Manufacturing Industry Principles

What truths shape how value gets created and captured across the manufacturing industry?

The Value-Creation Function

The manufacturing industry transforms a set of inputs into outputs that produce measurable outcomes for customers. Name the function explicitly or every other dimension drifts.

INPUTS                  ──→  [ THE FIVE Ps ]  ──→  OUTPUTS              ──→  OUTCOMES
─────                                              ───────                    ────────
energy                                             goods                       order fulfilled
materials                                          batches                     yield captured
know-how                                           assemblies                  scrap avoided
machinery                                          DPP attestations            license to operate held
data                                                                           customer retained
                                                                              ↑
                                              measured-against ────────────── ┘
                                                  Performance (the dream)

The five Ps are the transformation engine. Performance — the dream of a plant that improves every cycle, hits its OEE target, ships on time, scraps nothing, attests its carbon, and keeps its operators safe — is what every outcome is measured against.

The Five Inputs, named

• Energy — the kWh the line consumes. Historically billed monthly. Now metered per-machine, per-batch, with carbon attached. Energy is the second BOM.
• Materials — the BOM. Increasingly required to be traceable to source by mandate (ESPR, CSRD, conflict-minerals rules).
• Know-how — the routing codes, the standard work, the recipe IP. The asset that compounds across cycles when it's codified; that decays when it lives only in operator heads.
• Machinery — the PLCs, the robots, the AGVs, the line equipment. Increasingly identity-bearing — cryptographic DIDs that let a machine attest, transact, and settle.
• Data — the OPC UA stream, the vision frames, the vibration spectra. The output that is also the input — every cycle's data improves the model that runs the next cycle.

First Principles

Nine truths that shape every other decision in the industry. Hold these and the rest follows.

• Standardised interfaces compound. Bespoke decays. The factory that runs on OPC UA, ISA-95, MQTT, MTConnect can swap any vendor without breaking the loop. The factory that runs on bespoke integrations buys lock-in at every layer. The compounding asset is the standard, not the application running on top.

• Throughput is fragile; small constraints cascade. A 5-minute changeover delay on the bottleneck machine costs an hour of throughput downstream. Goldratt's theory of constraints is older than the digital factory and still rules it. Find the constraint. Subordinate everything to it. Elevate it. Repeat.

• OEE composite is the gauge. Availability × Performance × Quality is the irreducible factory metric. Optimise one in isolation and the other two break — push availability and quality collapses; push quality and availability dies. The composite forces honest trade-offs.

• Data that doesn't change a decision is cost. A sensor that streams 10Hz to a dashboard nobody reads is overhead, not asset. Every sensor must have a decision it serves; every dashboard must have an action it triggers. Read Information Arbitrage — speed × accuracy × uniqueness is what makes factory data valuable.

• Energy is the second BOM. Once you meter per-machine + per-batch, energy stops being overhead and starts being a cost-of-goods line item with optimisation levers. The factory that doesn't meter energy is flying blind on margin.

• Carbon is now an FX rate. CSRD (EU) and equivalent regimes elsewhere force scope-3 reporting. Suppliers that cannot attest their carbon footprint lose tenders. Carbon attestation is a procurement requirement, not an environmental nice-to-have.

• Machine identity is the precondition for autonomy. A machine that cannot prove who it is cannot transact autonomously. A cryptographic DID per asset unlocks machine-to-machine settlement, attestation, and zero-trust OT mesh. Without it, every machine action requires a human to vouch.

• The factory IS the agent surface for physical AI. Software AI is bounded by the screen. Manufacturing AI is bounded by the actuator. The factory is where AI gets to touch the world — vision agents that stop the line, scheduling agents that re-sequence the queue, energy agents that defer demand. Bounded by physics, not by attention.

• Every product is an experiment. The VVFL loop holds — Intention → Awareness → Impact → Measure → Reflect. The product going out the door teaches the next product. The factory that closes the loop fastest compounds; the factory that ships and forgets repeats every mistake.

