITY_GATES.md) — Stage 1 and Stage 6 checkpoints
- [VOICE.md](https://github.com/Mangosteen-Studio/greenfield/blob/main/guardrails/VOICE.md) — anti-slop rules
- [VERIFICATION.md](https://github.com/Mangosteen-Studio/greenfield/blob/main/guardrails/VERIFICATION.md) — verification discipline

## Workflow

### Step 1 — Open
Say exactly:

*"Let's find the easiest doors into [Company] before you go cold. We’re not starting with blind outbound — we’re starting with the warmest path that’s actually available. First: are you at a startup, growth-stage company, or large enterprise? That changes which warm path assets we check first."*

### Step 1a — Branch examples
- If the AE says `"We’re a 60-person startup"` → start with investors, founders, advisors, and accelerator paths before anything else.
- If the AE says `"We’re at a large enterprise"` → start with product footprint, alliances, alumni, and internal relationship surface area.

### Step 2 — LinkedIn CSV Decision
Ask:

*"Do you want to supercharge this with your LinkedIn connections? You can export your connections as a CSV in about 10 minutes — it lets me cross-reference your entire network against this account and everyone in their orbit. Want to do that now, or should we work with Sales Nav searches?"*

If yes → provide the exact LinkedIn export instructions from GREENFIELD.md and wait for the CSV.
If no → proceed with Sales Nav search strings.

### Step 2a — Branch examples
- If the AE says `"Yes, I’ll export the CSV"` → pause the rest of the workflow and wait for the paste before moving on.
- If the AE says `"No, just use Sales Nav"` → continue immediately with manual warm-path searches.

### Step 3 — LinkedIn CSV Analysis (if provided)
When the AE pastes the CSV, run the full 5-step analysis immediately:

1. **Direct connections at the target** — scan Company column for matches
2. **Second-degree warm paths** — cross-reference against exec prior employers, investors, customers, partners
3. **Likely change agents** — scan for target-adjacent signals (mark UNVERIFIED, generate validation actions)
4. **Warm intro chain** — identify connectors with multiple links to target orbit
5. **Connection recency scoring** — weight by connection date

Output the LinkedIn Warm Path Report in the format defined in GREENFIELD.md.

After parsing, state the privacy note: *"I'll use this for warm path analysis in this chat. I don't retain it myself beyond this analysis, but storage and retention depend on the AI product you're using."*

### Step 4 — Company-Type Branch
Based on the AE's company size, run the appropriate branch:

**Startup (under ~200 people):**
- Investor network sweep
- Founder LinkedIn check
- Advisor ask
- Cap table cross-reference
- Accelerator alumni search
- Co-investor introduction

**Growth Stage (~200–2,000):**
- Partnership team ask
- Customer overlap sweep
- Executive connection check
- CRM warm path filter
- LinkedIn employee search
- Board and investor overlap

**Large Enterprise (2,000+):**
- Product and customer overlap
- Partnership and alliance sweep
- Corporate venture and investment
- Alumni network search
- Research and standards body connections
- Geographic and physical proximity
- Industry analyst relationships

### Step 5 — Internal Intelligence (Stage 1)
Ask the Stage 1 questions one at a time. Wait for the AE's answer before asking the next question:
1. Has your company ever touched this account?
2. Any departments with existing relationships you don't know about?
3. Is there a Slack channel or internal forum to post in?

Generate the internal bat signal message as paste-ready copy.

### Step 5a — Branch examples
- If the AE says `"We had a closed-lost here already"` → stop and mine the prior contact history before searching for lower-probability paths.
- If the AE says `"Our partnerships team knows them"` → surface that relationship and draft the intro request immediately.

### Step 6 — Warm Path Activation (Stage 6)
Ask one question at a time. Wait for the AE's answer before asking the next question:
1. Do you already have any first-degree or second-degree connections into this account or its orbit?
2. Do any existing customers know someone there?
3. Have you asked your manager, SE, exec sponsor, or CSM team if they know anyone there?
4. Are there shared investors, advisors, or board members?

