A running log of decisions taken on Zone, written so they can be reviewed without a terminal. Each entry states what is proposed, shows real data from the live system rather than an invented example, and names the part that needs a ruling.

• What a cascade is, with real examples 2026-09-30 — draft. Working definitions of a pattern cascade and a policy set for the War Machine, seven real policy sets drawn out, and what they show that is wrong or unsettled.
• What is a mined proof good for? 2026-09-20 — what 121 extracted argument graphs can already offer a mathematical reader, what more papers would add, and why neither intact structure nor a passing gate makes a proof trustworthy.
• Five War Machine walkthroughs before construction 2026-09-15 — conceptual review drafts: prediction; patterns and G; the connected learning loop; a qualifying run; purpose and the published account. Build dispatch is held while we examine the explanations.
• One cascade, computed both ways: what G over cascades actually decides 2026-09-15 — a real zaif board (six chips, two wirings) scored under both readings of Joe's Alexander ruling, with worked numbers: order changes G by ~380×, and a long language of good patterns is free under one reading and metered under the other. The fork is shown, not settled.
• Teaching the construction: patterns, proof cascades, and an APM TA 2026-09-11 — a worked finite-net proof, what the TA–Student pilot actually showed, and the first steps toward a pattern-first APM V4.
• The index proved a coverage it had never scanned 2026-08-17 — we merged two evidence stores by replaying 7,249 documents, and the index then wrote a formal proof of completeness over six days it had never looked at. Every component behaved correctly. Why the claim was still false, and whether the store should refuse to make it.
• The proof assistant held the refutation, and was never asked 2026-08-14 — a sorry-free Lean file could have refuted a claim we made in LaTeX. It never did, because in Lean the claim was not false but inexpressible. Eleven lines of Clojure caught it. What that says about proofs versus bridges, and the three cheap practices we are adopting.
• Why deleting a pattern needs :relations 2026-08-13 — one pattern is 15 documents; we can currently delete 8 of them. What a one-line change to the retraction allow-list buys, and the one reason it might not be a one-liner.