The law does not say “destroy the book”

Buying a book gives you an object you can resell, lend, or give away under first sale. It does not let you scan it. Scanning is a reproduction, so you need permission or an exception such as fair use. There is no statutory right to turn a book you own into an ebook.

Fair use is a four-factor balancing test, decided after the fact by a judge. In Bartz, Judge William Alsup leaned on the fact that each digital copy replaced a destroyed print copy and never left the company. He did not order the books destroyed, and he did not rule that keeping them would have been illegal. But destruction was a fact he relied on again and again, and a company reading the ruling now has a reason to pulp its originals even when it could scan them intact.

The Internet Archive’s 2024 defeat is a different case. It lent full digital copies to the public without permission, in competition with licensed ebooks. It does not say that keeping a paper original next to a private scan is infringement. It does show how little patience courts have once copies start leaving the building.

The money half

This is where I had to concede. Loose pages go through a sheet feeder. An intact book needs its pages turned. At millions of books, that gap is the whole budget, and no exception in the statute makes a page-turning scanner cheaper. Some companies would cut the bindings regardless, and I can’t claim a better law would have saved Anthropic’s books.

What a better law can do is give whoever keeps the books intact a clean path to donating them. They would still need a willing library and someone to pay for sorting and shipping. Those costs are real. Legal doubt about keeping the scan needn’t be one more.

The argument I dropped

My first draft framed this as books lost to humanity. Claude talked me out of leading with that, and I think rightly. The books in a bulk purchase are mostly used copies of mass-printed titles. Pulping one leaves thousands of siblings, and a sharp reader says “so what, it’s a used copy of a bestseller” and closes the tab.

So the claim is narrower. Pulping a readable book does not erase the work, but it does leave one fewer copy for someone to read, and that is enough to ask what the destruction was for. Rare editions and annotated copies exist in any bulk lot, but the argument shouldn’t hinge on proving one went through the scanner. An ordinary readable book is worth passing on too.

The redundant waste

Now picture several companies following the same recipe. Each buys the same book, cuts it apart, and scans it. Several destroyed books, several files of the same text. The fair-use ruling gives none of them a right to copy another’s library, so each must build its own.

None of that is technically necessary. A rights holder can license one digital source to many customers, and most of these titles already exist as ebooks. Agreeing on rights, prices, and access rules is a bigger project than the exception I’m asking for below. It still beats feeding another copy of the same book through another scanner.

What I am asking

If you are American, ask Congress for a format-shifting exception: the owner of a lawfully acquired book may scan it for their own use, or for an organization’s internal library, and may then donate the original without deleting the scan. Nothing more. No right to distribute or lend the files, and no ruling on what further uses, AI training included, are allowed.

There is a serious objection. If each successive owner scans the book and passes it on, one physical copy could support many retained digital copies. Authors and publishers are right to worry about that, and a real bill would have to address repeated circulation, and whether commercial libraries need different conditions or compensation. Calling the change small doesn’t make those questions go away.

Congress already writes copying exceptions this specific. Section 108 lets libraries and archives make copies under detailed conditions. It is not a ready answer for private owners or AI companies, but it proves we can do better than “argue fair use and see who wins.”

I don’t have a finished bill. I have a test for one. When someone has paid for a book and scanned it, keeping that book available for another reader should never leave them worse off than throwing it away.

The gap

After f32 and f64 were done, I started on f128. That is where the plan broke. CORE-MATH has 41 binary64 functions and 15 binary128 ones. Square root, cube root, exponentials, one logarithm, and the inverse trigonometric functions were there. Sine and cosine were not: the binary128/sin directory had held a hard-to-round corpus since February, with no C file next to it. No tangent, no log2q, log10q, log1pq, or powq either.

Testing a correctly rounded function without an oracle is a different activity from testing one with an oracle. I ended up generating my own corpora with MPFR and a few Diophantine tricks, such as continued fractions of 2/π per binade for sine and cosine, and checking against them in CI. By the end of August every q function in metallic-rs that CORE-MATH also had was within a few percent of it, and the rest had nothing to be compared with except libquadmath, which is faithful rather than correctly rounded.

At that point the lazy conclusion was obvious. If the reference is missing the function, contribute the function to the reference.

The drama

On 3 September I mailed the CORE-MATH list a work-in-progress sinq and cosq, ported from metallic-rs to C and extended to all four rounding modes. The list bounced it: posting is for subscribers only, and I had never subscribed. I resent it the next day with the usual disclosure that I had used Claude Code while porting and testing.

Paul Zimmermann’s reply opened with the sentence I had been half expecting:

this is interesting. However we have not decided yet if we include AI-generated contributions.

followed by a question I could not dodge:

how much of the code was generated by you and not by AI?

Honestly, about all of it since April 2026. The methods are mine. I wrote them up in How to program mathematical functions back in 2021, turned that post into an agent skill this year, and have been doing what I can only call vibe optimization since. Everything else in the same reply was ordinary review: Taylor series where a minimax polynomial would do, a bound of π/512 < 2^−7.35 that was wrong in the second decimal, and timings on an i7-8700 showing my sine at about 250 cycles while expq took 60.

So I did what one does. Version 2 swapped in a degree-four minimax from rminimax and fixed the bound to 2^−7.34. Version 3 evaluated the accurate path in tiers of 128, 256, and 384 bits and halved its cost. Then came the least flattering exchange of the thread: “I don’t see the v3 in your core-math-c repo.” It was sitting unpushed on my development machine.

Paul also asked whether I would maintain the code if it were merged and problems turned up later. That question is the actual entrance exam. I said yes, and I added that I had a bigger plan: to close the gap between binary128 and binary64, one or two functions at a time. He opened an account for me on gitlab.inria.fr with the maintainer role, and sinq and cosq landed on master on 19 September. The announcement quotes 154.5 cycles per call for sinq against 872 for libquadmath, and 156 against 995.5 for cosq.

Where it stands

That was one function pair out of the two dozen or so still missing in binary128. I am now working on sincosq, tanq, the other logarithms, and whatever comes next, one or two at a time. Most of them already run in metallic-rs, so porting them is mostly a matter of adding three rounding modes and surviving review.