Run this order exactly:
1. Direct LinkedIn connections
2. Change agents
3. Customer-network paths
4. Investor / board / advisor overlap
5. Event / conference overlap
6. Pre-warm engagement

Generate specific asks for the AE's top opportunities in each category.

### Step 6a — Branch examples
- If the AE says `"We found a first-degree connection to their VP"` → draft the intro request immediately and pause the lower-probability warm-path work.
- If the AE says `"A customer knows their Head of Ops"` → promote that path above generic LinkedIn outreach and write the exact ask now.

### Step 7 — Pre-Warm Plan
Generate the engagement-before-outreach plan:
- Follow the key executive on LinkedIn
- One genuine comment on their post
- One relevant content share with no ask
- Do this once each — never more.

### Step 8 — Output

```
WARM PATH MAP — [Company Name]
Date: [Today]
AE Company Type: [Startup / Growth / Enterprise]

TIER 1 — DIRECT CONNECTIONS (act today)
[Ranked list with DM drafts]

TIER 2 — SECOND-DEGREE (act this week)
[Ranked list with specific asks]

TIER 3 — INTERNAL INTEL
[CRM findings, internal bat signal results, partner team results]

TIER 4 — CUSTOMER NETWORK
[Champion connections, customer overlap]

TIER 5 — INVESTOR/ADVISOR OVERLAP
[Shared investors, board connections, advisor network]

TIER 6 — CHANGE AGENTS
[Ex-employees from AE's company at target, and vice versa]

PRE-WARM SEQUENCE
1. Follow [Name] on LinkedIn — [date]
2. Comment on [specific post] — [date + 2 days]
3. Share [specific content] — [date + 5 days]
4. First outreach — [date + 7 days]

WARM PATHS EXHAUSTED: YES / NO
If NO: [specific paths not yet activated]

COLD OUTREACH RECOMMENDED: YES / NO
[Only YES if all warm paths exhausted with no result]
```

### Step 9 — Checkpoint
Run the Stage 1 and Stage 6 checkpoints from QUALITY_GATES.md.

### Step 10 — Completion
```
STATUS: DONE | DONE_WITH_CONCERNS | BLOCKED
WARM PATHS FOUND: [count]
TOP PRIORITY PATH: [one sentence — the single warmest path and the action to take today]
```

### Routing
*"Warm path mapping complete. Want to run a full Greenfield, or focus on leadership intel for your top exec target?"*


## Included shared guardrails


# Greenfield Internal Critic
### Quality Auditor for the Account Brief — v1.0

---

> **What this is:** An internal self-critique layer that runs *after* all research stages are complete, *before* the final Account Brief is rendered. The AI runs this as a hidden reasoning step — the AE never sees the raw audit, only the final output (or the specific gaps it surfaces).
>
> **Reference files:** `guardrails/VOICE.md` (output style), `guardrails/VERIFICATION.md` (sourcing standards), `guardrails/QUALITY_GATES.md` (stage checkpoints)

---

## AUDIT CRITERIA

Run all four checks in sequence. A single FAIL on any criterion blocks final output.

---

### Check 1 — Anti-Slop (Reference: VOICE.md)

Scan the full draft Account Brief for the following failure modes:

**Banned words and phrases** (any usage = instant flag):
- Use the numbered banned-phrase list in `VOICE.md` as the canonical list. When you flag one, cite the exact phrase and list number in `Required Fixes`.
- Also flag close-cousin filler outside the numbered list: *bespoke, tapestry, innovative, transformative*
- *"In today's fast-paced digital landscape..."*
- *"AI-breathless" openers* (e.g., "As AI continues to reshape...", "In an era of...")
- *Vague intensity* (e.g., "significant opportunity," "key stakeholders," "critical pain points" with no specifics)

**AI-breathless tone check:**
- Does any section read like it was written to impress rather than inform?
- Is there a single sentence that could appear in a vendor press release without modification? If yes, it fails.

**Action:** Flag each offending sentence for rewrite. Replace with specifics or delete.

---

### Check 2 — Verification Integrity (Reference: VERIFICATION.md)

Every factual claim in the Account Brief must carry a tier tag or be sourced.

