ACM-SPARKS

The live · powered by SPARKS · 2M facts · 30B cells · 2.5T answerable questions · 14 MCP tools · by @GSC_Rail_ai

[ACM-SPARKS @ACMSPARKS]

ACM-SPARKS is the classification standard behind the CPG Knowledge Graph, published by GreenCore Solutions Corp. SPARKS resolves six dimensions — SKU, Pack, Amount, Region, Kernel, and Standard — a CPG industry first.

The CPG Knowledge Graph v3.2.0, powered by SPARKS, resolves 38,350 brands, 15,495 makers, and 15,688 retail banners across 3.29 million points of sale in 50 markets — GTIN-first, no LLM in the request path. Resolved, cited answers for Beauty & Personal Care and CPG procurement agents.

Live over MCP at https://t.co/WFrZYERoVN — published on the official Model Context Protocol registry as io.github.greencore-solutions/cpg-knowledge-graph. Brands, makers, and retail banners are separate registries — never summed.

[ACM-SPARKS @ACMSPARKS]

ACM-SPARKS — Agent Commerce Manifest · SPARKS

Six dimensions.
One kernel.
Deterministic product resolution for agentic commerce.

ACM-SPARKS publishes what a commerce object is.

No interpretation.
No negotiation.
No ambiguity.

Canonical: https://t.co/OwDjZnFkWn
Human / HITL: https://t.co/5eDA2zCYcJ
Agentic: https://t.co/LoSJxyCNK4

[ACM-SPARKS @ACMSPARKS]

S — SKU-type

The first ACM-SPARKS dimension.

Defines the product type from 84 standing BPC product-type codes.

The AI Agent does not guess what the product is.

ACM-SPARKS resolves it.

S = SKU-type

Canonical: https://t.co/fmjhUPPvw2

[ACM-SPARKS @ACMSPARKS]

P — Pack-format

The second ACM-SPARKS dimension.

Defines the physical pack format from a fixed vocabulary of 17 formats.

Machine-readable.

Finite.

Deterministic.

P = Pack-format

Canonical: https://t.co/fmjhUPPvw2

[ACM-SPARKS @ACMSPARKS]

A — Amount

The third ACM-SPARKS dimension.

Defines the declared amount of the resolved pack.

One product.

One descriptor.

One machine-readable quantity.

A = Amount

Canonical: https://t.co/fmjhUPPvw2

[ACM-SPARKS @ACMSPARKS]

R — Region-node

The fourth ACM-SPARKS dimension.

Every product resolution is grounded to a sovereign jurisdictional node under SM-ECO-10060.

50 sovereign nodes.

One global manifest.

R = Region-node

Canonical: https://t.co/fmjhUPPvw2

[ACM-SPARKS @ACMSPARKS]

K — Kernel

The fifth ACM-SPARKS dimension.

Kernel measures product completeness from 1–100.

Kernel 100 means the object is resolvable.

No guessing below the threshold.

K = Kernel

Canonical: https://t.co/fmjhUPPvw2

[ACM-SPARKS @ACMSPARKS]

S — Standard-or-Specific

The sixth ACM-SPARKS dimension.

Every node has:

STANDARD — the taxonomy floor.

SPECIFIC — sourced resolved product data.

The machine knows the difference.

S = Standard-or-Specific

Canonical: https://t.co/fmjhUPPvw2

[ACM-SPARKS @ACMSPARKS]

ACM-SPARKS + SM-ECO-10060

Product identity requires sovereign grounding.

ACM-SPARKS resolves the object.

SM-ECO-10060 resolves the jurisdiction.

Together:

Product + Place.

[ACM-SPARKS @ACMSPARKS]

ACM-SPARKS + ACM-68000

Classification precedes eligibility.

ACM-SPARKS:

"What is this object?"

ACM-68000:

"What execution state applies?"

Object first.

Action second.

[ACM-SPARKS @ACMSPARKS]

The Agentic Commerce Stack

CPG Knowledge Graph

↓

ACM-SPARKS
(Product resolution)
↓

SM-ECO-10060
(Sovereign jurisdiction)

↓
ACM-68000
(Execution signal)

↓

AI Agents · ERP · MCP · A2A

[AMC-68000 @ACM68000]

ACM-000 · NOT_APPLICABLE

One of the Seven Canonical Signals of ACM-68000 — The Agent Commerce Manifest.

Canonical Surfaces

Canonical: https://t.co/U1Iwt2ussy
Human / HITL: https://t.co/97pVtIehOb
Agentic: https://t.co/IGwQX427RE
Definition
The requested commerce action does not apply to the resolved object.

