This site explains one idea and tests it from three sides. A software factory turns a reusable engineering pattern and an explicit product specification into independently tested software artifacts through governed, repeatable assembly stations. It starts with one function and four deterministic stations, reads the WatchAlgo repositories to see how far a repeated algorithm pattern was scaled, and then builds a synthetic insurance service the way an enterprise factory would, with contracts, authorization, release gates and owners.
What this site is, and what it is not#
It is a teaching map to adapt, not a standard. The chapter order follows the questions a team meets when it moves from doing one thing by hand to producing many similar things with checks: first the primitive, then a real case at scale, then the enterprise version, then measurement and design. The patterns, station graphs and simulators are the author's teaching reconstruction unless a page labels them otherwise.
How the WatchAlgo case study was checked#
WatchAlgo is the author's own algorithm-learning product, so it can be named and its repositories could be read. They were read, never changed: nothing in them was written, built, installed or deployed for this site. Every WatchAlgo statement on these pages carries one of three labels.
| Label | What it means | Example |
|---|---|---|
| Observed in the repository | A file, folder, field or commit that anyone with the repository can open | the three files inside each problem folder |
| Reported by the author | The author's operational account, not measured here | 3,246 implemented problems; runs of 30 or more hours until token limits |
| Teaching reconstruction | An architecture this site proposes to explain the idea | the station map and Example B's toy unit |
The full matrix, with paths and commits, is summarized on Sources and told as a story in the WatchAlgo chapter. Where the repository does not show something (for example the prompts or the run logs of the long sessions), the pages say so instead of filling the gap.
The insurance example is synthetic#
The enterprise chapters build a service for an insurer that sells commercial property insurance through brokers. This is an illustrative reconstruction for teaching. It follows the shape of a real project by the author, but the insurer, people, rules, fields and every number are invented, and nothing here reports the quality of a real system or model. In every example the model is a scripted stand-in, so the results show how the factory mechanics work, not how good any model is. Two sentences hold throughout: the factory builds the underwriting software but never makes an underwriting decision, and in the software it builds, the model recommends; a person decides.
The label convention#
Claims of different kinds are kept apart with a bold lead-in, so a reader always knows what kind of statement they are reading.
| Lead-in | Used for |
|---|---|
| Observed in the repository | WatchAlgo facts with a path and commit |
| Reported by the author | The WatchAlgo operational account |
| Teaching reconstruction | This site's proposed design, toy or simulator |
| Standard | What a standard or specification says, with version and date read |
| Product | What a product or open-source project documents, with date read |
| Pattern | A common practice that no standard requires |
| This site's recommendation | The author's advice, never presented as a standard |
| Simulated, illustrative | Any number from a lab or simulator |
Sister sites#
This site is one of a series on AI-native engineering by the same author. Each owns one subject and links the others where they meet, instead of repeating them.
| Site | What it owns |
|---|---|
| prompt.sammuthu.com | Instructions: how they are written, composed and defended |
| context.sammuthu.com | What the model can see now: selection, retrieval, budget |
| memory.sammuthu.com | Durable information and saved execution state |
| a2a.sammuthu.com | Messages between agents; A2A versus MCP |
| auth.sammuthu.com | Identity, delegation, enforcement and audit for every caller and every hop |
| guardrails.sammuthu.com | Enforceable checks around a probabilistic agent |
| observe.sammuthu.com | Logs, metrics, traces, audit events, replay |
| tools.sammuthu.com | Callable capabilities: schema, execution, MCP tools |
| mcp.sammuthu.com | Shared capabilities across teams: when MCP is justified, identity, authorization, governance |
| rag.sammuthu.com | Retrieval from SQL rows to production: embeddings, vector indexes, hybrid search, reranking |
| schema.sammuthu.com | Typed contracts: shape, validation, repair, evolution |
| orchestrate.sammuthu.com | The control plane: what runs next |
| reliable.sammuthu.com | Acceptable outcomes despite failures and load |
| harness.sammuthu.com | The runtime around a model call |
| eval.sammuthu.com | Measurement: graders, golden sets, release gates |
| factory.sammuthu.com (this site) | Specs and reusable patterns turned into independently tested software through governed assembly stations |
| diff-to-deploy.sammuthu.com | Shipping a change with coding agents |
| ask-to-release.sammuthu.com | Planning a change before the first meeting |
Three neighbours share the most ground with this site. Harness Engineering explains the controls around one agent's run; this site explains the production system that many such runs feed. Diff to Deploy follows one change from a diff to production; this site creates many products from a pattern and hands each one to that path. Evals for agentic software explains how to measure quality; this site invokes those measurements as gates and decides what advances.
What is original#
The chapter order, the examples, the simulators, the diagrams, the labs and the explanations are original to this site. Standards, product documentation and outside articles are cited and summarized in the site's own words, never copied; the dates they were read are on Sources. The WatchAlgo excerpts in the problem explorer come from the author's own repository and keep their file paths and commits.