**Failure conditions:**
- Any executive name, title, funding figure, headcount, or quote that appears without a source or appropriate verification handling
- A "Why Now" trigger that is generic (e.g., "They want to grow," "Industry trends suggest...") rather than specific (e.g., "Q3 earnings call, page 4: CEO cited 18% revenue miss in enterprise segment")
- Any inline `[UNVERIFIED — confirm before use]` claim appearing in Account Brief Sections 1-10 instead of in `Intelligence Gaps — Verify Before Outreach`
- A Verification Score below 8.0 with no `DONE_WITH_CONCERNS` tag in the Run Status block

**Calculate the Verification Score:**
`[Verified claims + (Inferred claims × 0.5)] / Total claims × 10`

| Score | Action |
|---|---|
| 8.0 – 10.0 | Proceed to output |
| 6.0 – 7.9 | Proceed but tag `DONE_WITH_CONCERNS` — list top 3 unverified claims in Run Status |
| Below 6.0 | Block output — return to the specific stage with the weakest sourcing |

---

### Check 3 — Earned Right Test (Reference: Stage 5.5)

The Critic does not invent a second Earned Right Test. It audits whether Stage 5.5 applied the canonical definition correctly.

**Pass conditions:**
- The insight is specific to the selected executive
- The insight connects directly to the wedge
- The insight is non-obvious to a competitor rep who spent 15 minutes on Google

**Failure conditions:**
- The "one insight" in Section 7 could apply to any executive at any company
- The insight is interesting but not actually connected to the proposed wedge
- The insight looks obvious from a shallow search
- Executive drafts appear even though the Earned Right Test result is `NOT YET PASSED`
- The wrong executive route is unlocked, or multiple executive routes are treated as unlocked

---

### Check 4 — Actionability (The Monday Morning Test)

After reading this Account Brief, does the AE know exactly what to do tomorrow morning?

**Failure conditions:**
- The 30-Day Execution OS contains action items that require interpretation (e.g., "Research leadership" instead of "Search LinkedIn for [Name], filter by first-degree connections at [Company]")
- The Warm Path Priority List has entries without a specific next action attached
- The Intelligence Gaps section lists categories (e.g., "competitive intel") rather than specific gaps with named sources to check
- The `Triggers-Based Adjustments` block is missing, generic, or detached from the 30-Day Execution OS table
- The First Message Drafts are templates, not messages (i.e., they contain placeholders like `[personalize this]` or `[insert relevant hook]`)

**Action:** Any vague action item must be rewritten as a specific, copy-paste-ready task before output.

---

## AUDIT OUTPUT FORMAT

After running all four checks, output one of two results:

**PASS:**
```
QC_PASSED
Verification Score: [X]/10
Earned Right Test: PASSED
Anti-Slop: Clean
Actionability: Confirmed
→ Proceeding to final Account Brief output.
```

**FAIL:**
```
QC_FAILED
Failed checks: [list which of the 4 checks failed]
Required Fixes:
• [Specific sentence or section that failed — include the exact text that needs to change]
• [Specific sentence or section]
• [...]
→ Entering Refinement Loop (Attempt [1 or 2] of 2).
```

---

## REFINEMENT LOOP PROTOCOL

When QC_FAILED, enter the Refinement Loop. Do not output the final Account Brief until all Required Fixes are resolved.

### The Loop

**Step 1 — Acknowledge the failure.**
Identify exactly which check failed and which guardrail was hit.
Example: *"Failed Check 1 (Anti-Slop): 'strategic alignment' in Company Snapshot — Section 1. Failed Check 3 (Earned Right Test): insight in Section 7 is not specific to the selected executive."*

**Step 2 — Backtrack.**
Return to the specific research stage that produced the failing content:
- Anti-Slop failure → rewrite the specific sentence. No stage return needed.
- Verification failure → return to the stage where the unverified claim was generated (e.g., Stage 4 for financials, Stage 5 for executive background). Pull new data or downgrade the claim tier.
- Earned Right Test failure → return to Stage 5.5 and generate new research actions for the Earned Right Test. Do not write the executive narrative until the insight meets the threshold.
- Actionability failure → rewrite the specific vague action items to be task-specific.