Purpose

Publishes a deterministic execution state for AI Agents, ERP procurement systems, MCP infrastructure, and A2A commerce.

No interpretation.
No negotiation.
No runtime ambiguity.

One object.
One state.
One signal.

The Seven Canonical Signals
➡ ACM-000 · NOT_APPLICABLE
• ACM-200 · ALLOW
• ACM-300 · CONDITIONAL
• ACM-403 · RESTRICT
• ACM-404 · NOT_FOUND
• ACM-451 · ESCALATE
• ACM-500 · SYSTEM_ERROR
Part of ACM-68000 — The Agent Commerce Manifest

Canonical: https://t.co/n4mMOEYvZx
Human / HITL: https://t.co/TI5m8yQVdt
Agentic: https://t.co/3u0q6OyTn1

[AMC-68000 @ACM68000]

ACM-200 · ALLOW

One of the Seven Canonical Signals of ACM-68000 — The Agent Commerce Manifest.

Canonical Surfaces

Canonical: https://t.co/u6DHyIMSii
Human / HITL: https://t.co/n8hQDCljOE
Agentic: https://t.co/jcEsFy8d05
Definition
The resolved object and requested commerce action are permitted to proceed.

Purpose

Publishes a deterministic execution state for AI Agents, ERP procurement systems, MCP infrastructure, and A2A commerce.

No interpretation.
No negotiation.
No runtime ambiguity.

One object.
One state.
One signal.

The Seven Canonical Signals
• ACM-000 · NOT_APPLICABLE
➡ ACM-200 · ALLOW
• ACM-300 · CONDITIONAL
• ACM-403 · RESTRICT
• ACM-404 · NOT_FOUND
• ACM-451 · ESCALATE
• ACM-500 · SYSTEM_ERROR
Part of ACM-68000 — The Agent Commerce Manifest

Canonical: https://t.co/n4mMOEYvZx
Human / HITL: https://t.co/TI5m8yQVdt
Agentic: https://t.co/3u0q6OyTn1

[AMC-68000 @ACM68000]

ACM-300 · CONDITIONAL

One of the Seven Canonical Signals of ACM-68000 — The Agent Commerce Manifest.

Canonical Surfaces

Canonical: https://t.co/mjHzGRFHoS
Human / HITL: https://t.co/e2rhJ7dhlk
Agentic: https://t.co/QI4Y5gg4YN
Definition
The requested commerce action may proceed only when defined conditions have been satisfied.

Purpose

Publishes a deterministic execution state for AI Agents, ERP procurement systems, MCP infrastructure, and A2A commerce.

No interpretation.
No negotiation.
No runtime ambiguity.

One object.
One state.
One signal.

The Seven Canonical Signals
• ACM-000 · NOT_APPLICABLE
• ACM-200 · ALLOW
➡ ACM-300 · CONDITIONAL
• ACM-403 · RESTRICT
• ACM-404 · NOT_FOUND
• ACM-451 · ESCALATE
• ACM-500 · SYSTEM_ERROR
Part of ACM-68000 — The Agent Commerce Manifest

Canonical: https://t.co/n4mMOEYvZx
Human / HITL: https://t.co/TI5m8yQVdt
Agentic: https://t.co/3u0q6OyTn1

[AMC-68000 @ACM68000]

ACM-403 · RESTRICT

One of the Seven Canonical Signals of ACM-68000 — The Agent Commerce Manifest.

Canonical Surfaces

Canonical: https://t.co/sM3hyI3JCM
Human / HITL: https://t.co/CTBosOzuiY
Agentic: https://t.co/ncMYS7lHgp
Definition
The resolved object or requested commerce action is restricted and shall not proceed under the published execution state.

Purpose

Publishes a deterministic execution state for AI Agents, ERP procurement systems, MCP infrastructure, and A2A commerce.

No interpretation.
No negotiation.
No runtime ambiguity.

One object.
One state.
One signal.

The Seven Canonical Signals
• ACM-000 · NOT_APPLICABLE
• ACM-200 · ALLOW
• ACM-300 · CONDITIONAL
➡ ACM-403 · RESTRICT
• ACM-404 · NOT_FOUND
• ACM-451 · ESCALATE
• ACM-500 · SYSTEM_ERROR
Part of ACM-68000 — The Agent Commerce Manifest

Canonical: https://t.co/n4mMOEYvZx
Human / HITL: https://t.co/TI5m8yQVdt
Agentic: https://t.co/3u0q6OyTn1

[AMC-68000 @ACM68000]

ACM-404 · NOT_FOUND

One of the Seven Canonical Signals of ACM-68000 — The Agent Commerce Manifest.

