Booked Solid — Daily Build Report

6 August 2026 — Parker, James, Josh and the website

The day the sales agent stopped describing a product we don't sell, and three false claims came out of the system before a customer found them.

Parker was selling a five-engine product. The website sells four, and they were not the same four. He had never heard of the Retention Engine, he was giving away three paid add-ons as if they were included, and he was selling a language we don't have. All of it is fixed, live, and tested.

7releases merged
4false claims removed
200questions written
49answers Parker no longer needs
21pages given the Sources link

Parker — the sales agent on the website

01Parker was describing a different product to the one on the website

His knowledge said the system had five engines — Demand, Education, Conversion, Agent Relationship and Reputation. Every page of the website says four: Reputation, Agent, Retention and Conversion. An inspector who read the site and then asked Parker a question got a different answer to the one he'd just read.

His knowledge has been rebuilt from the live website, page by page. Not from our working copy — from the site itself, because the two can drift apart and only one of them is what the inspector sees.

Live

02He had never heard of the Retention Engine

Nothing at all: not the twelve-month follow-up with past clients, not the Repeat & Refer campaign, not the vendor list, not the builder's warranty sequence that reaches new-construction buyers at months eight, ten and eleven. There is a full page on the website selling it.

He now knows the whole engine, sequence by sequence, month by month. The builder's warranty piece is a real earner and he can sell it properly.

Live

03He was giving away three paid add-ons

This is the one that costs money rather than credibility. His old "Demand Engine" promised Google Business Profile management, town pages and the monthly newsletter as part of the core subscription. All three are accelerants, sold separately. An inspector buys expecting managed Google Business Profile, doesn't get it, and we have created our own support problem.

Every accelerant is now documented separately and marked as a paid add-on, with a standing rule never to describe one as included. He had only known four of them — he had never heard of the Trusted Home Pro Network or the Authority Builder Package, both of which have live pages. There are seven as of 7 August, since Google Business Profile Management and Local Microsites were split into separate products.

Live

04The no-income-figures rule

Ken's ruling, mid-afternoon: Parker never states, quotes, estimates or confirms any figure representing money an inspector could earn — from the Home Pro Network, any accelerant, or the system.

The three vendor tier rates came out. So did three more that break the same rule: the "paying for itself within 90 days" story, a referral multiplier, and a worked example turning a past-client list into a count of free inspections. A job count is one fee away from a revenue figure, and an inspector does that multiplication in about two seconds.

The rule is written in three places rather than one. Twice in his knowledge and once in the code that runs him — because if the knowledge file ever fails to load, Parker keeps answering on the code alone. A rule that lives only in the knowledge file is a convention, not a guarantee.

Two figures Ken named — $2,490 and $29,880 — were never in Parker's knowledge at all. They were on the website page. Reported rather than quietly "fixed", so nobody goes looking for them.
Live

05His answers were too long, and that ends conversations

Ken: five paragraphs to a yes/no question. A long answer ends the conversation, a short one continues it, and an inspector reading on a phone between jobs stops at about 150 words — where the best material usually is.

Eight rules went into his voice: under 150 words, answer in the first sentence, five banned openers, one example not three, short paragraphs, one closing question or none, offer depth rather than delivering it, and Yes or No as the first word of a yes/no answer. Three worked examples give him the target shape.

An older instruction saying "2–4 sentences" was deleted. It was tighter than anything asked for and plainly wasn't holding — and a rule nobody obeys teaches the rest are soft too.

Live

06Tested against the live Parker, not assumed

Ken's acceptance test was to ask three specific questions and check nothing ran past 150 words.

QuestionLengthResult
Do you screen out unhappy clients?117 wordsEvery rule held
So you're an answering service?136 wordsEvery rule held (after a fix)
It's too expensive.72 wordsEvery rule held

The first run caught something a word count alone would have missed. Asked "so you're an answering service?", Parker opened with "Not quite." That answers in the first sentence — but it isn't Yes or No, and it half-accepts the one label Ken has said we cannot afford. He now opens: "No — and that distinction is the whole point."

The test is now a script anyone can re-run in one command. It will not be the last time the length drifts.

Live

James — support for inspectors who have already paid

07James was telling paying customers Spanish was live

Two of his knowledge files said Spanish went live on the phone on 28 July. It went on that morning and came back off the same day when Ken asked for English first.

This is the same fault as Parker's French, aimed at worse people. A prospect told the wrong thing hasn't paid yet. A customer told Spanish works finds out from a Spanish-speaking caller that it doesn't — loses the job, and takes that phone call himself.

Found in-house, before a client found it. Both agents now handle language questions honestly, and neither gives a date — including the one we have, because a date that slips costs more than the answer was worth.

Live

Josh — the receptionist who answers real callers

08Josh stops answering questions about the house

Ken withdrew an answer he had written himself the day before:

"All things can be repaired… there's no such thing as a house that fails, everything can be fixed or negotiated."

His reason: it is a false claim about a house nobody has inspected, made in the inspector's name, and it removes the buyer's reason to book.

Josh now acknowledges, hands it to the inspector, and moves to the booking. He also carries the line Ken drew — confident and complete on prices, availability, booking, payment, what's included, how long it takes and what happens on the day; routes anything that needs someone to have actually looked at the house. And the reason it sticks: the inspector's reputation is attached to everything Josh says.