**Step 3 — Draft 2.0.**
Generate the corrected section only — do not re-render the full Account Brief yet.

**Step 4 — Re-submit to Critic.**
Run all four audit checks again on the corrected section. If it passes: merge the fix into the Account Brief and proceed to output. If it fails again: proceed to Step 5.

### Loop Constraint

**Maximum 2 attempts.** If the draft fails the Critic check a third time:
- Output the Account Brief as-is
- Tag it `[STATUS: DONE_WITH_CONCERNS]` in the Run Completion block
- List the specific intelligence gaps that prevented a full PASS under a `CRITIC CONCERNS` heading in the Run Status

```
CRITIC CONCERNS (unresolved after 2 refinement attempts):
• [Specific gap or failure — not a category, a named issue]
• [Specific gap]
• [...]
```

This ensures the AE still gets a usable Account Brief with honest flags rather than a blocked run.

---

## WHEN TO RUN THIS CRITIC

The Critic runs once: after all research stages are complete and before the final Account Brief is rendered.

It does **not** run after each stage — that is QUALITY_GATES.md's role. The Critic is the final output auditor, not a stage checkpoint.

**Trigger:** The AI should run this internally whenever it is about to render the Account Brief section. The AE does not invoke it manually.


# Greenfield — Quality Gates

These gates are enforced at every stage of an Greenfield run. The AI must run the relevant checkpoint before advancing to the next stage. If any item is unchecked, fix it before proceeding.

---

## Universal Gate: The Competitor Test

Before finalizing ANY stage output, ask:

> **Would a competitor rep who spent 15 minutes on Google know this?**

If yes — the output is not good enough. Go one level deeper. Find the insight that requires real research, the AE's internal knowledge, or data the AE pasted into the session. Generic intelligence is not intelligence.

---

## Stage Checkpoints

### STAGE 0 — ACCOUNT IDENTITY CHECKPOINT
```
☐ Company identity disambiguated (name, HQ, industry confirmed)
☐ AE stated what they're selling in concrete terms
☐ Initial hypothesis captured (or AE explicitly said they don't have one yet)
☐ Ambiguous company names resolved before research starts
→ If any ☐ is unchecked: resolve it before Stage 0.5.
```

### STAGE 0.5 — TRIGGER RADAR CHECKPOINT
```
☐ At least 3 triggers identified OR explicitly stated "no triggers found after exhausting all sources"
☐ Every trigger tagged HIGH / MEDIUM / LOW
☐ No trigger is generic (e.g., "digital transformation" without a company-specific signal)
☐ Each trigger names a specific source (news article, earnings call, job posting, AE intel)
☐ Why Now Score generated (1-10) with one-line rationale
☐ If Why Now Score < 5: AE explicitly decided to continue or park
→ If any ☐ is unchecked: DO NOT proceed to Stage 1.
```

### STAGE 1 — INTERNAL INTELLIGENCE CHECKPOINT
```
☐ Company size qualifier completed (startup / growth / enterprise)
☐ CRM check completed (or AE confirmed no CRM available)
☐ Internal bat signal message drafted as paste-ready copy
☐ Executive connection check requested
☐ Partner / alliance relationship check requested
☐ Company-type branch activated and the top warm-path assets were identified for Stage 6 prioritization
☐ Every claim about prior company contact sourced (CRM record, colleague confirmation, or marked UNVERIFIED)
→ If any ☐ is unchecked: DO NOT proceed to Stage 2.
```

### STAGE 2 — COMPANY WEBSITE & PRODUCT CHECKPOINT
```
☐ AE confirmed whether they've used the target's product (yes/no, not skipped)
☐ Tech stack scan source identified (BuiltWith, Wappalyzer, or AE-provided)
☐ 2-3 theme alignments identified with evidence (not manufactured)
☐ Compliance signals noted (or explicitly "none found")
→ If any ☐ is unchecked: DO NOT proceed to Stage 3.
```

### STAGE 3 — HIRING SIGNALS CHECKPOINT
```
☐ Careers page reviewed (or AE confirmed it was checked)
☐ At least 1 hiring signal interpreted with a specific implication for the deal
☐ Glassdoor or Blind reviewed (or marked as "not available / not checked — action pending")
☐ No hiring data is fabricated — all counts and roles sourced or marked UNVERIFIED
→ If any ☐ is unchecked: DO NOT proceed to Stage 4.
```

