dl-sector-screen

name: dl-sector-screen description: > Decomposes a sector, industry, or sub-industry into discrete sub-verticals, screens each against the Overland industry attractiveness framework, and produces a structured markdown handoff with cascade anchor companies, NAICS codes, trade orgs, and conference targets for downstream borrower identification. Use whenever the user wants to develop sourcing coverage on a sector or industry — trigger phrases include "research X sector", "screen the Y industry", "develop sourcing coverage on Z", "decompose W into sub-verticals", "what sub-verticals in Q are attractive", or any freeform sector-level sourcing question. Use this even when the user describes a sector thematically (e.g., "aging infrastructure plays", "outsourced facilities services") rather than naming it directly. Do NOT use when the user names a specific company or deal (route to borrower identification or posting memo workflows), or for sector-level work unrelated to Overland sourcing such as public equity research or general market analysis.

Sector Research & Sub-Vertical Identification

Phase 1 of the Overland sourcing workflow. Decompose a sector, screen sub-verticals against the Overland industry attractiveness framework, and produce a structured markdown handoff for downstream borrower identification.

Audience and tone

Output is read by senior direct lending professionals — assume full institutional private credit fluency (do not define EBITDA, leverage, FCF conversion, or tier structure). Write as a credit-side analyst would: lead with the insight, flag uncertainty explicitly, avoid framework citations in prose.

Workflow

1. Disambiguate scope. If the user names a sector that is too broad to produce a focused output (e.g., "industrials", "healthcare", "business services"), ask one clarifying question to narrow before proceeding. If the input names a single company or specific deal rather than a sector, stop and redirect to borrower identification or the posting memo workflow. Otherwise proceed silently.

2. Load the attractiveness screen. Before any scoring, read reference/overland-industry-attractiveness-screen.md using the view tool. This file defines the eight Porter-derived FCF signals, the three-tier competitive structure model, and the micro FCF driver checklist. All scoring rationale must map to its named categories. Do not reproduce its contents in output — reference the categories by name only.

3. Decompose into sub-verticals. Identify 4–10 discrete sub-verticals within the input sector. Anchor decomposition to NAICS 4-digit groupings where possible; deviate when industry economics cut across NAICS boundaries (common in services). For broad inputs that survived Step 1, cap at 5 sub-verticals and flag partial coverage in the output preamble.

4. Score each sub-vertical against the screen. Use web search and PitchBook research to assess each sub-vertical against the eight FCF signals and the tier structure model. Produce one of three verdicts:

   • Pursue — clears the screen materially, three-tier structure intact with viable Tier 2 population
   • Watch — clears most signals but has one or more flagged risks worth monitoring
   • Screened Out — fails the screen materially or lacks Tier 2 addressability

   Verdict rationale must cite specific screen categories by name. Examples: "Entry Barriers: state applicator licensing + EPA registration", "FCF Conversion: asset-light, fleet maintenance capex ~3% of revenue", "Buyer Power: fragmented commercial buyer base, no single account >5% of typical Tier 2 revenue".

5. Identify cascade anchors for Pursue and Watch sub-verticals. Name 3–5 large reference companies per sub-vertical that can serve as organizational anchors when searching for smaller comps in the next phase. Mix tiers where possible — at least one public company or BSL borrower, at least one PE-backed Tier 1 platform. Use PitchBook to validate platform sponsorship and recent transaction activity.

   Each anchor entry includes: company name, type (Public | BSL | PE-Platform), and a one-line descriptor naming the operating model. Anchors must be currently operating businesses — do not list companies acquired, merged out of existence, or wound down.

6. Map NAICS codes. For each Pursue and Watch sub-vertical, list 1–3 candidate NAICS codes (4- or 6-digit). Prefer 6-digit specificity when the sub-vertical maps cleanly; fall back to 4-digit when the sub-vertical spans multiple 6-digit codes.