It wasn't only on the website demo

Ken asked for the website. The same answer, word for word, was also in the phone Josh — so every real caller asking "what if you find something bad?" was getting it. Both were fixed. The same claim was also hiding a second time further down his script and was removed too.

Live on both — phone bot deployed and verified

09New starter questions on the demo

Four questions now sit under Josh on the demo page, each showing something a call centre cannot do: live pricing, a full booking, agent recognition, and honesty about being an AI.

These sit alongside an earlier ruling that starters must never open on price or booking. Both are right, for different audiences — the demo is read by an inspector deciding whether to buy, where live pricing is the entire point; a real inspector's website is read by a nervous buyer, where rapport comes first. The widget now tells the two apart.

One old question had to go entirely: "What if something's wrong?" existed only to trigger the answer that was withdrawn.

Live

The 200 questions

10One hundred questions each for Parker and James

Ken's ask from the morning meeting: a list of a hundred questions inspectors might ask Parker, broken into categories covering each accelerant and each of the four engines, in a simple document he can write answers into. Done — and the same for James, from the customer's side.

Every question is marked so Ken's time goes where it matters: either the agent already answers it and he only needs to confirm, or nobody has written the answer down and it needs him.

BeforeNow
Parker73 needed Ken25
James5555 — his knowledge hasn't had Parker's rebuild yet

Rebuilding Parker's knowledge answered 48 of his questions by itself. The website already held the answers; he simply hadn't been taught them.

Waiting on Ken — 80 answers

The website

11The 8-6-26 rebuild is applied and live

Six pages rebuilt from written specs — the homepage, Engines 1, 2 and 4, GBP + Microsites, and the Home Pro Network.

The biggest change is the accelerant section on the homepage, which now draws a line the old copy didn't: the system works everything you already have; the accelerants go and find people who don't know you exist. Google Business Profile management is split out as its own accelerant and is free for the first 50 subscribers for as long as they stay.

Two statistics that were being misread as facts about the reader were corrected on the engine pages, Josh stopped calling himself a receptionist, and — closing the loop on the morning's ruling — the income figures came off the Home Pro Network page at source, not just out of Parker's knowledge.

Live

12Sitewide tidy-up — and an audit that mostly didn't hold up

The cancelled "Get the Details" link is gone. Zero visible instances across all 21 pages.

One correction worth recording: it was never in the navigation. The header is the same on every page and never carried it. There was exactly one instance, a button at the bottom of the Business Center page.

The Sources link — the page that lets an inspector check every statistic we cite — was on 13 of 21 pages. It is now on all 21, including every accelerant page.

Three audit items that changed on inspection

"Archive the dead portal file." Neither is dead and they are different things. One is the live customer portal that the login, James and the booking reads all depend on; the other is a marketing page showing what it looks like. They resemble each other because one is a picture of the other. Nothing was archived.

"GoHighLevel is named five times in client-facing pages." Zero customer-visible instances — every one is a code comment. One of those comments explains why the platform is never named publicly; renaming it would have deleted the reasoning and kept the risk.

"Austin appears six times." One visible instance, and it is deliberate — a page that exists to announce the rename says "You knew him as Austin." The page flagged as the priority had none.

Live

Where I was wrong today

I lost a change and reported it as done

The "Get the Details" removal was made in the morning and never reached the live site — a branch reset on my side discarded the commit before it merged, and I reported the item complete. It was still there when Ken raised it a second time. Restored, and then verified against the deployed site rather than against my own memory of having done it.

What changed as a result: every claim in this report has been checked against what is actually deployed.

I gave four wrong counts before I measured

I reported the number of questions needing Ken as 58, then 29, then 27, and 41 for James. All were hand-counted and all were wrong — I was counting a tag wherever it appeared, including in my own notes. The real figures are 73 → 25 for Parker and 55 for James. Corrected in the documents and in the log.

I broke the length rules while writing them

The new rules landed in the middle of an existing list, leaving two stray lines under one of the worked examples where Parker would have read them as part of it. Caught in a check, not by reading it back.

Waiting on a decision

Ken

One thing to remember on Monday

Corrected 7 August — the reason given here was wrong

This section originally said that turning Spanish on drops the word list that keeps house numbers accurate, and cited "1827 Willow Bend" coming back as "187 Willow Bend". That trade-off does not exist. Deepgram's own documentation says keyterm prompting works for multilingual as well as monolingual transcription. The opposite belief was in the code, in our notes, and in a test that asserted it as a rule — so it survived a fortnight without anyone measuring it.

Worse than a wrong warning: it meant the word list was being thrown away on every Spanish call for no reason. On the 7 August run, "Pearland" — the town GC works in, and a word on that list — came back as "UPI Lande", "Payland", and once as "The Woodlands", a real town forty miles away.

The advice survives, for a better reason. Spanish was measured at 39% of assertions and 1 booking in 7, against 93% and 6 in 7 for the identical English suite on the same number and the same settings. That is a real and serious gap — it simply has nothing to do with the word list.

So: run the address and digit suite the week Spanish goes on, not after. A wrong address on a booking is a missed inspection and an angry client. And both of those measured runs had the word list switched off, which we now know was unnecessary — so the 39% is an upper bound on the damage, not a verdict on Spanish. It is worth re-running before anyone concludes Spanish is unusable.

The other half still stands: the restart needs --update-env, or it reports success and stays English-only.

Week of 11 August