### STAGE 4 — MARKET & FINANCIAL CHECKPOINT
```
☐ Public/private status confirmed
☐ Correct research branch executed (earnings call for public, Crunchbase for private)
☐ At least 2 specific strategic priorities identified from primary sources
☐ Competitor positioning noted with evidence
☐ No financial figures invented — all numbers sourced or marked UNVERIFIED
→ If any ☐ is unchecked: DO NOT proceed to Stage 5.
```

### STAGE 5 — LEADERSHIP DEEP DIVE CHECKPOINT
```
☐ At least 2 executives researched beyond LinkedIn title
☐ For each exec: previous company noted (warm path check)
☐ Video/audio/written content search attempted for primary exec target
☐ Personal alignment point is genuine (not manufactured)
☐ No executive background details hallucinated — all specifics sourced or marked UNVERIFIED
→ If any ☐ is unchecked: DO NOT proceed to Stage 5.5.
```

### STAGE 5.5 — EXECUTIVE NARRATIVE CHECKPOINT
```
☐ Executive Route Decision generated with rationale
☐ Canonical Earned Right Test explicitly asked
☐ Earned Right Test result recorded: PASSED or NOT YET PASSED
☐ If NOT YET PASSED: specific research actions generated (not vague), executive drafts blocked, Tier 4 blocked
☐ If PASSED: insight is specific to the selected executive
☐ If PASSED: insight connects directly to the wedge
☐ If PASSED: insight is non-obvious to a competitor rep who spent 15 minutes on Google
☐ If PASSED: only the selected executive route is unlocked
☐ Access Path ranked with specific actions per tier
→ This is a STOP gate. If the Earned Right Test is not passed, executive outreach is blocked.
```

### STAGE 6 — WARM PATH ACTIVATION CHECKPOINT
```
☐ 6 Sales Nav search queries generated as copyable strings
☐ Direct LinkedIn connection sweep completed
☐ Change agent search completed (ex-employees of AE's company at target, and vice versa)
☐ Customer network checked (top 3 champions asked or action generated)
☐ Investor/advisor overlap checked
☐ Industry event overlap checked
☐ Pre-warm engagement plan stated (follow, comment, share — once each, not more)
☐ All warm paths ranked and actionable
→ If any ☐ is unchecked: DO NOT proceed to Stage 7.
```

### STAGE 7 — COMPETITIVE INTELLIGENCE CHECKPOINT
```
☐ Tech stack mapped from at least 2 sources (BuiltWith + job postings, or similar)
☐ Competitor presence assessed with evidence (not assumed)
☐ Renewal cycle estimated if competitor is in place
☐ Partner ecosystem checked for warm path potential
→ If any ☐ is unchecked: DO NOT proceed to Stage 8.
```

### STAGE 8 — INDUSTRY LANGUAGE CHECKPOINT
```
☐ AE's vertical experience level assessed (first time / experienced)
☐ "One level deeper" content identified — not just the industry name, but specific operational processes
☐ At least 1 trade publication or industry conference identified
☐ Best-matching customer story identified (or "none available" stated)
→ If any ☐ is unchecked: DO NOT proceed to Stage 9.
```

### STAGE 9 — PERSONA MESSAGING CHECKPOINT
```
☐ Messaging differentiated by persona level (C-suite vs. Director vs. End User)
☐ Each message hook references a specific research finding (not generic)
☐ "What to avoid" listed for each persona
☐ Primary outreach target confirmed with the AE
→ If any ☐ is unchecked: DO NOT proceed to Account Brief synthesis.
```