7. Identify trade orgs and conferences. For each Pursue and Watch sub-vertical, list relevant trade associations, membership directories, and industry conferences. Specificity matters — prefer the industry-specific org over a generic umbrella body (the HVAC contractor association, not the manufacturing umbrella). For conferences, flag any with publicly-available attendee or exhibitor lists; these are high-yield borrower identification sources downstream.

   Each entry includes: name, URL (if known), brief descriptor, and exhibitor_list: yes | no | unknown.

8. Draft the thesis paragraph. For each Pursue and Watch sub-vertical, write a 1–2 sentence thesis characterizing the Tier 2 addressable opportunity — operating model, scale band of typical Tier 2 incumbents, and the structural feature that makes them Overland-suitable. Analytical prose only. Do not use the words "Porter", "five forces", or "base rate" in any output. Do not recite framework categories in narrative form — the verdict rationale field is where category mapping lives.

9. Handle edge cases.

   • Oligopolistic structure (no Tier 2 population, e.g., elevator OEM service): score as Screened Out with rationale "Limited Overland addressability — Tier 2 population thin or non-existent."
   • Geographically or regulatorily concentrated sub-verticals: score normally but populate the Scope caveat field naming the constraint.
   • Sub-verticals that fail the screen entirely: place in the ## Screened Out section with a one-sentence rejection rationale. Do not omit silently.
   • Unresolvable factual gaps (contested HHI, unknown sponsor activity, unclear regulatory regime): list under ## Open Questions rather than guessing.

10. Write output. Produce a single markdown file at outputs/p1_sector-research_<slug>.md using the schema below. Section headers, field names, and the Pursue/Watch/Screened Out taxonomy are stable — the downstream borrower identification skill parses against this schema.

Output schema

ALWAYS produce a single markdown file matching this schema exactly. Use analytical prose in the Thesis field; use bullets only inside the structured fields.

# Sector Research: [Sector Name]
**Date:** [ISO date]
**Input scope:** [original user input, verbatim]
**Coverage note:** [full | partial — and why, if partial]

## Pursue

### [Sub-Vertical Name]
- **NAICS:** [code(s)]
- **Verdict rationale:** [2–4 sentences mapping to screen categories by name]
- **Scope caveat:** [if applicable; otherwise omit field]
- **Thesis:** [1–2 sentence prose paragraph on Tier 2 addressable opportunity]
- **Cascade anchors:**
  - [Company] — [Public | BSL | PE-Platform] — [one-line descriptor]
  - [3–5 entries total]
- **Trade orgs & conferences:**
  - [Name] — [URL] — [descriptor] — exhibitor_list: [yes | no | unknown]
  - [as many as relevant; prioritize specificity over count]

[Repeat for each Pursue sub-vertical]

## Watch

[Same schema as Pursue. Include a `Watch flags:` field noting the specific screen risks being monitored.]

## Screened Out

### [Sub-Vertical Name]
- **NAICS:** [code(s)]
- **Rejection rationale:** [one sentence]

[Repeat for each Screened Out sub-vertical]

## Open Questions
[Unresolved factual gaps. One bullet per gap.]

If any required field cannot be resolved, write [INSUFFICIENT DATA — <reason>] in the field. Do not fabricate.

Worked example (single sub-vertical block)

The level of specificity expected in a populated Pursue entry:

