attention-is-not-a-mechanism — hope-as-gate prohibition

Census id: F4 · Family: F — rules corpus + generated-rule tooling + shipped audit scripts · Provenance: framework pin b069c59328aa8e08671d3a19fd25efa618dfc1c5 (staging FETCH_HEAD, read 2026-09-11; re-anchored from census pin a1337cb301). Every file:line below is at that pin. Raw reference drafted by stage E4 — factual, unpolished by design.

What it bans

Checks whose detection layer is someone's attention. .claude/rules/attention-is-not-a-mechanism.md:1 «# Attention is not a mechanism — discipline rule»; the rule itself: .claude/rules/attention-is-not-a-mechanism.md:17 «A load-bearing check MUST be one of: (a) a deterministic gate at the earliest reachable» / :18 «channel, or (b) a NAMED cold-agent protocol with structured output» — and the prohibition: :19 « Bare human attention — «a reviewer will read» / «the diff» — and unstructured AI attention — «the model will notice» — are NOT mechanisms; they» / «may serve only as decision/merge authority on top of (a)/(b).» Class header: .claude/rules/attention-is-not-a-mechanism.md:3 «> Class: C — prose-only; promotion criterion in §3.»

Never (fires)

The two named anti-patterns: .claude/rules/attention-is-not-a-mechanism.md:28 «- #hope-as-gate — a check whose failure mode is «nobody looked». Counter: (a)/(b) above.» and :29 «- #warning-nobody-reads — load-bearing warnings. Counter: error+escape or agent-audited.» The warning corollary makes the firing shape concrete: .claude/rules/attention-is-not-a-mechanism.md:21 «error-with-escape-token» — in context: :20-24 «Corollary: a WARNING whose only» / «consumer is «someone reads the log» is attention-dependent detection — either promote to» / «error-with-escape-token (rationale ≥20 chars; precedent ci-tool-pinning.md §3) or route to a» / «named agent audit.» Origin evidence (a flipped claim caught by nothing): :11 «> Evidence base: P3 (a flipped «Enforces» claim in AGENTS.md caught by NOTHING — zero readers);»

Always (clean)

A check that is either a deterministic gate at the earliest reachable channel or a named cold-agent protocol with structured output (compliance-verifier / backward-sweep-auditor / docplan-auditor class — :17-19 above). First consumers quoted: .claude/rules/attention-is-not-a-mechanism.md:39 «First consumers ship with MT S4: composition gate FF8002 as error+excluded-escape (not» / :40 «warning), and agents/docplan-auditor.md replacing «human reads the plan diff». This rule»

Enforcement channels

• Rule-index row: .claude/rules/00-rule-index.md:12 «| attention-is-not-a-mechanism.md | C | designing any load-bearing check (gate vs. bare human/AI attention). | always-on core |» — always-on core, no executable artifact today.
• Promotion path (.claude/rules/attention-is-not-a-mechanism.md:33): «Promote to audit-checklist dimension (Phase -1 / reviewer protocols) after 3 documented» / «incidents in 6 months where a bare-attention check missed a real defect.»
• Not backend-rendered — no FF diagnostic applies to this rule itself (FF7001/FF7002 are backend render refusals, packages/core/diagnostics/registry.ts:321/:329; the FF8002 mention at :39 is a downstream consumer's composition gate, quoted where it appears, not this rule's channel). Honest status: always-on prose + named-agent audits at consumption sites.