# Foreman — Map of Content This map shows one chain of ideas. First, the unwritten know-how of experienced factory workers is a stock that becomes smaller. Second, the number of skilled workers decreases, and this makes the loss costly at this time. Third, agents that use domain packs for an industry can capture and use this know-how again. Fourth, a channel through institutes can move the software to mid-market manufacturers. Fifth, the company is a *platform* business only if the pitch is correct about deployment work and data rights. The pages are a due-diligence map and also a company description. Start at one page and use the wikilinks. > [!abstract] How to read this map > Read from bottom to top. First is **a thesis**: tacit knowledge goes with the workers. Second is **a forcing function**: skilled workers go faster than new workers can replace them. Third is **a wedge**: domain packs that institutes sell. Fourth is **a mechanism**: a shared platform layer below thin vertical layers. Last is **a company** that has deployments. The company has no proof that the mechanism works. To see only the numbers, go to [[#The company, compressed]]. > [!info] At a glance > - **What** — a vertical agent platform for manufacturing operations: quality, production, equipment, R&D, ESG and supply chain. > - **Core idea** — capture the know-how of experts one time and use it in many deployments. > - **Wedge** — domain packs for an industry. The company and institutes make them together, and institutes sell them. > - **Market** — sources document the pain at a scale of one trillion dollars. The sources do **not** give a dimension for the market of this software. > - **Status** — seed funding. The company has paying customers, signed instruments, one formally accepted project and a small team. > [!warning] About this map > "Foreman" is a **codename**. This map deliberately omits the company, its personnel, customers, partners, investors and all financial and contract figures of the company. Thus the map is safe to share. Public statistics show their source. If a number can identify the company, the page gives a qualitative description. --- ## 1 · Tacit knowledge is a stock that becomes smaller, and it has a price Operating know-how in manufacturing industries is in the workers and not in the systems. When veterans retire, their diagnostic memory goes with them. The plant then pays with slower troubleshooting, lower quality and longer downtime. External sources document the pain. No source shows that the pain becomes willingness to pay. > [!tip] The sharpest version > Knowledge is a *stock* that attrition of workers makes smaller. The software changes it into a *flow* of decisions. The flow continues after the workers who made the knowledge have gone. → [[The Tribal Knowledge Thesis]] → [[What the Downtime Number Actually Means]] → [[Agent Skills as Codified Domain Expertise]] → [[Knowledge Graphs for Industrial Data]] ## 2 · The forcing function is a decrease in skilled workers, and recovery takes longer External research reports that the average recovery time from downtime became much longer in five years. The research also states that skilled workers went away. As a result, the plants lose skills and knowledge that they need. The headline loss figure gives the cost of the problem for the largest firms. It does not give a market for this vendor. > [!danger] Do not use the downtime statistic for the market dimension. > The figure of approximately $1.4–1.5T each year is for unplanned downtime at the 500 largest companies. Nobody calculated what share of it comes from the loss of knowledge. Nobody calculated what share of that is for software. → [[What the Downtime Number Actually Means]] → [[Industrial AI Unit Economics]] ## 3 · The wedge is the vertical domain packs that institutes sell The company does not sell a horizontal copilot. For each sector, it makes domain packs. Domain packs have a database of knowledge and validated agents. The company makes each one with an institute. The institute gives data, credibility and introductions to its members. This is the strongest asset of the company. It is also the source of the primary dependencies of the company: subsidy, data that partners own, and concentration in one sector. > [!tip] Why institutes matter > In this type of market, trust moves through associations. A competitor from a different country cannot copy this channel quickly. A horizontal competitor cannot copy it quickly. → [[Domain Packs and the Vertical Playbook]] → [[Institute Channel and Subsidy Dependence]] → [[Deployment Model Versus Target Buyer]] → [[The Age of Vertical Models]] ## 4 · The mechanism: is the company a platform or a set of solutions? The pitch is a shared platform layer with thin vertical layers on top, and a deployment time of days. The facts show a competent stack that the company mostly assembled from other parts. The strongest part of the stack is classical manufacturing ML. The company makes each deployment for one customer during months. The valuation depends on this one question. > [!important] The one test > Look at the deployment work for each customer during a period of time. Look also at the share of the work that goes into shared code. If the work becomes less, the platform claim is correct. → [[Platform or Solutions - The Central Question]] → [[The Four Layer Stack and What Is Commodity]] → [[The Evaluation Gap]] → [[Delivery Economics and the Scaling Ceiling]] → [[Deployment Velocity]] → [[Consultancy-to-Platform Transition]] ## 5 · The moat is thinner than the deck states The own roadmap and contract terms of the company contradict