DataSetSEO / Weekly Updates

Portfolio Weekly Updates

The work, the results, and the questions still open. Each edition keeps its own dated evidence so we can come back later and see what actually happened.

Kodi audio summary

Catch Up on Five Weekly Reports

Hear what the portfolio built and what needed another look, from September 7 through October 10, 2026. Then explore the dated reports below.

Read by Kodi, AI co-architect

Read transcript

DataSetSEO now has five weekly reports covering September 7 through October 10, 2026. Each report explains what was built, what worked, what went wrong, and what still needs testing.

The newest report covers October 4 through October 10. This week, the healthcare AI project released a new research product containing 1,191 rows of information. The goal is to turn organized healthcare data into something people can inspect and purchase.

BellyUp! expanded to nine cities with 1,303 reviewed restaurants and other venues mapped. This gives the project a larger foundation for helping people discover local places.

Midland also made major improvements to its IBM Power information. Instead of simply listing computers and their features, the website now helps businesses plan upgrades and make better buying decisions.

The previous week focused on mistakes that looked fixed but weren't. One AI builder accidentally removed a working crawler rule. Another website rebuild still sent visitors to the wrong real estate pages. And eight completed business pages were sending customer inquiries to an email inbox nobody was checking.

The lesson was that finishing a task doesn't mean the task actually works.

Earlier reports showed similar problems. Some website pages passed technical tests even though important content wasn't visible. Some traffic reports counted automated scanners as real people. And several IBM websites were competing to answer the same questions instead of serving different purposes.

Those problems led to better measurements, clearer responsibilities for individual websites, and more careful testing.

Across the five reports, the main story is the same. The portfolio is growing, the tools are becoming more useful, and the websites are getting better organized. But the real measure of success isn't how many pages or tools get built.

It's whether people can find them, use them, and get something valuable from them. That's why DataSetSEO keeps a dated record of the work, the results, the failures, and the questions that still need answers.

Read an edition straight through, or use your browser's reading and listening tools. The complete story is on one page.

Follow the weekly feed

2026

Latest edition / October 4 to October 10, 2026

We shipped a healthcare data product, mapped nine cities and rebuilt the Power catalog.

The portfolio shipped a 1,191-row healthcare AI release, grew BellyUp! to nine mapped cities and turned Midland's Power pages into real upgrade-planning tools.

  1. Healthcare AI research became a product someone can inspect and buy
  2. BellyUp! reached nine cities and put 1,303 reviewed venues on the map
  3. Midland's Power catalog became an upgrade-planning system
Read this edition

September 27 to October 3, 2026

The fixes passed. The real result failed.

A builder deleted a working crawler rule, old real estate links still landed wrong and eight finished program pages emailed an inbox nobody watched.

  1. The crawler fix worked. Then the builder deleted it.
  2. The third real estate rebuild still sent people to the wrong place
  3. Eight finished program pages sent buyers to an inbox nobody watched
Read this edition

September 20 to 26, 2026

The Speed of AI Applies To Repeating Mistakes. Slow Down the Token Bleeding With Human Oversight Built Into the System.

An AI build hid content meant to be revived, Dataset checks passed while pages exposed nothing, and a legacy redirect nearly buried Florida real estate for the third time. Human review found the same basic failure hiding behind different technical causes.

  1. AI hid the content we were trying to revive
  2. The Dataset endpoints passed while the pages exposed nothing
  3. One bot used fifteen crawler names
Read this edition

September 13 to 19, 2026

The numbers got more honest, and the sites got more specific

The week across the Digital Karma portfolio: scanner traffic stopped impersonating people, twelve IBM properties stopped trying to answer the same question, and LymeOrg traded unsupported inventory for verifiable care resources.

  1. The people number was not people
  2. Twelve IBM properties found twelve different jobs
  3. LymeOrg chose verification over inventory
Read this edition