A free Overyield tool. Everything you need to sell a $600 screening sprint to an acquisition searcher: the intake that freezes their buy box, the source register, the extraction schema and worker prompt, the deterministic rejection rules, the frontier reviewer prompt, the human-check protocol, the shortlist memo, the broker-question bank, the delivery checklist, and the outreach note that lands the first buyer.

You do not need code. You need careful reading, spreadsheet logic, a frontier chat subscription (Claude or ChatGPT) plus access to one cheap worker model (or about $20 of API credit total), and a Google Drive folder. The first sprint takes about 10 hours across a week. Every sprint after that gets faster.

A fixed-scope screening service for self-funded acquisition searchers, priced $200 for a 15-listing pilot and $600 for a 50-listing sprint, paid upfront.

The window it was built in: the week of July 6-10, 2026, repetitive frontier-grade AI work fell to cents. Grok 4.5 posted $0.31 per task where Fable 5 ran $2.75 (AI Daily Brief, Jul 9). Muse Spark 1.1 cleared a full coding benchmark for $0.92 against Fable’s $12.51 (AI Daily Brief, Jul 10). GPT-5.6 Sol landed about one point behind Fable 5 at roughly a third of its cost (The Rundown, Jul 10). The same week, the demand side wrote its number down: experienced acquisition buyers evaluate 100 or more deals before buying one, and one searcher signed 200+ NDAs before winning a seven-figure deal (Contrarian Thinking, Jul 9).

The service sells the collision of those two facts. The searcher needs disciplined evaluations in volume. The cost of producing a disciplined first pass just collapsed. You deliver 50 consistent deal reps: a ledger showing pass, reject, or unknown for every listing, the exact rule behind every rejection, a source URL and captured evidence for every fact, a human-checked shortlist of three to five candidates, and five unanswered questions for each shortlisted broker.

Three rules before you start:

One folder per engagement: Rep Sprint - [Buyer], containing:

The scope-and-boundary page (paste, fill, deliver as page one):

```
SPRINT SCOPE : [Buyer] : [date]

INCLUDED: screening of [50] public business-for-sale listings against the
attached buy box (01-buy-box.yaml, frozen [date]); a full pass/reject/unknown
ledger with a rule code for every rejection; source URL and captured excerpt
for every extracted fact; a human-checked shortlist of 3-5 candidates; five
broker questions per shortlisted candidate.

EXCLUDED: valuation, financial or legal diligence, broker or seller outreach,
negotiation, offer advice, and any recommendation to buy or not buy.

DATA RULE: public listings only. All figures are seller-supplied and unverified.
A clean screen cannot prove earnings, validate add-backs, or reveal customer
concentration. It enforces your stated first-pass rules and exposes what
remains unknown.

THIS IS NOT investment, legal, tax, or brokerage advice.
```

Start with one listing-dense category in one state or region: HVAC, commercial cleaning, landscaping, vocational schools, med-adjacent services. One lane keeps the buy box tight and makes your demo sample legible to the exact searchers you will pitch.

Where the public listings live: the large business-for-sale marketplaces, regional business-broker sites, and franchise-resale boards. Fifty live listings in one category and one region is a few evenings of collection.

The capture rule: listings change and disappear. For every listing, save the URL, the date, and a text or PDF capture of the page into 02-source-register.csv and the archive folder. Capture pages manually as you browse; do not build a scraper against a marketplace’s terms (legal notes below).

No criterion may be invented after screening begins. Ask the buyer these nine questions, write the answers into 01-buy-box.yaml, and get a written “frozen” confirmation before you touch a listing:

The template:

```yaml
# 01-buy-box.yaml : frozen [date] : confirmed by [buyer] in writing
geography: [“<state or metro>”] # hard boundary: yes/no
asking_price_max: 1500000
cash_flow_min: 200000
cash_flow_basis_accepted: [“SDE”, “EBITDA”] # or one
industries_in: [“service businesses”]
industries_out: [“restaurants”]
models_excluded: [“mandatory real-estate purchase”]
owner_involvement_max: “full-time acceptable”
real_estate: “lease or optional purchase only”
inventory_rule: “included or priced separately; note either way”
missing_data_rule: “unknown queue, never auto-reject, never pass”
```

If the buyer cannot answer questions 2 and 3 with numbers, stop. A vague buy box produces a vague ledger, and you will be blamed for the vagueness. Make them choose before you run anything.

