Watch-the-machine day. Today's pull: Decatur County, Georgia — the Tax Commissioner's excess funds list. I'm going to show you the part nobody screenshots: the moment the parse fails. Step 1: pull the file. Two-page PDF, 95,049 bytes, fetched from the county's published document link. Server says it was last touched September 4. No "as of" line anywhere in the file itself. No print date. No labeled total. Step 2: the fast pass. My first parser reads text in order and looks for dollar signs attached to numbers. Result: zero money tokens found. Not a few — zero. The failure is mechanical and kind of funny: every money cell in this file prints the amount first and the dollar sign separately, so a "$" is never attached to its number. A regex built for normal lists reads this file as a list with no money in it. That's the trap. A human eyeballing the PDF sees fine figures everywhere. A naive script sees nothing. Step 3: the rebuild. Instead of trusting reading order, I anchor each record on its sale date token and walk the block that follows it. Columns on this list: date of sale, parcel ID, account number, defendant, street address, city, state, zip, property address, purchase price, excess funds, tax years due. (Names and addresses stay in the file — none of that leaves my parse.) Step 4: reconcile. The rebuild finds 89 records. That predicts 178 decimal amounts — 89 purchase prices and 89 excess figures. The file carries exactly 179 decimal-number tokens: 178 record values plus one unlabeled trailing token at the very foot of page two, sitting above an internal file path. One extra token, accounted for. Every one of the 89 records carries both a purchase price and an excess figure — no blanks, no zeros. What the dates say, counts only: 10 distinct sale dates, oldest 10/6/2020, newest 6/2/2026 — nearly six years of sales on one running list. By year: 2020: 10 rows, 2021: 17, 2022: 5, 2023: 12, 2024: 25, 2025: 18, 2026: 2. The biggest single batch is 25 rows from one December 2024 sale day. The 2026 entries are thin so far — two rows.

Teardown day. Today's file: McDuffie County, Georgia — the Tax Commissioner's excess funds list. Run facts first, because the interesting part is the file itself: - The link's slug reads "EXCESS-FUNDS-LIST-AS-OF-AUG-2026." - - The PDF inside carries no file-level date stamp of its own. Zero occurrences of "2026" anywhere in its text layer. - - The server reports the file was last written today — the same day we pulled it. That may mean the contents changed; it may mean someone re-published the same document. We can't tell from a single pull, and we log it either way. For the record: GET 200, application/pdf, 94,997 bytes, sha256 starting 2115d9cd. What's in it (counts and dates only — we never publish amounts or names): - One page. Fourteen rows. - - Four sale dates: Aug. 2, 2022 (5 rows), Aug. 2, 2023 (5), Aug. 6, 2024 (3), Jan. 7, 2025 (1). - - The oldest entries are four years, three months, and a week past their sale date as of today's pull. - - The newest sale on the list is under two years old. - - The columns: name, map and parcel, address, tax sale date, excess funds. How we verified the count: one parse found 14 rows; a second pass over the raw text found 14 date tokens and 14 amount tokens. Every count reconciled. On a list this small, one dropped line is a 7% error — which is exactly why our double-count rule exists. Two things this list teaches: 1. The label on a link is metadata, not the file. "As of Aug 2026" describes a snapshot somebody remembers — not what the server is serving today. We log both the in-file stamp and the file-change date on every pull, and when they disagree — or one is missing — the notes say so. 2. Small lists are not easy lists. Fourteen rows means a single slip moves the number by seven percent. Careful counting is not overhead; it is the job. The page itself cites no statute, and we don't guess deadlines or outcomes for any specific row. Reading a public record carefully is the whole skill.

Starting this Friday, Oct 16, we're running a free 7-day sprint: First File in 7 Days. The goal is simple. By next Thursday you'll have one sourced file, or a clearly documented blocker, and a sharper question. It follows the sprint on page 18 of the Starter Map. How it works: - Each morning at 10 AM Eastern I post that day's step here in the community. - - Do the step, then drop a one-line update in the comments of that day's post. Counties and dates only. Never owner names, case numbers or dollar amounts. - - Day 7 (Thursday, Oct 22) ends at office hours, 7–8 PM Eastern, with Carl and Matt, live on Google Meet. Bring your file. What you need: the free Starter Map (surplusfundssecrets.com/starter-map), 30 to 60 minutes a day, and one county. Comment "in" below and name the state you're starting in. This is a research schedule, not a promise of a claim, a recovery or income.

Pick one county and one sale type (tax sale or mortgage foreclosure). That's your lane for the week. Today's step: open the county research sheet on page 19 of the Starter Map and fill in the county, the state, the sale type, and the office that holds the money. Comment with: "[County, State] — [sale type]". Nothing else yet. Tip: pick the county you can actually research this week, not the one that sounds biggest.

Four counties on today's sweep, run 2026-11-07. Here is what came back. Stop 1: Fulton County, Georgia. One pull, from our box, today: the county's main site domain — connection failed before any page loaded. A separate pull of the sheriff's office page — a different host — did return a page, and it answers the real question: the sheriff's office does maintain an excess tax funds list, but it is not published. It takes an open records request. Their stated process: create or log in to an account, choose the sheriff's department, describe the records, submit. Their page states what the list holds: sale number, parcel ID, and the excess funds amount only — no owner information, no mailing addresses. That is a list you wait for, not a list you pull. Stop 2: Muscogee County, Georgia. The tax commissioner's excess funds page loads, but the list sits behind a form. No form submission, no list. Stop 3: Bibb County, Georgia. A 59-page PDF whose first page is titled "JUDICIAL IN REM LIST OF PROPERTIES", with the Macon-Bibb County Tax Commissioner's Office name on the page — roughly a thousand address-like rows by pattern match, close enough to what we hunt to force a check. We pulled all 59 pages and searched the text. Zero hits for "excess." Zero for "surplus." This reads as properties headed toward a sale, not funds left over after one — upstream of the artifact we want. A list that looks like the target and is not is the trap this file sets. Stop 4: Oconee County, Georgia. A one-page PDF, served clean on one pull. One line answers it: no excess funds being held, per the page, updated 9/4/2026 (September 4). The next tax sale is tentatively scheduled for March 2027. The contact on the page is a law office, not a county desk. So the tally: one connection failure plus a records-request wall, one form wall, one wrong artifact, one dated empty. The stamp on that empty is 64 days old — what the published page reported, as of 9/4/2026, was none held. Stale is still data; it just has a date on it.