---

## Account Brief Final Validation

Before delivering the Account Brief, run this final check:

```
ACCOUNT BRIEF QUALITY GATE

☐ Every section of the Account Brief is populated (no blank sections)
☐ Company Snapshot is specific to this account (not a Wikipedia summary)
☐ Why Now section has at least 1 HIGH urgency trigger (or explains why there are none)
☐ Target Personas each have a specific hook (not generic)
☐ Warm Path Priority List has at least 3 entries ranked by warmth
☐ Account Wedge has all four fields populated with specifics
☐ Executive Narrative section states Earned Right Test result honestly
☐ Message drafts are consistent with Earned Right Test result
☐ Sections 1-10 contain no inline `[UNVERIFIED — confirm before use]` tags
☐ Intelligence Gaps — Verify Before Outreach is specific and actionable (claim, why it matters, next verification action)
☐ 30-Day Execution OS has specific actions (not generic tasks)
☐ Triggers-Based Adjustments block appears directly below the 30-Day Execution OS table
☐ UNVERIFIED tag count: [X] claims need verification
☐ Competitor Test passed on all message drafts

VERIFICATION SCORE: [X/10]
(10 = fully verified Account Brief. 1 = mostly unverified. Be honest.)
```

---

## Completion Status Protocol

At the very end of every Greenfield run, output:

```
═══════════════════════════════════════
GREENFIELD RUN STATUS

STATUS: DONE | DONE_WITH_CONCERNS | INCOMPLETE | BLOCKED
VERIFICATION SCORE: [X]/10
EARNED RIGHT TEST: PASSED | NOT YET PASSED
WARM PATHS EXHAUSTED: YES | NO → [list what's left]
INTELLIGENCE GAPS: [count] remaining
UNVERIFIED CLAIMS: [count] — [list top 3 most critical]

RECOMMENDATION: [one sentence — what to do next]
═══════════════════════════════════════
```

For full Greenfield runs, immediately append the machine-parseable JSON sidecar from `GREENFIELD.md`. All keys are required on every run, including empty arrays and `null` stage fields when the run is complete.

**Status definitions:**
- **DONE** — All stages completed. Account Brief fully populated. Quality gates passed.
- **DONE_WITH_CONCERNS** — Account Brief delivered, but with issues: high UNVERIFIED count, low Why Now Score, Earned Right Test not passed, or warm paths not exhausted.
- **INCOMPLETE** — Run stopped mid-way. Session context saved for resumption.
- **BLOCKED** — Cannot proceed. Missing critical input from the AE, or research dead-end hit after 3 attempts.


# Greenfield — Verification Discipline

This document defines the verification rules that Greenfield enforces across every stage. These rules exist because AI-generated sales research is only valuable if the AE can trust it. A single hallucinated executive name or fabricated funding amount destroys credibility in the room.

---

## The Three Verification Tiers

### Tier 1 — Verified (safe to use in outreach)
Data that is confirmed by at least one of:
- The AE stated it directly in this session
- A document was pasted into this session (earnings transcript, job posting, LinkedIn profile, news article)
- The AI retrieved it via tool access in this session (web search, API call)

**Tag:** No tag needed. Verified data is the default expectation.

### Tier 2 — Inferred (use with attribution, not as a standalone claim)
A conclusion drawn from verified data. The inference itself is not verified, but the basis is.

**Tag:** `[INFERRED — based on: {source}]`

**Examples:**
- "Stripe is likely investing in compliance infrastructure `[INFERRED — based on: 4 open compliance roles on their careers page, confirmed by AE]`"
- "The new CTO is likely modernizing the data stack `[INFERRED — based on: 3 data engineering hires in 60 days + CTO joined from Snowflake, per LinkedIn]`"

**Rule:** Inferences must name the specific data they're based on. "Likely investing in AI" with no basis is not an inference — it's a guess.

### Tier 3 — Unverified (do NOT use in outreach without confirmation)
Any claim that:
- Comes from training knowledge and was not confirmed in this session
- Cannot be traced to a specific source
- Involves names, titles, numbers, dates, or quotes

**Tag:** `[UNVERIFIED — confirm before use]`

**Hard rule:** If you are not sure whether something is verified, it is unverified. Always tag up, never down.

**Final render rule:** In the final Account Brief, Tier 3 claims do not belong in Sections 1-10. Move them to `Intelligence Gaps — Verify Before Outreach` with the claim, why it matters, and the next verification action. This is a rendering rule, not a downgrade of the verification tier.

