docs(readme): add Who This Helps and US manufacturing impact sections

README.md

@@ -24,7 +24,72 @@ Off-the-shelf AI tools are not designed for the shop floor. They are:
 - Not CMMC-aligned (non-starter for defense-adjacent suppliers)
 - Not auditable (no governance lineage, no human-in-the-loop discipline)
 
-Gartner estimates **40%+ of agentic AI projects will be cancelled by 2027** due to unclear value and inadequate risk controls. This framework is the missing reference implementation that makes agentic AI safe to deploy in manufacturing.
+Gartner projects 40%+ of agentic AI projects will be cancelled by 2027 due to unclear value and inadequate risk controls. This framework is the missing reference implementation — UNS-aware, OT-aware, governance-aware — that makes agentic AI safe to deploy in manufacturing.
+
+---
+
+## Who This Helps
+
+This framework is designed for U.S. manufacturers being asked to do more with 
+fewer experienced people, under tighter compliance requirements, on shorter 
+timelines.
+
+**Small-to-mid manufacturers (SMMs)** — approximately 99% of U.S. manufacturers, 
+including Tier-2 and Tier-3 suppliers feeding the CHIPS Act semiconductor 
+ecosystem. SMMs cannot afford enterprise-scale data and AI teams; this 
+framework gives them a reference architecture deployable on a laptop with 
+Ollama, scalable to AWS Bedrock or Azure OpenAI when production-ready.
+
+**Defense industrial base subcontractors** — need CMMC Level 2 / NIST SP 800-171 
+alignment as a non-negotiable prerequisite for DoD contracts. The framework's 
+policy engine, Ed25519-signed audit trails, and Purdue-zone enforcement are 
+built to support that posture.
+
+**CHIPS Act and reshoring brownfield plants** — inherit legacy SCADA, MES, and 
+historian estates. The UNS Context Broker bridges legacy OT and modern AI 
+without rip-and-replace, preserving capital while accelerating Industry 4.0 
+maturity.
+
+**NIST MEP Network centers and Manufacturing USA institutes** — need free, 
+open, standards-aligned reference architectures they can recommend to their 
+thousands of SMM clients without endorsing any one vendor.
+
+**Academic and research labs** — get a reproducible benchmark suite 
+(IABENCH-v1) and a synthetic UNS dataset to advance industrial agentic AI 
+research without negotiating proprietary data access.
+
+---
+
+## How This Advances U.S. Manufacturing
+
+The framework is designed to move five concrete needles:
+
+**1. Knowledge cliff mitigation.** The Tacit-Knowledge Curator captures 
+retiring-expert tribal knowledge into a governed, queryable layer before it 
+leaves the shop floor. A junior technician gets senior-engineer answers 
+on-demand, grounded in cited sources. This addresses the 30+ years of 
+institutional knowledge projected to retire by 2030.
+
+**2. AI safety by construction.** Purdue zoning, OPA policy gates, Ed25519 
+audit trails, and human-in-the-loop escalation mean operators retain 
+accountability. This is the missing layer behind the 40%+ agentic-AI project 
+cancellation rate Gartner projects through 2027 — most failures are 
+governance failures, not model failures.
+
+**3. Standards-first compliance.** ISA-95, OPC UA, MQTT Sparkplug B, NIST 
+SP 800-82, CMMC L2, and NIST AI RMF aren't bolted on — they are architectural 
+primitives. Suppliers can demonstrate compliance posture without expensive 
+retrofitting, lowering the cost of entry into the defense industrial base.
+
+**4. Open-source multiplier.** As a free reference implementation under MIT 
+license, the framework gives MEP centers, system integrators, and academic 
+groups a starting point they can adapt rather than rebuild. Public 
+infrastructure for industrial AI.
+
+**5. CHIPS-era supply chain resilience.** Tier-2 and Tier-3 suppliers gain 
+operational visibility and decision intelligence comparable to what their 
+large-customer fabs already have, narrowing the resilience gap that has 
+historically slowed reshoring of high-complexity manufacturing.
 
 ---
 
@@ -129,7 +194,7 @@ LLM_PROVIDER=ollama make demo
 | Safety guardrail compliance | Block-rate | TBD | TBD | TBD |
 | Hallucination rate | % | TBD | TBD | TBD |
 
-*Results populated in Phase 7.*
+*Headline benchmark results will be published alongside the v0.1.0 release. See benchmarks/industrial_agent_benchmark.md for the full task specification and benchmarks/runner/ for the reproducible harness.*
 
 ---