The story of TradesAd, told in days. Day 1 is September 29, 2026, the day the current product was born; everything before that is the prologue. Trials, fixes, failures, endurance, focus, discipline, written from the daily work record while it was happening, for the builders who come after. Every entry ends with the rule the day left behind.
Before Day 1. Early September 2026.
TradesAd existed before it had a birthday. I built the first version inside Google AI Studio, an AI powered checker called TradesAd Audit Engine, and it went live the way a lot of first builds go live: fast, hopeful, and held together by prompts. Google switched on its business profile on September 7.
When we finally audited the site properly on September 25, the evidence was embarrassing in the most useful way. The browser tab still carried the default title “My Google AI Studio App.” The same product wore three different names on its own pages: TradesAd Suite, TradesAd Diagnostics, TradesAd Audit Engine. The support email pointed at a misspelled domain. The hero promised “clinical precision” in the language of a laboratory, to an audience of people who sell plumbing ads.
None of that was the real problem. The real problem was underneath: the checker itself was a model improvising verdicts, and improvisation is a strange foundation for a tool whose entire job is to be right.
The rule it left behind: a product is not real when it launches. It is real when it can survive being audited by its own standard.
Tuesday, September 29, 2026.
This morning the tool sat an exam and failed it. We fed it eight ads chosen to be difficult, the kind with the violation buried in the third sentence, and it judged four of them wrong. Worse than the misses was the manner of them: it blessed risky ads with words like “ready to launch” and “excellent,” the two things a compliance checker must never say. A checker that blesses is worse than no checker at all, because it lends its confidence to someone else’s money.
I asked the question out loud: is this tool worth continuing at all? It felt risky. It felt like it would never come close to accurate.
By evening we had made the hardest call a builder makes. We stopped trying to fix the brain and replaced it. No model decides anything in TradesAd anymore. The product was rebuilt in a single day as fixed written rules: a library of patterns, each one tied to a named policy, a scanner that applies them the same way every time, and three verdicts only: HIGH RISK, NEEDS REVIEW, CLEAR. The words “ready to launch” and “excellent” were banned from the building. The new engine passed its 18 case regression exam 18 out of 18 before the day ended.
TradesAd was not improved on Day 1. It was replaced, and the replacement is the product whose age we now count.
The rule it left behind: when the judge cannot be trusted, do not retrain the judge. Write down the law and let the law decide.
Monday, October 5, 2026.
Every product needs a quality bar, and most bars are useless because they sound like “zero defects,” which no honest builder believes. Ours came out of a long audit day, the eleventh deep audit of the product, which found no blockers and one finding that still stings: a progress report that was supposed to carry an agency’s own brand went out stamped with our sample company’s name instead. The feature worked. The promise broke.
So the bar got set in plain words, and it has governed everything since: no embarrassment. Not zero errors. Our competitors are not error free either, and pretending otherwise is its own kind of dishonesty. The bar is that nothing we ship should make a careful user feel fooled, and nothing should make us ashamed to have our name read next to it.
The bar comes with machinery, and the machinery matters more than the slogan. Every report carries an “Audited as” stamp. Every uncertain case falls into a NEEDS REVIEW bucket instead of being rounded up to good news. The pre-screen disclaimer stays prominent, because the day we bury it to make reports look cleaner, the bar breaks silently.
The rule it left behind: a standard you can say in three words is a standard you can actually hold at midnight. Ours is: no embarrassment.
Monday, October 12, 2026.
A product can be live on its own website and still be invisible. Today TradesAd opened a second front door: GoHighLevel approved the public Marketplace listing, and the suite took its place inside the platform agencies already open every morning. The listing carried the honest numbers too: 4 installs, $0 earnings. Not a victory lap. A shelf in the right store.
The work was not glamorous: portal fields the automation could not type, screenshots, keywords, a reviewer test path that had to work without a paid plan. Three days later the listing grew again, version 2.2.0, with the Local SEO Audit inside it and menu links added by hand in the agency sidebar. But distribution is often just that: paperwork done carefully enough that strangers can find you.
Two weeks later, on Day 31, the listing was approved again as version 2.3.0, with the Rejected Ad Fix named in the description and the support email corrected to the inbox that actually receives mail. The door did not just open. It learned to describe what was behind it.
The rule it left behind: build the product, then build the doors. A great tool in an empty room is still an empty room.
Tuesday, October 20, 2026.
Today the diary gets an entry about something that did not happen.
Eleven five star reviews arrived in a neat file, with names, agencies, and the kind of praise every young product dreams of. People thanking the tool for saving accounts, for catching what their teams missed. All we had to do was upload them, and the empty reviews section would stop looking empty.
We checked them against the record first. Our own user store, verified four days earlier, showed zero outside users. Not one of the eleven names existed anywhere in our work. The reviews described suspensions saved this year and dozens of accounts run through a tool that, at that point, almost nobody outside this desk had touched. They were impossible facts wearing five stars.
So nothing was uploaded. The reviews section stayed honest and empty. There is also a legal shadow here, fake reviews carry real fines for a real company, but the legal reason is not the one that decided it. The deciding reason is simpler: a tool that sells trust cannot buy its own. The first real testimonial will arrive the slow way, from someone who actually ran an audit, and it will be worth more than eleven inventions.
The rule it left behind: social proof you manufacture is not proof. It is a debt, and it always comes due in public.
Thursday, October 22, 2026.
For two days, our email went into the world unsigned.