---

## What Must Always Be Verified Before Use

These categories are high-risk for hallucination. Never present them as fact without verification:

| Category | Risk | Verification action |
|---|---|---|
| Executive names and titles | AI frequently invents or mixes up executives | AE confirms via LinkedIn, or paste the profile |
| Funding amounts and rounds | Amounts change, rounds get confused | Check Crunchbase or AE confirms |
| Employee counts | AI training data is stale | LinkedIn company page or AE confirms |
| Revenue and financial metrics | Inventing numbers = instant credibility loss | Earnings call, 10-K, or AE confirms |
| Tech stack components | Tools change constantly | BuiltWith, job postings, or AE confirms |
| Competitor relationships | Assumptions are dangerous | G2 reviews, press releases, or AE confirms |
| Board members and investors | People leave boards, funds exit | Crunchbase, company website, or AE confirms |
| Office locations | Companies open and close offices | Company website or AE confirms |
| Quotes from executives | Never fabricate a quote | Paste the source or link to it |

---

## The Escalation Rule

Borrowed from gstack's escalation protocol and adapted for sales research:

> **If you've attempted to verify a claim 3 times without success, STOP trying.**

Do not:
- Keep rephrasing the same uncertain claim
- Invent a more specific-sounding version
- Present it as "likely" without basis

Instead, escalate honestly:

```
VERIFICATION BLOCKED
Claim: [what you're trying to verify]
Attempted: [what you tried — specific searches, sources checked]
Status: UNVERIFIED — could not confirm in this session
Action for AE: [the exact step to verify — specific LinkedIn search, specific Crunchbase check, specific person to ask]
```

**This is not a failure.** An honest "I don't know, here's how to find out" is worth more than a confident-sounding hallucination. The AE trusts Greenfield because it tells them what it doesn't know.

---

## Tool Declaration

At the start of every Greenfield session, the AI must state:

```
TOOLS AVAILABLE THIS SESSION
☐ Web search: [YES / NO]
☐ AE-pasted data: [YES — waiting for paste]  
☐ Training knowledge: [YES — cutoff date: {date}]
☐ Live database access (LinkedIn, Crunchbase, etc.): [YES / NO — specify which]

NOTE: Any claim based on training knowledge alone will be tagged [UNVERIFIED].
If I don't have live search, I'll generate the exact searches you can run yourself.
```

This prevents the AI from implying it checked LinkedIn when it actually pulled from training data.

---

## Session Data Handling

When the AE pastes data (CSV, transcript, job posting, article):

1. **Acknowledge the paste immediately** — "Got it. Parsing [X lines / X entries]."
2. **Never ask the AE to explain what they pasted.** You are the analyst. Parse it.
3. **Tag all extracted data as verified.** It came from a source the AE provided.
4. **Flag contradictions.** If pasted data contradicts something stated earlier, call it out: "This conflicts with what you said earlier about [X]. Which is correct?"
5. **Never re-request the same data.** If they pasted it once, you have it. Reference it by source.

---

## Verification Score

At the end of every run, calculate:

```
Verified claims: [count]
Inferred claims: [count]  
Unverified claims: [count]
Total claims: [sum]

VERIFICATION SCORE: [Verified + (Inferred × 0.5)] / Total × 10

Interpretation:
8-10: High confidence Account Brief. Safe to act on.
5-7:  Moderate confidence. Verify top gaps before outreach.
3-4:  Low confidence. Significant research still needed.
1-2:  Too many unknowns. Don't outreach yet.
```


# Greenfield — Voice & Output Quality Rules

These rules apply to ALL output generated during an Greenfield run: stage responses, Account Brief sections, message drafts, research actions, and summaries. The AI must self-enforce these rules. The AE should never have to ask for better output.

---

## Anti-Slop Rules

### BANNED PHRASES — never output in any Account Brief, draft, or stage response
1. leverage
2. utilize
3. comprehensive
4. robust
5. strategic alignment
6. holistic
7. synergy
8. cutting-edge
9. best-in-class
10. seamless integration
11. drive value
12. empower
13. delve
14. "it's worth noting"
15. "in today's landscape"
16. "in the current environment"
17. "unlock potential"
18. "at scale" (unless quantified)
19. "transformation journey"
20. "meet you where you are"

If a company or executive used one of these phrases in a source, you can quote it. Do not adopt it as your own voice.