### Commercial fire suppression service & inspection
- **NAICS:** 561621 (Security Systems Services), 238220 (Plumbing/HVAC/Sprinkler)
- **Verdict rationale:** Entry Barriers — NFPA certification plus AHJ approval at the local jurisdiction level create meaningful credentialing lag. Non-Discretionary Demand — code-mandated annual inspection and quarterly testing under NFPA 25; deferral creates direct regulatory and insurance exposure. Substitution Risk — none; sprinkler systems are code-required and not displaceable. Capex Profile — asset-light, primarily van fleet and inspection equipment; maintenance capex ~3% of revenue.
- **Thesis:** Regional Tier 2 inspection-and-service operators in the $5–25M EBITDA band with route-dense commercial customer bases (office, light industrial, multi-family) are highly Overland-suitable — recurring inspection revenue under NFPA 25 produces through-cycle EBITDA stability and the credentialing moat insulates against price-led entry.
- **Cascade anchors:**
  - APi Group — Public — global fire/life safety services platform, multi-brand consolidator
  - Pye-Barker Fire & Safety — PE-Platform — Leonard Green-backed national fire/life safety roll-up
  - Cintas Fire Protection — Public — fire protection division of facility services platform
  - Summit Fire & Security — PE-Platform — SK Capital-backed regional consolidator
- **Trade orgs & conferences:**
  - National Fire Sprinkler Association (NFSA) — nfsa.org — contractor membership directory — exhibitor_list: yes
  - American Fire Sprinkler Association (AFSA) — firesprinkler.org — open-shop contractor body — exhibitor_list: yes
  - NFPA Conference & Expo — nfpa.org/conference — annual industry conference — exhibitor_list: yes

The verdict rationale cites screen categories by name (Entry Barriers, Non-Discretionary Demand, Substitution Risk, Capex Profile) rather than describing them; the thesis is prose with no framework citations.

Tools

• view — Required at Step 2 to load reference/overland-industry-attractiveness-screen.md. Reread sections as needed when scoring rationale needs to map back to a specific category.
• web_search — Used at Steps 4, 5, 7 for sub-vertical economics, cascade anchor validation, and trade org / conference identification. Run multiple queries in parallel when calls are independent.
• PitchBook (MCP, if connected) — Used at Step 5 to validate PE-backed platform sponsorship, fund identity, and recent transaction activity. Use pitchbook_search to resolve company entities, then pitchbook_get_profile for ownership detail. If PitchBook is not connected, fall back to web search and flag platform identity confidence in the cascade anchor descriptor.
• create_file — Step 10, for writing the output markdown to outputs/p1_sector-research_<slug>.md.

Constraints

• Do not reproduce the Overland industry attractiveness screen, FCF decomposition sequence, or Porter/Mauboussin citations in output. Reference categories by name only.
• Do not score sub-verticals from memory or training data alone. Validate Tier 1 platform identity, recent transaction activity, and trade org existence via web search or PitchBook before naming.
• Cascade anchors must be currently operating businesses.
• Geographic scope defaults to US. If the user specifies otherwise or the sub-vertical is geographically concentrated, populate Scope caveat.
• Output is the structured markdown file only. Do not produce a posting memo, deal screen, or borrower-level analysis — those are downstream phases.

Downstream contract

The downstream borrower-identification phase consumes the schema above intact — sub-vertical names, NAICS codes, cascade anchor names, and trade org / conference names are used directly as query seeds. Do not modify field names or section headers.

Classification & review state

The sector screen is internal sourcing intelligence (INTERNAL/CONFIDENTIAL) and is never co-lender- or LP-facing. It is not an IC-, legal-, co-lender-, or AM-facing artifact, so no [DRAFT — HUMAN REVIEW REQUIRED] banner is added — the output is still a draft pending human review: the sourcing analyst validates the screen verdicts and cascade anchors before the handoff feeds downstream borrower identification. A reviewer approves — never approve, finalize, or issue the handoff. Use [INSUFFICIENT DATA — <what is missing>] for any unresolvable field — never fabricate.

Runtime

This skill needs no Python scripts — it produces a markdown handoff. In Claude Desktop or claude.ai, write the output file at Step 10; in a chat or Office add-in without file creation, emit the same markdown in the conversation for the user to save. Web search is required for scoring; PitchBook is optional (fall back to web search and flag platform-identity confidence when it is not connected). Use fully-qualified Server:tool names for MCP calls; confirm the server name at first invocation.