The Notebook · week by week
The daily page, one week at a time.
The same letter readers get by email every Sunday. Get it in your inbox.
- 2 November 2026 to 8 November 2026 · one idea llm upgrade testing: the new model passed the old tests
- 26 October 2026 to 1 November 2026 · 7 ideas the answers survived the squeeze and the reasons did not · the retry was safe and the side effect was not · the retriever was fine and the pdf was not · the questions looked alike and the answers did not · the shortlist decided the score before the reranker did · the numbers were right and the units were not · the longer list looked better because guessing got easier
- 19 October 2026 to 25 October 2026 · 7 ideas pinned to zero and still changing its mind · the summary kept the answer and dropped the hedge · the reasoning stayed honest and the answer went rogue · the attackers went straight for the rows the benchmark left blank · the speed-up went backwards once we asked for json · it found the needle every time and still lost count · it deferred every ticket and called that caution
- 12 October 2026 to 18 October 2026 · 7 ideas the forgetting was in the words your data never used · the answer was correct and the procedure never ran · the skill that won the argument wasn't the one you installed · the copy came out steadier than the thing it copied · the cache turned out to be slower than thinking about it again · the guardrail passed because nobody tried the door · amortising the housekeeping made the recall worse, not cheaper
- 5 October 2026 to 11 October 2026 · 7 ideas nobody asks the whole question on the first turn · the rehearsal went well because nobody else was in it · the backslash went missing somewhere between the model and the shell · the rule at the top of the query turned out to be more query · the steadiest judge was wrong the same way every time · the thinking-off switch turned out to be a suggestion · the limits were in the file the agent was allowed to edit
- 28 September 2026 to 4 October 2026 · 3 ideas the pass rate went up and so did everything else · we kept pruning the memories that never got a turn · the vaguer the question, the more it helped itself