Plain-language definitions

If a term here is not clear, that is a bug in the writing — tell us.

Filter terms

Clover and the five stages

Clover: This framework. A way of working with System, Human, and AI to produce meaningful outcomes, through Context, Direction, Execution, Outcome and Growth. Five stages, describing how human intent becomes an outcome the environment confirms, and what the next attempt starts from.
Stage: One part of the system cycle. Each stage has one job, and there are five. In the clover mark each stage is drawn as a leaf.
Leaf: A leaf of the clover mark. The picture has five. In the documents the parts of the cycle are called stages, and the fifth leaf represents Growth.
System: The reality the work happens in. Outcomes are the changes that get applied to it.
Actors: The two participants in any cycle — the human, who holds Direction and accountability, and AI, which supplies capability and execution and determines how work happens within human Direction. They work inside the system, which is the reality the outcome has to exist in.
System cycle: The five stages the actors run — Context → Direction → Execution → Outcome → Growth. A team runs it over and over.
Accountability: Being answerable for the outcome afterward. It cannot sit with AI, which can perform work and report on it but cannot carry the consequence. When execution moved to AI, accountability tended to move out of scope with it. Clover establishes it back in the system, on the human actor who can truly take up the role.
Context: Everything the work reasons from, and the stage the cycle starts in. It may be an existing system or the reality already established while a system is being built: data, behavior, history, constraints, evidence, and previous cycles.
Direction: The human-defined purpose and destination: what matters, the desired outcome, priorities, constraints, boundaries, what must not happen, and accountability. With real context available, Direction also points at where the answer probably is.
Execution: Where the outcome is pursued by working with the system, using the Direction decided and the Context available. AI determines and carries out how the work should happen, and the system's boundaries bind both actors — neither the human nor AI may violate them.
Outcome: What actually happened, as the real system shows it. It is both the failure and the success; an unfavorable Outcome is still an Outcome. A closed task, a passing build, or a confident report sits outside this on its own.
Growth: The fifth stage of the system cycle. Whatever the Outcome taught, at any size, carried back into Context. One wrong answer, understood and written down, counts. No repetition and no scale are required. A system that does not retrospect its growth will not produce better outcomes. What accumulates over time can sit with humans, AI systems, the system being worked on, teams and organizations, and none of it is owned by one actor.
The common clover: Three stages — Direction, Execution, Outcome. How AI is used almost everywhere today.
The lucky clover: Four leaves, with Context added and running first. The step that made the cycle work.
The growth clover: Five leaves. Growth is the fifth stage: preserve what the cycle taught and promote what has held. This is the framework.
The fifth leaf: Growth, drawn as the fifth leaf of the mark. The leaf also carries the unknown boundary of how far capability and learning may develop.

Everything else

Orchestration: Coordinating people, AI, tools, and context so that work produces an outcome the environment confirms, rather than each part doing its own thing.
Orchestration environment: The access layer between AI and the systems an organization already uses. Read-only connections to whatever those systems keep — their records, their history, their measurements, the working environments. For software that usually means MCP servers in front of the repositories, datasources, logs and environments. It is what feeds the Context stage.
MCP server: The software route into a system: a small service that gives an agent a scoped way to read one of them — a repository, a datasource, a log store, an environment. Read-only, and scoped to what the human driving the work already has access to. Systems that are not software need the same thing by some other means.
Capability: Anything that can do work: a human, a team, an AI, an agent, a tool, a system. Capability does not by itself grant authority or accountability.
Intent: What a human actually wants to achieve. The outcome, and not the task.
Output: What got produced — a file, a patch, a report. Distinct from the Outcome, which is the change in the real world that was wanted. Work can produce output and reach no outcome.
Evidence: What was actually done to check a claim — an assertion, one manual look, a repeatable test, a before-and-after measurement, or the original signal gone from the real environment. Say which. More on Outcome.
Experience: What was learned from one cycle — what was tried, what happened, what the evidence showed.
Expertise: A reusable pattern that emerges from several validated experiences. One cycle is not expertise.
Ownership: Within a task or piece of work, the named human accountable for the outcome they direct. Work can be delegated; ownership cannot. This does not transfer accountability for a model, product or deployment away from the organization that builds, releases or operates it.
Agent: An AI system that can take actions and use tools, rather than only produce text.
Agentic workflow: A designed loop of agent steps that repeats a known process. Orchestration differs in that it keeps what the outcome taught it.
Autonomy: In Clover, this refers only to how much of the path AI is allowed to determine inside human Direction. It never means ownership of purpose, acceptable risk, priorities, boundaries, or accountability. A more capable model does not create authority over the destination.
Delegated execution: The amount of operational work a human or organization chooses to have AI perform inside human Direction. It can expand or contract by context and evidence. It does not transfer purpose, acceptable risk, priorities, boundaries, the destination, or accountability to AI.
Telemetry: The signals a running system emits about itself. In software: logs, metrics, traces, error rates. In any other system: whatever it records about its own operation while it runs.
Blast radius: How much damage a change could do if it is wrong. A bigger blast radius means more human approval.
Non-production: Any setting that is not the live system — a copy, a test bench, a rehearsal, and in software local, test and staging. It can still contain sensitive data and is not automatically safe to expose.
Guardrail: A constraint that keeps work inside safe boundaries: a required approval, a scope limit, a check that must pass.
Boundary: A limit set by human Direction — how far the work may go, what it must not do, what needs approval first. The accountable human can widen it or narrow it as evidence comes in.
Obligation: A requirement that comes from law, regulation, or an organization's own rules. It sits outside Clover, so nobody inside a cycle can trade it away and no Direction can lift it. Adopting Clover neither satisfies an obligation nor removes one.
Root cause: The underlying reason a problem occurs. Distinct from the symptom, and from the workaround that hides it.
Workaround: Something that stops the pain without fixing the cause. Legitimate for stabilizing an incident, and not a destination.
Thrashing: Repeated confident attempts at a fix, none of which work. The signal to stop fixing and go back to Context.

Source of truth: docs/glossary.md