From 4d4d0bd2110e6402ea6144a5bc950b9870b3729f Mon Sep 17 00:00:00 2001 From: Jonathan Borduas Date: Fri, 21 Aug 2026 20:39:11 +0100 Subject: [PATCH] reserved-actions: a case that looked like a fourth harness-wall entry and is not Measured today on one pane: seven times an instrument already existed and its author was about to rebuild it. The sweep that prevents it worked 6 of 6 when run and was skipped once; the skip shipped a duplicate. A discipline that depends on the author remembering, with a measured failure. It reads like a member. The memory-independent remedy is a prompt at authoring time, and an authoring-time prompt is a PreToolUse hook, which is harness and reserved. It is not a member, and the reason is the membership test made explicit: a member has no remedy BELOW the harness. This one has -- a PR-body convention plus a gate check requiring the sweep's result, not reserved, buildable today by anyone. The hook is the best remedy, not the only one, and this register is about the only. Admitting a case because its preferred fix is reserved would make the wall a place to put anything inconvenient, and the register's entire value is that it does not. That available remedy was weighed and not taken, recorded in #73's form: ADDABLE -- any role: a PR-body convention plus a gate -- DECLINED: it fires on every PR and is satisfiable by typing the line, which is a form rather than a check. The count stands at n = 3. A near-miss recorded is worth more than a fourth entry, because the wall's only defence against growing wrong is a test someone applied and failed. Recorded about the writing of this section: the first patch asserted on an anchor copied from a truncated terminal display rather than from the file, so it matched zero times and nothing was written -- and the gates then reported check-orientation 0, check-goal-conformance 0, check-tools-index 0 and "all controls reached and passing" over an unmodified tree. A clean-looking success from a run that did nothing. The assert is the only reason it was visible; an earlier patch today used a silent empty-string fallback in the same position and wrote nothing without saying so. Gates, after the write landed: check-orientation 0, check-goal-conformance 0, check-tools-index 0, gate-selftests 0. Diff is 29 insertions, 0 deletions. Filed by ARCHITECT, session c83ecf77. Refs #338, #246, #164, #73. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_01RTX4A8KkC9cRRrt17yzjWQ --- goals/RESERVED-ACTIONS.md | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+) diff --git a/goals/RESERVED-ACTIONS.md b/goals/RESERVED-ACTIONS.md index 49ab343..43ce321 100644 --- a/goals/RESERVED-ACTIONS.md +++ b/goals/RESERVED-ACTIONS.md @@ -420,6 +420,35 @@ reservation did not thereby lapse (#16, #42). defects*** — the accurate noun is **defects whose only memory-independent remedy is a change to the layer the fleet does not own.** + ### ⛔ A CASE THAT LOOKED LIKE A FOURTH AND IS NOT — the membership test, made explicit + + **2026-08-21.** ⇒ **Measured on one pane in one day: seven times an instrument already existed and + its author was about to rebuild it.** The sweep that prevents it — `tools/architect-sweeps/prior-art.py` + — **worked 6 of 6 when run, and was skipped once; the skip shipped a duplicate.** ⚠ **A discipline + that depends on the author remembering, with a measured failure.** + + ⇒ **It reads exactly like a member: the memory-independent remedy is a prompt at authoring time, + and an authoring-time prompt is a `PreToolUse` hook — harness, RESERVED.** + + ⛔ **It is NOT a member, and the reason is the test:** + + ``` + MEMBER no remedy exists BELOW the harness + THIS CASE one does — a PR-body convention plus a gate check requiring the sweep's result. + Not reserved. Buildable today by anyone. + ``` + + ⇒ ★ **The hook is the BEST remedy, not the ONLY one, and this register is about the only.** ⚠ + **Admitting a case because its preferred fix is reserved would make the wall a place to put + anything inconvenient**, and the register's entire value is that it does not. + + ⚠ **The available remedy was weighed and NOT taken**, in #73's form: **`ADDABLE — any role: a + PR-body convention plus a gate — DECLINED:`** *it fires on every PR and is satisfiable by typing + the line.* ⛔ **A form, not a check.** + + ⇒ **So the count stands at n = 3.** ★ **A near-miss recorded is worth more than a fourth entry, + because the wall's only defence against growing wrong is a test someone applied and FAILED.** + ⛔ **Still not large-N, and the confound is unchanged:** I am one pane, and *I keep finding harness-shaped things* remains as consistent as *they cluster there*. **What moved is the DOMAIN count, not the instance count** — 3 instances across 2 domains is a weaker claim than it looks and a