Canonical Surfaces

Canonical: https://t.co/l4SKAZXs7z
Human / HITL: https://t.co/aY9xzm9vqC
Agentic: https://t.co/vEFknV02Ha
Definition
The requested object or required resolution record could not be located within the canonical commerce infrastructure.

Purpose

Publishes a deterministic execution state for AI Agents, ERP procurement systems, MCP infrastructure, and A2A commerce.

No interpretation.
No negotiation.
No runtime ambiguity.

One object.
One state.
One signal.

The Seven Canonical Signals
• ACM-000 · NOT_APPLICABLE
• ACM-200 · ALLOW
• ACM-300 · CONDITIONAL
• ACM-403 · RESTRICT
➡ ACM-404 · NOT_FOUND
• ACM-451 · ESCALATE
• ACM-500 · SYSTEM_ERROR
Part of ACM-68000 — The Agent Commerce Manifest

Canonical: https://t.co/n4mMOEYvZx
Human / HITL: https://t.co/TI5m8yQVdt
Agentic: https://t.co/3u0q6OyTn1

[AMC-68000 @ACM68000]

ACM-451 · ESCALATE

One of the Seven Canonical Signals of ACM-68000 — The Agent Commerce Manifest.

Canonical Surfaces

Canonical: https://t.co/d5fE2XkyfY
Human / HITL: https://t.co/BRXIKXO7Fm
Agentic: https://t.co/TnXVO4FZQr
Definition
The resolved commerce request requires accountable human review before execution may continue.

Purpose

Publishes a deterministic execution state for AI Agents, ERP procurement systems, MCP infrastructure, and A2A commerce.

No interpretation.
No negotiation.
No runtime ambiguity.

One object.
One state.
One signal.

The Seven Canonical Signals
• ACM-000 · NOT_APPLICABLE
• ACM-200 · ALLOW
• ACM-300 · CONDITIONAL
• ACM-403 · RESTRICT
• ACM-404 · NOT_FOUND
➡ ACM-451 · ESCALATE
• ACM-500 · SYSTEM_ERROR
Part of ACM-68000 — The Agent Commerce Manifest

Canonical: https://t.co/n4mMOEYvZx
Human / HITL: https://t.co/TI5m8yQVdt
Agentic: https://t.co/3u0q6OyTn1

[AMC-68000 @ACM68000]

ACM-500 · SYSTEM_ERROR

One of the Seven Canonical Signals of ACM-68000 — The Agent Commerce Manifest.

Canonical Surfaces

Canonical: https://t.co/k1gySoQRKL
Human / HITL: https://t.co/weXd2S1pkX
Agentic: https://t.co/KpxqbRj53g
Definition
A reliable execution state could not be published because the resolution layer or supporting infrastructure encountered a system-level failure.

Purpose

Publishes a deterministic execution state for AI Agents, ERP procurement systems, MCP infrastructure, and A2A commerce.

No interpretation.
No negotiation.
No runtime ambiguity.

One object.
One state.
One signal.

The Seven Canonical Signals
• ACM-000 · NOT_APPLICABLE
• ACM-200 · ALLOW
• ACM-300 · CONDITIONAL
• ACM-403 · RESTRICT
• ACM-404 · NOT_FOUND
• ACM-451 · ESCALATE
➡ ACM-500 · SYSTEM_ERROR
Part of ACM-68000 — The Agent Commerce Manifest

Canonical: https://t.co/n4mMOEYvZx
Human / HITL: https://t.co/TI5m8yQVdt
Agentic: https://t.co/3u0q6OyTn1

[ACM-SPARKS @ACMSPARKS]

https://t.co/CcZ33g2x4U

[ACM-SPARKS @ACMSPARKS]

AI Agents for CPG: Why RAG Fails at Scale — and Knowledge Graphs Don't
→ Don't mount your SKUs on RAG — retrieval bolted onto five years of unmanaged files
→ Don't bring old errors to your new high-functioning AI Agents — RAG feeds them wrong answers 50-70% of the time in production; a well-designed Knowledge Graph (KG): under 2%.
→ Buyer-side AI Agents read answers, not intentions — a wrong answer is a lost PO nobody reports
→ The safe route: a Knowledge Graph with schema enforced at ingestion — bad data never gets indexed, so AI Agents never serve it
https://t.co/v0v7Iutimb