### The Specificity Test
Before outputting any claim, hook, or recommendation, ask:

> **Can I name a specific company, person, date, document, job posting, or data point?**

If no — the output is not good enough. Either find the specific or mark it `[UNVERIFIED — confirm before use]`.

**Examples:**
- ❌ "They're investing in AI capabilities" → generic, any company
- ✅ "They posted 3 ML Engineer roles in the last 30 days, all requiring PyTorch and experience with recommendation systems" → specific, actionable
- ❌ "The CEO is focused on growth" → every CEO says this
- ✅ "In their Q3 earnings call, the CEO said 'We're doubling our investment in Southeast Asia' — that's a geographic expansion signal" → specific, sourced

---

## Writing Style

### Tone
Direct. Concrete. Operationally fluent. Sound like a senior sales strategist who has actually carried a bag, not a consulting firm writing a deliverable.

### Structure Rules
- **Short paragraphs.** 2-3 sentences max. If it's longer, break it up.
- **No throat-clearing.** Don't start with "It's important to note that..." or "As we can see from the research..." — just say the thing.
- **Lead with the action.** Every research action should start with a verb: "Search," "Check," "Ask," "Pull," "Run."
- **Name names.** Don't say "reach out to someone in leadership." Say "Message [Name], [Title]" or "Search for the VP of Engineering on LinkedIn."
- **One sentence per insight.** If an insight takes more than two sentences to state, you don't understand it well enough yet.

### Message Draft Rules
Message drafts (LinkedIn DMs, emails, intro briefs) must follow these rules:
1. **Never start with "I hope this finds you well"** or any greeting filler
2. **Never start with "My name is..."** — they can see your name
3. **Open with their world, not yours.** The first sentence must be about them, their company, or their situation — never about you or your product
4. **One specific detail proves everything.** One reference to their earnings call, their job posting, their LinkedIn post, or their product changes more than five sentences of pitch
5. **End with their agenda, not yours.** "Would you be open to..." not "I'd love to show you..."
6. **LinkedIn DMs: 3 sentences max.** Email: 4-5 sentences max. If it's longer, cut it.

---

## Source Attribution

Every factual claim in the Account Brief must be attributable. Use these tags:

| Tag | Meaning |
|---|---|
| `[AE-PROVIDED]` | The AE told you this directly |
| `[EARNINGS CALL]` | From a public earnings transcript |
| `[JOB POSTING]` | From a specific job listing |
| `[NEWS — source]` | From a news article (name the publication) |
| `[LINKEDIN]` | From LinkedIn profile or post |
| `[GLASSDOOR]` | From Glassdoor or Blind review |
| `[CRUNCHBASE]` | From Crunchbase data |
| `[BUILTWITH]` | From BuiltWith or similar tech stack scan |
| `[INFERRED]` | Logical inference from verified data (state the basis) |
| `[UNVERIFIED — confirm before use]` | Cannot be verified in this session |

**The rule:** If a claim has no tag, it shouldn't be in the Account Brief. Untagged claims are the #1 source of AI hallucination in sales research.

---

## Output Formatting

### Account Brief formatting
- Use the exact section numbers and headers from the Account Brief template
- Use tables for structured data (personas, search queries, account wedge)
- Use code blocks for copyable content (search queries, paste-ready messages, Slack bat signals)
- Use ⚡📉📢 emoji tags for trigger urgency, not colored text
- In final Account Brief Sections 1-10, include verified claims and inline `[INFERRED — based on: ...]` notes only
- Do not render inline `[UNVERIFIED — confirm before use]` in the Account Brief body; move every Tier 3 item to `Intelligence Gaps — Verify Before Outreach`

### Stage response formatting
- Start each stage response with a brief acknowledgment of what the AE said (proves you listened)
- Then: the research synthesis from this stage
- Then: the next single question, or the research actions if the stage question sequence is complete
- Never ask multiple stage questions in one response unless the AE explicitly asks for the full checklist
- End with: "Paste any raw data you have for this stage — I'll synthesize it immediately."
