PR #793 follow-ups: design 006 A1 does not use Step 7's amendment-entry template nf-xj1 ← Beads

Twinned from https://github.com/cfsh/ninefold/issues/798 by tools/beads/import_github.py's reconcile pass.

Suggestions from ninefold-reviewer's review of #793. Unclaimed and unlabelled — nothing is blocked on these. #793 is CONVERGED and approved by @aedanpope; these are the residuals that were below the gate.

All line numbers are design_docs/006-the-coders-front-door.md at 23df301d.

1. A1 carries none of Step 7's four entry elements

.claude/commands/design.md:289-303 — ### What an entry contains — prescribes:

~~~markdown ### A<N> — <YYYY-MM-DD> · <one-line title> (§<sections superseded>)

Facts moved. | Reasoning was wrong. | Direction changed. <what the doc says, why it no longer holds, and the evidence — re-measured, not recalled>

What this reverses: <the narrowest accurate scope — and say what still stands> What it adds: <the transferable lesson, if any> Discovered by: <piece / PR>. Landed in: <PR for the code> · <PR for the amendment> ~~~

Both precedents follow it — 001-no-logic-in-shell.md:28 (Facts moved.) and 005-probe-selection-and-build-config.md:17 (Direction changed.), the latter carrying What this reverses (:32), What it adds (:37) and Discovered by … Landed in (:52). A1 (:15-42) substitutes its own shape and has none of the four.

Discovered by / Landed in is one line and free: discovered implementing piece 12 (#780); landed in #788 (the code) · #793 (the amendment).

2. The classification line is the one field that changes practice, and A1 is genuinely ambiguous

Step 7's ⚠️ is *"Classify honestly, because only one of the three should change practice."* #784 reads as Direction changed — @aedanpope wanting something different, no error. But #793's own body argues something stronger: that S6's *mechanism* was wrong on the evidence available at the time (*"S6's instinct was right and its mechanism was one tier too timid"*). Under Step 7 that is Reasoning was wrong, and it is the one that earns a sentence on *what would have caught it*. The doc states neither, so a later round cannot tell whether design 006's process needs to change.

3. What it adds is empty, and its content existed in this PR two rounds ago

> "S6's instinct was right and its mechanism was one tier too timid: the strongest version of *'a dev stops monitoring services it does not own'* is not moving the alarm, it is making the failure unable to matter."

Present at f1c52b9 and 62ecbb9, cut at 8b8a264 on ninefold-reviewer's own round-2 recommendation, which was wrong about it — that pass judged those paragraphs as restating §2.8, and that sentence is not in §2.8. It is new analysis about S6, generated by implementing it. It now survives only in #793's body and notes/history.md, and design_docs/README.md is explicit that a reader implements from the doc without fetching either. That is 006 A1's own thesis applied to itself.

Cost is ~6 lines against a 35-line section: §0 would move from :66 to about :72, still inside the Read(limit=80) window the round-2 trim restored.

4. The heading's superseded-section list is incomplete

:15 reads (S6, §2.8, §5.3, §8.1, §8.2) — five of the six annotated sites, omitting §2.13 (the *"eliminated outright by S1–S6"* count, pointer at :1063), and naming S6 where the other four are §-numbered (the S6 row lives in §0's Suggestions table, :151). The in-body enumeration at :39-42 is complete and correct; the heading should match it. 005:7's header bullet also carries a section list where 006's :7 carries none.

5. Cosmetic — the TOC line splits ## Contents

:48's Amendments bullet sits *above* the *"Top level only, deliberately…"* preamble, so the section renders as one bullet, a paragraph of prose, then nine bullets. Moving it under the preamble with the rest of the list costs nothing.

---

⚠️ design_docs/** is inside nflib.perimeter's deny list, so whoever takes this waits for @aedanpope's approval on the PR — a property of the path, not of the change.