Research Notes¶
Everything else in this knowledge base is finished. A page in another bundle has been researched, cited, argued over by a review panel, and audited before you see it. It states what is known and expects you to rely on it.
These pages are different, and they say so at the top of each one. A research note is the current understanding of a question that is still open. The findings in it are real and sourced. The conclusion is provisional.
Notes exist because some questions are worth capturing before they are worth publishing. A topic keeps surfacing, the reading turns up something genuinely useful, and the honest state of the work is "here is what I found, and here is what is still unresolved." Forcing that into a finished guide would mean pretending to a confidence the evidence does not support.
What a note guarantees¶
The apparatus around notes is lighter. The factual standard is not.
- Every number is cited to its primary source, the same as any other page here, and lands in this bundle's evidence ledger.
- Every stated fact is verified against that primary source before the note is published.
- Uncertainty is named where it exists, rather than smoothed into confident prose.
A note holds its conclusion loosely. It does not state a fact loosely. Where the evidence runs out, the note says so and names the question that would settle it.
What happens to a note¶
Notes do not sit here indefinitely.
- A note that holds up graduates. When the question settles, it gets rewritten as a proper page in the bundle where it belongs, through the full authoring process. What remains here is a pointer to it.
- A note that turns out to be wrong is marked superseded, in place, with a link to whatever replaced it. That record stays. Being wrong in public, dated and traceable, is a reasonable thing for a knowledge base to keep.
How to read one¶
Treat a note as a well-sourced starting point rather than an answer. The citations are worth following, the open questions are the genuinely interesting part, and the conclusion may not survive the next thing that gets published on the topic.
In this section¶
| Page | Last updated |
|---|---|
| Rethink the Process, or Just Add AI? Tests the advice to redesign a process for AI rather than automate it, against the evidence and the record of 1990s reengineering. |
Updated 2026-09-13 |
| What an AI Test Can and Cannot Tell You A working note on AI evals: what testing an AI system can tell you, what the evidence shows, and what nobody has shown yet. |
Updated 2026-09-13 |
| Why Your AI Gives Three Different Answers A working note on shared definitions — why the same question returns different numbers, and how much semantic modeling is actually worth doing. |
Updated 2026-08-10 |
| Glossary Plain-language definitions of the terms used across this bundle. |
Updated 2026-08-10 |
| Validation How these notes are validated — sourced claims are checked and each sweep is recorded here. |
Updated 2026-08-10 |