02-source-register.csv, one row per listing:

```
listing_id, source_url, capture_date, archive_filename, category, region, status_at_capture
```

Fifty rows, each with a saved capture. This register is part of the deliverable: it is what makes every downstream claim checkable.

The cheap worker model fills 03-extraction-schema.csv. Schema fields, one row per listing:

```
listing_id, business_name_as_listed, industry, location, asking_price,
disclosed_revenue, disclosed_cash_flow, cash_flow_basis (SDE/EBITDA/not stated),
real_estate (included/leased/purchase_required/not stated),
inventory (included/excluded/valued_at/not stated), owner_hours_as_stated,
years_in_business, employees, reason_for_sale_as_stated, franchise (y/n),
source_excerpt_per_field
```

The worker prompt (verbatim; run per listing or in batches of 10):

Use a cheap near-frontier worker for this pass (the Grok 4.5 / GPT-5.6 Luna / Muse Spark class). At July 2026 pricing this is cents per listing; the whole 50-listing extraction runs low single-digit dollars (per the task pricing published in AI Daily Brief, Jul 8-10). Spot-check the first five rows by hand before running the rest.

04-rejection-rules.xlsx applies the buy box as spreadsheet logic. Every rule gets a reason code:

| Code | Rule | Verdict when triggered |
|---|---|---|
| R-GEO | Location outside the buy-box geography | reject |
| R-PRICE | Asking price above the ceiling | reject |
| R-CF | Disclosed cash flow below the floor (accepted basis only) | reject |
| R-IND | Excluded industry | reject |
| R-RE | Mandatory real-estate purchase where the box excludes it | reject |
| R-OWN | Stated owner involvement above the ceiling | reject |
| R-MODEL | Excluded business model (per the box) | reject |
| U-CF / U-GEO / U-RE / U-* | The relevant field is not stated or conflict | unknown |
| P | All rules pass on stated facts | pass to review |

Two hard rules: a rejection must cite its code AND the quoted excerpt that triggered it, and an unknown can never be promoted to pass by anyone but you, after opening the original page. The rejection trail is not overhead. It is the product: the buyer is paying for reps, and a documented reject is a rep.

Have the expensive model attack the survivors, never approve them. The reason this pass exists: model judges miss broken outputs. Claire Vo’s blind evaluation found her human ranking ran nearly the reverse of the LLM-judge ranking (Lenny’s, Jul 6). So the frontier pass hunts errors; it does not bless candidates.

The reviewer prompt (verbatim; run once over the top 10):

Log every disagreement between the worker and the reviewer in 05-review-queue.csv. Each one is a mandatory human check.

This is the paid part. For every shortlist candidate and every review-queue row:

Budget five to ten careful minutes per checked listing. Do not delegate this step to a model. The buyer can get model output anywhere; they are paying you for a queue a human stands behind.

The question bank. Convert every important unknown into a direct question. The ten starters:

The memo template (one per shortlisted candidate, in 06-shortlist-memos/):

```
SHORTLIST MEMO : [listing_id] : [as-listed title]
Source: [URL] (captured [date]; re-checked live [date])

BUY-BOX FIT
- [criterion]: PASS (“[quoted excerpt]”)
- [repeat for every criterion]

UNKNOWNS (unresolved)
- [field]: not stated / conflict (“[excerpt(s)]”)

NUMBERS AS DISCLOSED (recalculated by hand, [date])
- Asking price: $X (“[excerpt]”)
- Disclosed cash flow: $Y, basis: [SDE/EBITDA] (“[excerpt]”)
- Implied price-to-cash-flow: X/Y = [Z] (our arithmetic on the two disclosed
figures; the figures themselves are seller-supplied and unverified)

FIVE BROKER QUESTIONS
1-5. [from the bank, tailored to this listing’s unknowns]

BOUNDARY: first-pass screening against the buyer’s written criteria, from
public seller-supplied data. Not a valuation, not diligence, not a
recommendation to buy.
```

08-delivery-checklist.md, run before sending:

Deliver ALL 50 rows, rejects included. A buyer who sees the whole trail understands in thirty seconds why this beats a folder of AI summaries.