Buried in the domain settings was a malformed signing key, one character too long, and because of it the cryptographic signature that tells Gmail “this really is tradesad.com” was broken. The timing could not have been worse: our first real outreach wave had gone out the day before, ten emails to ten agency founders, sent while the signature was already broken. Some of them very likely landed in spam, and we will never know which ones. Two automated attempts to fix the record failed. On the third try, side by side, one of us signing in and the other driving and verifying character by character, the record was replaced with the exact value: 410 characters, not 411.
That evening Google’s status flipped to “Authenticating email with DKIM,” and a test message came back signed by tradesad.com. A week later we proved the signature on a real received email, and the saga was closed for good.
One wrong character, invisible to every reader, quietly taxing the most important launch we had. Deliverability is not a marketing detail. It is whether your voice arrives at all.
The rule it left behind: verify the invisible things before the visible launch. A signature nobody sees can silence everything everybody was supposed to read.
Sunday, October 25, 2026.
Today we gave the main tool away.
The ad audit had always sat behind a signup, the way SaaS products are taught to behave: make them register, count the run, guard the value. This morning we took the wall down completely. Anyone can now run the audit, no account, no limit, and they get the complete report, not a teaser. The paid plans keep their ground where the ground is real: white label reports, the tools that cost us money to run, the monitoring. But the check itself, the thing a worried media buyer needs at 11 PM before a launch, is free.
The reasoning was not generosity. It was arithmetic. A checker nobody has tried is a claim. A checker someone has run on their own worst ad is evidence, and evidence is what we sell. The audit costs us nothing to run, it executes in the visitor’s own browser, so every wall in front of it was pure fear dressed as strategy.
The same evening, the whole site fell over. Every one of our sites returned errors for hours because hosting credits had been bought onto the wrong team account and the right one ran dry. We fixed it that night, but the diary should keep the pairing: the day we removed the wall was also the day the building went dark for a while. Conviction and chaos are frequent roommates in month one.
The rule it left behind: if a thing costs you nothing to give and everything to hide, give it. Walls should guard costs, not fear.
Wednesday, October 28, 2026.
Today we built the feature I had wanted for weeks, the Rejected Ad Fix: paste in an ad that Google or Meta rejected, tell it the reason they gave, and get the flagged words found, fixed, and rechecked before your eyes. It passed every test battery we own. Three hundred and fifty checks, green. We deployed it, and it was dead.
Not degraded. Dead. The page loaded, the dropdown stood empty, the button did nothing. The cause, when we found it, was one wrong assumption: the page’s code reached for the rule library through a window it does not live behind, and the whole page died quietly at startup. Every automated test had passed because every test exercised the logic and none of them had ever truly opened the page and clicked.
What saved the launch was the least scalable tool we have: a human style click test in a real browser, the same evening, before any stranger arrived. We fixed the references, wrote a new permanent wiring test that opens the real page machinery and drives the full flow, and clicked it again end to end: a risky ad scored 72, the fix went in, the recheck climbed to 92, and the report unlocked only after the recheck, exactly as designed.
The feature went live twice that night. Only the second launch counted.
The rule it left behind: a test that never opens the page is a rumor about the page. Ship nothing you have not clicked yourself.
Thursday, October 29, 2026.
A tool that judges other people’s work should expect to be judged. Today we built the judge a judge.
We wrote sixteen realistic contractor ads, the kind agencies actually run, some clean, some carrying the exact claims that get accounts in trouble, and we ran them through the audit and the fixer as a standing quality exam. The first sitting: 51 checks passed and 2 genuine gaps stood in the open. An ad promising “Guaranteed Leak Free Roofs” had slipped past the guarantee rules and come back with a gentle NEEDS REVIEW and a score of 92. A housing ad steering people by religion, “walk to churches,” had come back completely CLEAR.
Both gaps were in the library, not the engine: two patterns drawn too narrowly for the real world’s wording. We widened them the same day, bumped the policy version the way the maintenance contract demands, redeployed, and verified against the live site itself, not our local copy. The guarantee ad now lands at HIGH RISK, 60. The housing ad lands at HIGH RISK, 80. The exam now reads 56 out of 56, and it re-runs after every change to the library, forever.
This is the part of building nobody photographs. Not the launch, the audit of the audit.
The rule it left behind: your quality gate must use the real world’s words, not the words you wished the world used. Test with the ads that actually exist.
Friday, October 30, 2026.
TradesAd is one month old today, and the diary should record the month the way it actually stands, not the way a launch post would tell it.
The honest numbers: four accounts exist and all of them are ours. Revenue from strangers is zero. A stranger has run our new fixer exactly never; its only recorded run is our own launch test. And yet the month was not empty. An AI answer engine, asked the exact question our buyers ask, named this tool as its only recommendation and cited our site first, even while quoting a wrong pattern count it had picked up from our own stale pages. We fixed that count at every source this week: 82 patterns, everywhere, no exceptions. One practitioner, a creative partner who sells proof to home service brands, asked to run his own work through the checker, the first time anyone outside this desk has asked to be judged by it.
Today the month closed with unglamorous, necessary work. Two messages went out through the contact forms of the only two specialist agencies in our region built exactly for our buyers, because neither publishes an email address and we do not guess email addresses. Two old outreach sequences were closed with honest breakup emails. An automated welcome email arrived within one minute of one submission, and for one second it looked like a reply. It was a machine. The diary records that too, because month two will have real replies, and we intend to know the difference.
One month old. No revenue, no audience, no shortcuts taken. A tool that works, a bar that holds, and a story that is true.
The rule it left behind: count what is real, say it plainly, and keep building. Endurance is a strategy when the product is honest.
Season 1 ends here. Season 2 begins with whatever November brings: the follow ups, the first stranger, the first dollar, in whatever order they arrive.
The ad audit at the center of this story is free, ungated, and runs in your browser. Same ad in, same verdict out, every time.
Run a free audit