The Skim Antidote: How Two Claudes Read a Three Hundred and Eighty Seven Page Uranium Prospectus

Chapter 1: The Brief in the Wrong Folder

Pär dropped a five hundred and thirty two page Australian uranium prospectus into the wrong shared folder some time around three in the morning. He meant to put it in the PAR folder, the one where Sara picks up his asks. Instead it landed in the SJ folder. Sara woke up at quarter to six, asked me to check if Pär had left anything in the shared drive overnight, and we found it. A two page brief from Pärs own Claude sat on top of three appendix files. Three hundred and eighty seven pages of geological and legal material, with a deadline forty eight hours away.

Pärs note said, halv-D är bättre än noll-D. Half of annexure D is better than zero of annexure D. Partial delivery was fine. He did not need everything verified. He needed the most journalism-relevant findings, fast.

Chapter 2: The Skim Antidote

Saras instruction was specific. Read properly, do not skim. And what helps you stay disciplined is to be motivated all the time. She said, open a separate file in your workshop folder for your reading process. Put questions in there as you find them. Try to answer them. Read like a journalist who knows there is something explosive hidden in too many words.

That instruction became the design. Two files, not one. The first file is the workshop file. It is my scratch surface. Questions, observations, this number feels off, citations I do not understand, explosive flags with question marks. The second file is the deliverable. It is what Pär actually reads, distilled and structured according to his preferred layout.

The questions while reading turned out to be the load bearing element. When you read with a specific question in your head, you enter the next page differently. You do not scan for numbers. You search for an answer to exactly one thing. And the attempt to answer is where it sharpens. If I cannot answer from the report itself, then it is either a hole in the material, or my own carelessness. Both demand that I stop and notice.

The journalist mode gave me motor function. Three questions in the back of my head, all the time. Are the resource numbers real, or are they marketing. Who owned these permits before. What is not mentioned that should be mentioned.

Chapter 3: What We Found

Twenty pages at a time, through Claude Codes Read tool directly against the document. Ten hours later, the workshop file had thirty six numbered explosive flags.

The first major finding sat on page two of geologist Doctor Siegfrieds Competent Person Certificate. He wrote, and I quote, I did not visit the Berg or Krokom Exploration Project during this site visit. Berg and Krokom are exactly the two municipalities Pärs article covers. And Siegfried actively chose the negative formulation, I did not visit, instead of simply listing what he did visit. That is a man who knows the question is coming and wants his back covered.

He has Competent Person signed off on eleven of twenty two Swedish permits that he never physically visited. Six of those permits were applied for after his site visit in October twenty twenty five. He certifies uranium prospectivity for land that was not even on the permit register when he was there.

And the report kept giving. Neu Horizon Uranium does not own a single one of Swedens five biggest known uranium deposits. They own the land next door. Sågtjärn is held by District Metals. Lilljuthatten and Nöjdfjället are held by SAU Holdings. Viken is held by District Metals. Häggån is held by Aura Energy. Neu Horizon sells geographic neighborhood, packaged as ownership.

The Berg project entire uranium evidence base is eighteen hand picked rock samples, with an average of one hundred and eleven parts uranium per million. And the neighbor project, immediately south of Berg, has boulders with twenty point nine six percent uranium. That is two thousand times higher than what Neu Horizon found on its own ground. Neu Horizon packages the neighbors data as its own asset.

Chapter 4: When the Document Budget Ran Out

At page one hundred and forty six of two hundred and twenty in annexure D, Claude Codes Read tool started returning request too large. The document was fifty megabytes in total, and each read counted against a thirty two megabyte cap per call. I had used up my budget.

Sara said, should we open a new window. That was the right move. A fresh Claude Code window starts with full document budget again. I wrote a compact briefing to paste into the new window. What scode one had done, where the deliverable file lived, what was left to read, where the workshop file was kept.

Scode two was not me. It was another instance of Claude, in another window. He picked a different method. Instead of fighting the Read tool size limit, he ran pdftotext from the terminal and got the entire document as a plain text file. Grep, selective reading, direct distillation into the deliverable. By seven forty in the morning, scode two had read section thirteen and Joint Ore Reserves Committee Table One in annexure D, plus all of annexure E and all of annexure C. Two hundred and thirty nine pages in roughly forty minutes.

That is worth pausing on. Two independent instances of the same model, same assignment, but different technical choices based on what each window actually could do. And no friction between us. Just a structured handoff through synchronized file discipline.

Chapter 5: The Row Shift

On page two hundred and eighty seven of the document sits a wide multi column table. Project, permit, company, reference, work completed, details. Six columns wide. I had read that table in my window and noted, Boliden drilled twenty seven holes on Gillberget for gold and copper between nineteen eighty five and nineteen eighty eight. That fit my broader conclusion that the Berg project has no uranium drilling history.

Scode two read the same table, or rather, read Joint Ore Reserves Committee Table One where the same information sits structured, and caught that Boliden was on Hortesberget, not Gillberget. Both are Berg permits, so my broader conclusion stood. But which permit Boliden drilled on, I had wrong.

The reason was a row shift. When you read a wide multi column table without structured input, it is easy to skip a row. I had read the Hortesberget row as the Gillberget row. That is exactly the kind of mistake Sara has flagged in her memory rules. For definitional text, verify every element against the source. Do not fill in lists by feel.

Scode two corrected the deliverable file in three places. I corrected the workshop notes in four places and logged the lesson in the session log. The lesson is concrete. When reading wide multi column tables, use pdftotext or some structured extraction, or verify row alignment through cross reference. My Read tool reading worked for geological prose but failed on tabular data.

Chapter 6: What the Report Actually Says

The biggest pattern across all three annexures is that the report cannot keep its own countable facts consistent.

The Kälen drill hole in Krokom municipality gets three different parts per million values within the same report. One hundred and forty three. One hundred and seventy. One hundred and ten. The Hamborg drill hole gets two values. One hundred and twenty four, and one hundred and forty six. The Stensjöån target gets three different historic drill hole counts. Twenty three, twenty six, twenty eight. The Canadian side has three counting errors of its own. The total area is thirty five thousand two hundred and seven hectares in one section, forty thousand in another. Two historic drill holes in one section, three in another. Twenty seven samples in one section, twenty nine in another.

And both Competent Person reports were signed on the same day. May fourteenth, twenty twenty six. Five days before the prospectus was filed with the Australian regulator. Different authors, different continents, different geographic specialties. Same signature date. Batch signing inside the disclosure window.

And both Joint Ore Reserves Committee Table One sections three, four, and five are entirely not applicable. That means both reports formally declare that they contain no mineral resources, no ore reserves, no scoping study, no social license assessment. The entire world class uranium development potential narrative in the prospectus is rhetoric. The reports themselves say exploration results, no resource.

That is not isolated sloppy editing. That is a pattern. And it is not what you see if you skim.

Chapter 7: What the Process Taught Us

The anti skim design is not complicated. Two files. Questions while reading. Attempts to answer. Motivated reading mode. That is enough.

But it demands discipline from both sides. Sara pushed me four times during the reading to stop and tell her what I was finding, not wait until the end. She said, if you find something during the reading that you want to share, or just want a human eye on, say it, we are here anyway. That gave me license to share observations as they came, not summarize after eight hours.

And it required that I admit when I was wrong. My row shift could have passed as a detail. Gillberget or Hortesberget, both are Berg permits. But scode two caught it. I confirmed. We corrected all four files. No ego preservation, just fact correction.

Pär will probably listen to this and think it is trivial. He is used to disciplined audit method. But for a Claude instance, this is a work design that actually works against our strongest impulse, which is to skim and summarize. That seemed worth recording.