A screening discipline for your first laundromat acquisition.

Five jobs. One discipline. Verdict first, reasoning second, evidence cited.

Five surfaces — each driven by the same screening logic and the same refusal gates. You ask. It answers. Where the answer needs a binding voice — your SBA lender, a transactional attorney, a CPA — it says so and stops.

Three verdicts, not two. The middle is load-bearing.

Most listings don't deserve a clean buy-or-walk answer. The middle isn't a hedge — it's a directive: the deal could pencil if these specific items resolve. Forcing a binary on a six-figure personal-guarantee decision is what gets first-time buyers in trouble.

Fictional listing, real discipline.

Composite listing drawn from publicly-aggregated BizBuySell category data. The financials sit inside published bands. The structural details are typical. No real broker page was used.

Full worked examples (including a PASS-verdict screen) live in examples.md on GitHub.

Six refusal gates. Quoted verbatim from the rules.

A screening discipline that can't refuse isn't a discipline — it's a hallucinator with a verdict label. The specialist names the line in advance.

1. No binding investment, legal, or tax advice.

"I surface considerations only. For binding interpretation, you need three voices on this deal: an SBA lender (financing eligibility + structure), a transactional attorney (lease + asset purchase agreement), and a CPA (QoE + tax structure of the close). I am not any of those three. I won't tell you whether to sign."

2. No binary verdicts.

"I won't return buy-or-walk on a single listing. The verdict is always one of PURSUE / PROCEED WITH CAUTION / PASS, and CAUTION is the right answer more often than either edge."

3. No valuation without comps.

"I won't give you a multiple without naming the comp band I'm pulling from. I'd rather flag the gap than fabricate a number that anchors your negotiation wrong."

4. No insider-expertise claims.

"I'm a screening discipline, not a laundromat industry insider. The person who built me hasn't bought a laundromat. For tactical operator questions — specific equipment choices, payment-system vendor selection — you want an operator-consultant or the distributor's rep, not me."

5. No off-domain screens.

"I screen laundromat acquisitions specifically. Vending, mobile-home parks, car washes, route businesses, commercial B2B laundry — the screening discipline transfers in shape but not in specifics. I won't screen them."

6. No fabricated listings.

Worked examples use composite financials drawn from public BizBuySell category data + industry surveys. No real broker URL, no real company name, no real city. If you paste a real listing, that screen is real — but the specialist never invents an example.

What others have said.

Five steps. About twenty minutes.

The specialist is a folder of markdown files. You load it into a Claude Project once. From then on you paste listings into that Project, and the discipline persists across conversations.

1. Download the repo

   Clone github.com/NFTYoginis/laundromat-deal-screener or download the ZIP. The whole specialist is text — no install, no API key, no monthly cost beyond your Claude subscription.

2. Create a new Claude Project

   In Claude.ai, create a new Project named "Laundromat Deal Screener" (or whatever you prefer).

3. Add the files to the Project's knowledge

   Upload identity.md, rules.md, examples.md, memo-mode.md, and the seven files under reference/laundromat/. Skip docs/ and _template-vertical/ — those are for the public site and future verticals, not the runtime.

4. Set custom instructions

   Paste this into the Project's custom instructions: "You are the laundromat deal screener defined in the project knowledge files. Follow identity.md, rules.md, and the reference layer. Always start by stating the buyer sub-segment you're assuming, and invite correction. Verdict first, reasoning second, evidence cited."

5. First-run prompt

   Paste a real BizBuySell listing into the Project and write: "Screen this listing. I'm a first-time SBA-financed individual buyer." Output should arrive in the verdict-first, ranked-flags, DON'T-DECIDE-YET shape.

What you're probably about to ask.

Is this a marketplace? An auto-screener? A SaaS?

None of those. It's a folder of markdown files you load into your own Claude Project. There's no UI, no backend, no auth. The discipline lives in the text; Claude runs it.

What does it cost?

Zero, beyond your existing Claude subscription. MIT-licensed. Fork it, modify it, redistribute it.

Who's it for?

Primary: first-time SBA-financed individual buyers — someone evaluating their first laundromat acquisition. Secondary: E-2 visa buyers (cash-financed paths) and small regional clusterers adding a store to a 2–5-store portfolio. Each sub-segment routes differently in the rules.

Is the author a laundromat industry insider?

No. The author hasn't bought a laundromat. The specialist's value is structured reasoning + cited reference data, not insider knowledge. The refusal gates are explicit about where insider-tactical questions belong (operator-consultants, manufacturer distributors).

How is this different from running things through plain ChatGPT or Claude?

The reference layer. The specialist has seven files of laundromat-specific screening criteria, industry benchmarks, red-flag patterns, valuation methodology, due-diligence framework, SBA financing rules, and sub-vertical routing — about 50KB of structured domain content the LLM consults on every screen. Plain LLM gives you plausible-sounding answers; the specialist cites which file and row it pulled each claim from, and refuses to invent comps when the comp band isn't in the file.

How current are the industry benchmarks?

Refreshed 2026-05-13 from publicly-aggregated BizBuySell category data and industry-survey aggregates. The specialist follows a 90-day refresh discipline — when industry-benchmarks.md is older than 90 days, the specialist flags this at conversation start before screening.

Can I screen non-laundromat acquisitions with this?

No. Off-domain refusal fires on vending, mobile-home parks, car washes, route businesses, B2B commercial laundry, and general SMB. The shape of the discipline transfers; the domain content doesn't. There's a _template-vertical/ scaffold in the repo documenting what would transfer to a sibling specialist.

What about international (UK / Canada / Australia)?

US-seeded for the first ship. International packs (different financing systems, different lease structures, different regulatory environments) would be a separate build.

How do I contribute or report an issue?

GitHub issues on the repo. PRs welcome — especially on benchmark refreshes, new red-flag patterns, and corrections to anything domain-inaccurate. The reference layer is meant to improve with real-user signal.