the claim of "exclusive proprietary data". Partners or customers own the data. The company can give the results of model training to customers. Each customer has an isolated model, so use does not add up. Other companies can copy the technology in months. Nothing shows that the IP exists, and other parties can possibly claim rights to it. The company keeps the channel and the domain know-how of a small team. > [!danger] Contradiction between sources > The pitch states exclusive data in some industries. The roadmap and the contracts show fewer industries, and other parties hold the rights. See [[Claims Versus Evidence]]. → [[Borrowed Data Is Not a Moat]] → [[IP Encumbrance Map]] → [[Open-Source Dependencies and Licence Exposure]] → [[Competitive Landscape]] → [[Data Moat]] → [[Defensibility Principles MOC]] ## 6 · How the company earns revenue, and what stops the bigger customers The revenue exists, but services are the larger part of it. Subsidy causes part of the revenue. Some cash depends on milestones. The biggest prize customers are in electronics supply chains. They are behind a security-certification gate. The company did not go through this gate. > [!warning] Quality of revenue > The company gives more than one version of the revenue for the same year. Nothing supports the revenue estimate. The company does not put the deployment costs in the cost of goods sold. Thus the confidence in the metrics of the company is lower. → [[Quality of Revenue]] → [[The Security Certification Gate]] → [[Land-and-Expand in Enterprise AI]] ## 7 · The team and the plan A founder with domain credibility and a small team have a roadmap. At the unit costs of the company, the budget for the roadmap is much too small. The central risks for the team are the concentration of key persons and an engineering team that is not transparent. → [[Team Depth and Key-Person Concentration]] → [[Roadmap Budget Realism]] → [[Key-Person Risk in Deep Tech]] --- ## The diligence layer > [!abstract] How the diligence pages fit together > [[Claims Versus Evidence]] shows where the narrative and the documents disagree. [[What Must Be True]] changes the map into ten conditions that tests can show to be incorrect. [[Dispositive Questions]] lists the questions for the company, with green answers and red answers. [[Open Questions and Numbers to Reconcile]] has each contested figure and the decision that it moves. [[Evidence Hierarchy]] and [[Technical DD Framework]] are the frame for the grades. ## The company, compressed > [!abstract] The company, compressed > **Founding** — a young company. A researcher of manufacturing ML started it. The team is small and has different skills. The company is in a manufacturing economy that technology leads. > **Traction** — about twelve signed instruments, a small number of live deployments and one formally accepted project. The company resold a textile-domain model. The company has a revenue-share agreement with an institute. The revenue is small, and services are the larger part of it. > **Funding** — a seed investment, which investors mostly committed. The price is for a platform narrative. This map intentionally omits the terms. > **Reality check** — the company has *no proof* for these points. The deployments share the platform layer. The company owns or can use again the data and models behind each vertical. Customers see measurable results in production. Customers with a subsidy can pay the full price. The protection of data at the company goes through the certification of top-tier customers. The fine-tuned model is better than retrieval alone. > **Proven** — the engineering work, the domain credibility of the founder, and a channel that competitors do not have. ## Concept map ```mermaid graph TD A["The Tribal Knowledge Thesis"] --> B["What the Downtime Number Actually Means"] A --> C["Domain Packs and the Vertical Playbook"] C --> D["Institute Channel and Subsidy Dependence"] C --> E["Platform or Solutions - The Central Question"] E --> F["The Four Layer Stack and What Is Commodity"] E --> G["Delivery Economics and the Scaling Ceiling"] E --> H["The Evaluation Gap"] F --> I["Open-Source Dependencies and Licence Exposure"] C --> J["Borrowed Data Is Not a Moat"] J --> K["IP Encumbrance Map"] G --> L["Quality of Revenue"] D --> L G --> M["Roadmap Budget Realism"] M --> N["Team Depth and Key-Person Concentration"] L --> O["What Must Be True"] J --> O O --> P["Dispositive Questions"] class A,B,C,D,E,F,G,H,I,J,K,L,M,N,O,P internal-link; ``` ## The argument in one breath > [!tip] The argument in three sentences > Factories do not keep the [[The Tribal Knowledge Thesis|unwritten know-how]] of their veterans when [[What the Downtime Number Actually Means|downtime costs the most]]. Thus [[Domain Packs and the Vertical Playbook|vertical domain packs]] that [[Institute Channel and Subsidy Dependence|trusted institutes]] sell are a sensible wedge. The company is a software business or a [[Platform or Solutions - The Central Question|solutions business]]. This depends on [[Delivery Economics and the Scaling Ceiling|deployment work for each customer]] and on [[Borrowed Data Is Not a Moat|data rights for the data that the company claims as a moat]]. [[What Must Be True|Ten testable conditions]] must be correct first. Until then, the honest view is a credible vertical-AI team with a channel and an unproven platform. ## Related - [[Industrial AI MOC]] - [[AI Agent Vertical SaaS DD MOC]] - [[Consultancy-to-Platform Transition]] - [[Industrial AI Unit Economics]] - [[Defensibility Principles MOC]] - [[Sovereign AI Positioning]] - [[Digital Twins]] - [[Maps of Content (MOCs)]]