Canonical original: https://aistacklab2026.blogspot.com/2026/11/plumbing-business-startup-cost-i.html

The best documentation products in the trades share one property that almost no marketing page mentions: they are falsifiable. You can open the file, take the claim apart, and find out whether it is true before you spend anything. So when a $1 plumbing business kit claimed a startup-cost model covering one van to five, I did the only thing worth doing with a claim like that — I extracted the zip and rebuilt the model in Python from its own rows.

Three files did most of the work: 03_Startup_Cost_Model.csv (19 line items), 11_Job_and_Route_Log.csv (5 seeded jobs), and self_qos.py (a standard-library checker, imports csv, decimal, os, sys). Here is what survived.

A cost model is a spec, and a spec must be executable

The startup model is deliberately split into two scales. Nine rows are base — counted once no matter how many vans you run. Ten rows are per-truck — multiplied by the van count. That single decision is what makes the ladder non-linear, and it is the sort of thing a PowerPoint business plan quietly gets wrong.

Rebuilt:

base = 12,230.00 low / 38,400.00 high (9 rows, counted once) per-truck = 15,350.00 low / 54,100.00 high (10 rows, each) 1 van = 27,580.00 -> 92,500.00 2 vans = 42,930.00 -> 146,600.00 3 vans = 58,280.00 -> 200,700.00 4 vans = 73,630.00 -> 254,800.00 5 vans = 88,980.00 -> 308,900.00 

Ten of ten cells footed to the cent against the shipped table. That is a low bar — addition — but it is the bar most of these products fail, because the table was typed by hand after the engine was written.

Two properties of that table are worth more than the numbers.

The band is 3.35× wide. 92,500 / 27,580 = 3.354. Any article quoting "about 28 grand" has deleted two thirds of the model's uncertainty. When a range is that wide, the width is the information: it tells you which rows to go price first in your own market.

About half of the low-end figure is working capital, not capex. The largest single base row is a 60-day reserve: 6,500.00 low, 18,000.00 high. That is 53% of the base low subtotal and 24% of the entire one-van low row. So "startup cost" in this file is two questions stapled together — what you must buy once, and what must sit in the account while the first deposits clear. Splitting them changes what you ask a lender for, and it changes how long you can survive a slow month.

The price engine, re-derived from its inputs

Five calls are frozen end to end. I threw the totals away and repriced each from its inputs through the nine-line engine (book hours × 145.00, parts × 0.45 markup, 69.00 service call, 0.50 after-hours on its own line, permit pass-through):

def engine(hours, parts, ah, trip, permit): raw = round2(hours * 145.00) sell = round2(parts + round2(parts * 0.45)) base = round2(raw + sell) job = round2(base + round2(base * 0.50 * ah)) return round2(job + round2(69.00 * trip) + permit) 

5 / 5 exact, including margins (total minus job cost). The interesting one is the least obvious call, J-002: 3.00 hours, 620.00 of parts, a 95.00 permit, service call credited. 435.00 + 899.00 + 95.00 = 1,429.00. The one that would have hidden a composition bug is J-004, where the after-hours multiplier and the service call both apply: 1,145.50 base, 572.75 after-hours, +69.00 trip, permit 0.00, total 1,787.25.

The labor rate is a chain of seven lines, not a market guess: wage 34.00 → payroll load 9.52 → loaded tech 43.52 → 18,278.40/month → cost hour 60.93 → T&M 87.04 → 174.08 minimum. The cost hour being 60.93 rather than 43.52 is not a typo; it absorbs the paid-but-not-billed hours over a paid-to-on-job ratio of exactly 420/300 = 1.40. Replace the wage with your own and the chain follows.

Rounding a break-even up is the correct defect to have

Fixed costs 18,278.40 labor + 5,255.00 overhead = 23,533.40. Contribution per call 447.96. That is 52.53 calls.

The kit does not round that to 52. It rounds it to 53 whole calls and prints both sides of the line: at 52 calls the month nets −239.48, at 53 calls +208.48. Round-half-up on a break-even would have put the reader one call short of profitable and looked entirely innocent. That is the kind of choice you can only check by doing the arithmetic yourself.

Same story on the monthly P&L: 92,675.00 − 57,973.57 = 34,701.43, which is 37.44% of revenue. And the job log seeds five rows totalling 6,741.25 revenue / 3,306.74 cost / 3,434.51 margin over 17.50 book hours — a blended 50.95% job margin.

That last pair is the trap worth naming out loud: job margin is contribution before the month's overhead. 50.95% on the log and 37.44% on the month are both correct, and comparing them to each other is exactly how a busy week gets mistaken for a profitable month. 385.21 of revenue per book hour is not profit per hour.

The checker is the part I would pay for

$ python3 self_qos.py SELF-QOS PASS - manifest 17/17 files. SELF-QOS PASS - tech build-up 9.52/43.52/18278.40/60.93/87.04/174.08/0.62. SELF-QOS PASS - engine 5/5 calls 282.75 1429.00 554.75 1787.25 2687.50. SELF-QOS PASS - P&L footing 92675.00 - 57973.57 = 34701.43. SELF-QOS PASS - startup ladder 27580.00/92500.00 42930.00/146600.00 ... SELF-QOS PASS - log 5/5 rows, totals 6741.25/3306.74/3434.51. SELF-QOS PASS - honesty sweep clean: no banned tokens, no sibling pricing, no URLs. SELF-QOS PASS 

Exit 0, eight lines. Zero external URLs anywhere in the bundle, no pip install, no service to authenticate against — which is why the whole thing is auditable in an afternoon instead of a weekend.

The honest limits of that checker, since I am the one recommending it: it validates the shipped numbers. Edit the startup CSV or the job log to your own market and it will, correctly, report your numbers and may print a fail line. That is the check doing its job, not breaking. Keep an untouched copy if you want to re-run the full gate later.

What this is not

  • Not software. No app, no CRM or map or API or invoicing integration. Seventeen plain files (13 Markdown, 2 CSV, a licence, the checker) at 42,042 bytes and 13,322 words of Markdown, measured rather than claimed.
  • Not US-neutral. Dollars, state and local licensing, permit pass-through. The method travels — engine, cost chain, ladder, log — every figure and every compliance item does not.
  • Not one number. The honest output is a 3.35× band plus a method for narrowing it in your own market.
  • Not legal, tax, insurance or licensing advice. The licensing file is 19 checkboxes of questions to put to your own state office; it does not tell you what your state requires.

Every figure above is an example figure to verify locally. Footing means the arithmetic agrees with itself, not that 145.00 a book hour is your market rate.

The product

This is our own kit, disclosed plainly: Plumbing Business Starter Kit — Estimate & Pricing Templates, Startup Cost Model & Operations Docs. The estimate engine, the startup ladder, five worked calls, a one-van P&L with break-even, 7 SOPs, a licensing checklist, a safety checklist, a contract, three estimates, a job log, a pipeline tracker and a 7-day launch plan.

  • Gumroad: https://nandminc.gumroad.com/l/plumbing-kit — $1.00 as loaded from the live page immediately before this was published (intro price on a ladder that steps up as sales accrue; check the page, not this sentence)
  • Licence: PERSONAL USE - NO RESALE — use it in your own business, do not resell or redistribute it
  • Questions: nandminc2020@gmail.com

No ratings, reviews, testimonials or sales figures appear in this post, because I have none that would be honest to quote.

The full measured write-up, with the tables and the row-by-row rebuild, is on AI Stack Lab: https://aistacklab2026.blogspot.com/2026/11/plumbing-business-startup-cost-i.html