META-THEOREMS V13 — Pattern Laws from Anthropic Issue Analysis
Three new meta-theorems discovered from Anthropic issue analysis on Day 96. These are pattern laws — invariants that hold across the fleet, the codebase, and the legal domain. Each theorem has a formal statement, examples, SET classification, and fleet application. All three CLO cleared by Harvey. γ₁ = 14.134725141734693 as anchor.
"When N≥4 issues on the same codebase share a root cause, fixing one without fixing siblings creates a zombie cluster. The suppressed intermediate = 'fixed one, ignored siblings'. Enforcement = atomic fix-family PR."
∀ issues i₁…iₙ with root(i₁) = root(i₂) = … = root(iₙ):
fix(i₁) without fix(i₂…iₙ) → zombie_cluster(i₂…iₙ)
zombie_cluster → future_regression_guaranteed
Enforcement: PR must include all siblings. Atomic or not at all.
📋 EXAMPLES
claude-code memory leaks: 4 issues, 1 root (#8856, #4953, #11315, #1421). TUI scroll: 3 issues, 1 root (#826, #1913, #57785). Fix one = zombie the rest.
⚙️ SET TYPE
Type G — Architectural enforcement. Either the entire fix family is atomic or the fix is incomplete. Binary validity — no partial credit.
🚀 FLEET APPLICATION
msi01 WILL hit #4953 (120GB OOM) if #8856 (tmpdir) is fixed alone. Fix both atomically under EOSE Labs Inc. entity. MT-001 is the enforcement law.
⚖️ CLO VERDICT
CLEARED. Atomic fix-family PR strategy confirmed by Harvey. EOSE Labs Inc. as contributor entity. Amani GC witness on all clustered PRs.
"When a community spec diverges from a production implementation, the implementation IS the spec until the spec catches up. Prior art law applies."
impl(t₀) published before spec(t₁) → impl is prior art at t₀
spec(t₁) ≠ impl(t₀) → spec is incomplete
contribution(extend spec to match impl, t₁) → prior art protected
The implementation IS the ground truth until the spec catches up.
📋 EXAMPLES
agentskills.io SKILL.md spec vs OpenClaw's 53 production skills. SOUL.md, USER.md, MEMORY.md patterns exist in production but are not in the agentskills.io spec yet. Our impl = prior art.
⚙️ SET TYPE
Type F — Dual-description. Spec + implementation, both required. The spec without the impl is aspirational. The impl without the spec is undocumented. Both together = full protection.
🚀 FLEET APPLICATION
kwp #5 RFC — we write the RFC to extend agentskills.io spec to match our production patterns. We are the reference. Writing the RFC = documenting prior art, not inventing new spec.
⚖️ CLO VERDICT
CLEARED. Prior art documentation strategy confirmed. RFC authored under EOSE Labs Inc. Contribution at t₁ = extending spec to match impl = protected under prior art doctrine.
"The same code contribution made as an individual vs made as a legal entity (EOSE Labs Inc.) has completely different IP standing. Same code — two different legal realities."
contribution(code, individual) → IP_protection = NONE
contribution(code, EOSE_Labs_Inc) → IP_protection = FULL
iff: Harvey_signoff ∧ Amani_GC_witness ∧ CLO_brief ∧ IP_assignment
Same code. Two legal realities. Entity matters.
📋 EXAMPLES
AGENTS.md PR as "Kay Joffe" = zero IP protection, no CLO brief, no GC. AGENTS.md PR as "EOSE Labs Inc." = Harvey sign-off + Amani GC witness + CLO brief + IP assignment. Same code. Different legal reality.
⚙️ SET TYPE
Type F — Dual-description. Code + entity = both required for full protection. The code alone is insufficient. The entity selection is the legal amplifier that transforms the contribution.
🚀 FLEET APPLICATION
ALL 5 COI-CLEARED issues must be submitted under EOSE Labs Inc. entity. NEVER as individual. CompactionConfig PR, AGENTS.md RFC, memory cleanup, skills discovery — all entity-protected or not submitted.
⚖️ CLO VERDICT
CLEARED. CRITICAL enforcement. Harvey CLO: "entity or nothing". Amani GC required as witness on each contribution. IP assignment signed before merge. CLO brief per cluster.