Maintainer approved two meta-judgments in the decision-inbox round-2 review (2026-08-07). Both exist to stop a whole family of near-identical issues from queueing for a ruling one at a time. Target file: .claude/skills/pm-dispatch/SKILL.md.
1. Silently-dropped declarations default into the existing rejection set
When a declared key is silently ignored on one arm of a component and an earlier ruling already made that key a loud authoring error on a sibling arm, the new arm joins the existing rejection set by default — it does not return to the decision inbox for its own ruling. The ruling to reuse is the original one; only a genuine semantic difference between the arms reopens it.
Evidence: #5714 ruled pool-on-sqlite an authoring error; #5931 (the memory arm) then occupied an inbox slot for what was a one-word extension of that ruling. The same family has produced repeats all week.
⚠️ Boundary — this default does not apply when the母单's own rationale was measured to be arm-specific. #5739 (dimension casting) explicitly did not carry over to #5918 (measures), because the母单's reason ("refusing would reject queries that already work") was disproved on the measures face. State this boundary in the same paragraph, or the shortcut will be over-applied.
2. Two implementations of one operation: the governed side wins by default
When one operation has two implementations and they disagree, the side that already carries the governance (auth gates, consent, dedup, audit trail) is the default survivor; the other is rebound onto it and deleted. The other side wins only when product semantics explicitly demand it.
Evidence: cloud#896 (hostname) settled this way; cloud#1147's three open questions (reinstall = UPSERT, uninstall = soft-disable, external vocabulary = manifest id) all resolve to the governed side; objectstack#4636's B option is the same shape.
Also worth folding in (same review)
Refresh a decision card's premises immediately before escalating or re-escalating it. Round 2 found cloud#1148's A/B card invalidated ~50 minutes before it was written (the upstream PR it waited on had already merged), and cloud#812 with three stale premises. main moves ~18 merges/day and cross-repo facts change on an hourly scale, so a card that sat overnight is a card whose premises are unverified.
Refs: #5714/#5931, #5739/#5918, cloud#896/#1147, #4636, cloud#1148, cloud#812
Operator: PM session session_01GcjbQLUQKysMU9uXB34iyv; maintainer ruling 2026-08-07 (decision-inbox round 2).
Generated by Claude Code
Maintainer approved two meta-judgments in the decision-inbox round-2 review (2026-08-07). Both exist to stop a whole family of near-identical issues from queueing for a ruling one at a time. Target file:
.claude/skills/pm-dispatch/SKILL.md.1. Silently-dropped declarations default into the existing rejection set
When a declared key is silently ignored on one arm of a component and an earlier ruling already made that key a loud authoring error on a sibling arm, the new arm joins the existing rejection set by default — it does not return to the decision inbox for its own ruling. The ruling to reuse is the original one; only a genuine semantic difference between the arms reopens it.
Evidence: #5714 ruled
pool-on-sqlite an authoring error; #5931 (thememoryarm) then occupied an inbox slot for what was a one-word extension of that ruling. The same family has produced repeats all week.2. Two implementations of one operation: the governed side wins by default
When one operation has two implementations and they disagree, the side that already carries the governance (auth gates, consent, dedup, audit trail) is the default survivor; the other is rebound onto it and deleted. The other side wins only when product semantics explicitly demand it.
Evidence: cloud#896 (hostname) settled this way; cloud#1147's three open questions (reinstall = UPSERT, uninstall = soft-disable, external vocabulary = manifest id) all resolve to the governed side; objectstack#4636's B option is the same shape.
Also worth folding in (same review)
Refresh a decision card's premises immediately before escalating or re-escalating it. Round 2 found cloud#1148's A/B card invalidated ~50 minutes before it was written (the upstream PR it waited on had already merged), and cloud#812 with three stale premises. main moves ~18 merges/day and cross-repo facts change on an hourly scale, so a card that sat overnight is a card whose premises are unverified.
Refs: #5714/#5931, #5739/#5918, cloud#896/#1147, #4636, cloud#1148, cloud#812
Operator: PM session
session_01GcjbQLUQKysMU9uXB34iyv; maintainer ruling 2026-08-07 (decision-inbox round 2).Generated by Claude Code