Why These Truths Are Immutable

Principle	Why immutable
Standardised interfaces compound	Network-effect physics — every additional standards adopter raises the value for every other
Throughput is fragile	Queueing theory — variance × utilisation = WIP, mathematically inescapable
OEE composite is the gauge	Three factors must multiply; isolating one cannibalises the others
Data without decision is cost	Storage + bandwidth + cognition costs are real; data is liability until it's leveraged
Energy is the second BOM	First law of thermodynamics; energy is materially consumed in production, must be priced
Carbon is an FX rate	Regulatory + financial-market dynamics; mandatory disclosure changes capital cost
Machine identity is the precondition for autonomy	Trust requires verifiable identity; cryptographic identity is the engineered minimum
The factory is the agent surface	AI is bounded by what it can affect; the actuator is the boundary
Every product is an experiment	Bayesian update — every cycle generates evidence; the choice is only whether to listen

What These Principles Disqualify

Naming the principles names what they reject.

• "Replace ERP with a chat interface." ERP holds the master data. Chat is an access layer. Confusing layer for system of record loses both.
• "AI optimises everything automatically." Without the OEE composite as the gauge, AI optimises whatever's loudest. Local optimisation breaks the system.
• "Ship and gather feedback in market." Closed-loop manufacturing fails fast on the line, not after the customer call. The factory is the feedback loop.
• "Custom integration is the moat." It is the anti-moat. Custom decays; standards compound. The moat is the recipe, the operator know-how, and the data history — never the integration code.
• "Decarbonisation is a corporate-PR program." Once carbon enters cost-of-goods, it is an engineering problem with operational levers. Treating it as PR loses tenders.
• "Every machine needs a generic LLM." No. Every machine needs a deterministic control loop with a probabilistic layer on top. Confusing the two ships unsafe products.
• "Decentralise the line." No. Decentralise the data plane. The line itself stays deterministic and local; the data and identity layer goes distributed.

How the Principles Cascade

Standardised interfaces compound
  └─► open data plane (UNS) is the moat
        └─► AI agents attach where the data is, not where the vendor is
              └─► the factory keeps optionality across vendor cycles

Throughput is fragile
  └─► find the constraint
        └─► subordinate everything to it
              └─► don't waste optimisation on non-bottlenecks

OEE composite is the gauge
  └─► compute it in real time on the line
        └─► operator sees it
              └─► micro-decisions get made on the same minute the event happens

Data without decision is cost
  └─► every sensor pays back via a named decision
        └─► dashboards only exist for actions
              └─► AI agents replace dashboards when the action is mechanical

Energy + carbon
  └─► meter per-machine, per-batch
        └─► price into cost-of-goods
              └─► engineer the levers
                    └─► attest the result on-chain for buyers and regulators

Machine identity
  └─► DID per asset
        └─► machine-payable APIs become safe
              └─► throughput-as-collateral becomes financeable
                    └─► autonomous production cells emerge

Boundaries These Principles Hold

Boundary	What the principle protects against
Vendor lock-in	Standardised interfaces compound — protects against single-vendor capture
Local optimisation	OEE composite — protects against pushing one factor at the cost of the other
Vanity metrics	Data without decision is cost — protects against dashboard sprawl
Greenwashing	Carbon is an FX rate — protects against unverifiable sustainability claims
Autonomy theatre	Machine identity — protects against "autonomous" systems that need humans
Optimisation without constraint focus	Throughput is fragile — protects against working on the wrong machine

Context

• Manufacturing Players — the community these principles bind
• Manufacturing Platform — the technology that enforces them
• Manufacturing Processes — the workflows that operationalise them
• Manufacturing Performance — the gauges that validate them
• Information Arbitrage — why factory data captures value
• Standards — the interoperability layer that makes principles compound
• DePIN Actuators — the infrastructure layer that machine identity unlocks

Questions

• Which of the nine principles is the one most factories actively violate today — and what would have to change for them to stop?
• "Data that doesn't change a decision is cost" — apply this to your last three dashboards. Which survive?
• If carbon is an FX rate, which line item on the P&L moves the most when carbon repricing actually arrives?
• When machine identity is engineered, what new commercial structures become possible that aren't possible today?
• Standards compound. So why does every plant tour still reveal a different stack? What forces resist standardisation?