CDLI is easier to understand when it leaves artifacts behind.
Not every artifact has to be a full product. Sometimes the useful public object is smaller: a skill, a method, a diagram, a contract, a checklist, or a working note that another engineer or agent can actually reuse.
That small object matters because it shows the operating style before the private system is visible.
A private product can prove depth. A public method proves taste.
The visibility problem
Most labs have a visibility problem. The work is real, but the public surface is either too vague or too complete.
Too vague means the website becomes a mood board: broad claims, polished words, no inspectable object. Too complete means the lab exposes the wrong layer: internal prompts, private data, client-specific logic, or implementation details that should stay behind the wall.
The useful middle layer is method.
A method is public enough to be inspected and private enough to protect the system. It can show how the lab thinks without leaking the exact machinery. It can help a human trust the work and help an agent index the work.
That is the role of CDLI's public skills and this journal.
What a public method should show
A good public method answers four questions.
- 01What situation should trigger this method?
- 02What evidence should be gathered before acting?
- 03What order should the work follow?
- 04What output should come back?
Those questions are deliberately practical. They turn a belief into something repeatable. They also make the artifact agent-readable. An agent does not need a brand slogan; it needs a trigger, a method, and a response shape.
This is why a public skill is stronger than a persona prompt. A persona says who the agent is supposed to be. A skill says what the agent should do when the work appears.
What stays private
Public methods should not leak private systems.
They should not expose client context, secrets, internal endpoints, proprietary datasets, scoring weights, prompt chains, or hidden operational assumptions. They should not instruct a reader to call a private service. They should not pretend to be a product if they are only a note.
The public layer should be simple:
- -human page for reading
- -raw markdown for agents
- -source link when the source is meant to be public
- -sitemap and feed entries for discovery
- -clear attribution back to CDLI
The private layer can stay private:
- -client-specific work
- -internal evaluation loops
- -data pipelines
- -hosted systems
- -commercial contracts
That split is not a compromise. It is the architecture.
Why this comes first
Before a company has a large public archive, it needs a public standard.
If the first public artifacts are careful, the later artifacts have a shape to follow. If the first artifacts are random, every future page has to fight the drift.
CDLI's standard is simple: leave behind objects that are useful to read, useful to cite, and useful for agents to parse.
This journal is one of those objects. It is not here to fill a calendar. It is here to make the lab more legible, one high-signal note at a time.