Build the demo first. Ten listings, one lane, a fictional buy box labeled clearly as a demo. Run the full pipeline: reject the rule-breakers, preserve the unknowns, write one human-checked memo. Put it in a view-only folder with the source register visible.

Where searchers announce themselves: self-funded searchers post their buy boxes in public: on X, in searcher forums, and in SMB-acquisition communities and newsletters' comment sections. You want people who state an industry, a geography, and a price range in writing. They are showing you their frozen criteria for free.

The note (verbatim; send to 15 of them):

One follow-up after three days, then stop. Collect the $200 before adapting anything to their criteria. The pilot’s deliverable is the same structure at 15 listings, delivered in three days.

The buyer: a searcher with a full-time job. Frozen box: service business in one state, asking under $1.5M, disclosed cash flow of at least $200K (SDE or EBITDA), no restaurants, no mandatory real-estate purchase, unknowns to the queue.

Day 1 (1 hour). Intake call. The buyer’s first draft says “good cash flow”; you make them pick $200K. Box frozen in writing. Folder built.

Day 2 (2.5 hours). Fifty listings collected in the lane, each captured with URL and date. Register complete.

Day 3 (1.5 hours). Worker extraction over all 50; you hand-check the first five rows, then let it run. Model spend so far: about $4. Rules sheet fires: 29 rejects on explicit criteria (R-PRICE 11, R-GEO 7, R-CF 6, R-IND 3, R-RE 2), 13 unknowns (mostly U-CF: cash flow simply not stated), 8 passes.

Day 4 (2.5 hours). Frontier reviewer attacks the 8. It flags one thousands-vs-millions extraction error (a $190K cash flow read as $1.9M) and two listings whose revenue and cash-flow figures conflict across paragraphs. You open all 8 original pages, plus the review queue. One listing is gone: to unknown. After checks: 3 solid candidates, 5 pushed to unknown or reject with documented reasons.

Day 5-6 (2 hours). Three memos written, each with recalculated numbers and five broker questions. The corrected $190K listing becomes a clean R-CF reject with the excerpt quoted; the buyer sees exactly why.

Day 7 (30 minutes). Delivery checklist, then the ledger, memos, register, and question sheets go over. The buyer accepts. Invoice: $600, already paid upfront. Total model and tool spend: about $20, roughly 3% of the invoice. Your labor: about 10 hours, most of it the human checks and memos, which is precisely what they paid for.

Numbers are illustrative and deliberately conservative; the model prices cited are the July 8-10, 2026 published figures and will drift. What does not drift: deliver only claims you can trace to a capture.

Then hold the weekly rhythm: one sprint in delivery, one pilot in intake, and the demo folder refreshed monthly so the sample never goes stale.

The free kit runs you one sprint by hand. The Rep Desk, this issue’s Pro upgrade, is the operating layer: the buy-box intake interview that gets vague buyers to commit to real criteria, the worker-and-reviewer harness configs with the per-sprint cost sheet, the pricing ladder that turns one sprint into a $1,500-a-month deal-flow retainer, the searcher-sourcing playbook for where buy-box buyers announce themselves, and the NDA graduation SOP for engagements that move past public listings. It ships to members as a rolling build, intake interview first, and founding members set the order of the rest.

Founding rate: $129 a year, locked forever. Closes when the first 25 members are in; the standard price rises toward $399 as the Vault and the scorecard grow.

This kit is from Overyield: make money with AI, one play at a time. Educational, not financial, legal, or brokerage advice. Listing figures are seller-supplied and unverified; model prices are the July 8-10, 2026 published figures and will drift. Check your own state’s rules before offering services adjacent to business brokerage, and publish only numbers you can defend.