The Loop Atlas: the record - Every prediction a run judged, every other check a run made, how each came out, and every miss with its reason

The Loop Atlas: the record is the companion volume to The Loop Atlas (10.5281/zenodo.24100672), the book that asks what a loop leaves behind when it comes home. The book tells the story and counts the results; this volume holds the receipts. It is made by script from the runs of record at every build, so that it cannot fall behind them, and it is made for checking rather than for reading through. What is in it. Part 1 prints every prediction about the toys that a run judged, 1,104 in all, day by day in the order the runs were made: 706 held, 369 missed and 29 were left undecided, each with the run that judged it, named by its tool and its stamp, the prediction in the run's own words and its verdict. Part 2 prints the 1,015 other checks the runs made: 183 rivals written to win if a prediction was wrong, 572 checks planted in each tool before its numbers were used, 243 lines of the toy pages' engine checked against the runs, 12 of the book builder's own checks and 5 checks of files and records. Part 3 prints every miss, 656 of them, numbered in the order they were found, each with what happened, what kind of miss it was, what it taught, whether a redesign was named, and whether the lesson has been proved by a run. Part 4 lists every one of the 830 run files in the archive, with its kind and where it is cited, and names the 12 cited runs that did not travel. The files. LOOP_ATLAS_RECORD_2026-10-29.pdf, with its source as md. The tools and the run files themselves are in the archive deposited with The Loop Atlas (LOOP_ATLAS_ARCHIVE_2026-10-31.zip, in the book's own record), beside the reader's edition and the toy pages. Limits. A preprint, not peer reviewed. Toy models with assumed numbers, not experiments: every number here was printed by a small computer model with values we chose. Made with Claude (Anthropic), which wrote and ran every tool and the scripts that print this volume; responsibility for the work rests with the named human author. Contact. Matthew "Bear" Schulz, Independent Researcher, Belmont, CA · ORCID 0009-0000-0956-5743 · mattschulz97@protonmail.com