Figure 6.11 The final portion of the Relationship Loop workflow decomposition, showing function A5. The diagram captures the learning loop that concludes each execution cycle by transforming approved communications, updated voice patterns, and decision records into governance outputs for future iterations. Following a human judgment checkpoint, the workflow produces the weekly archive, emerging signals for the thinking loop, and an updated voice file, illustrating how operational execution continuously strengthens subsequent cycles.

What makes it a different example is what A3 produces. The weekly session doesn't only output approved emails and updated deals. It also produces voice patterns, pricing rationale, and objection handling, extracted from the decisions the humans actually made. Those get written into stores: a voice file, a pricing guide, an objection bank. And those stores generate the controls that govern the next run of A2. The voice file is the clearest case. It leaves A3 as an output and re-enters A2 as a control, the voice guide that governs how the next batch of drafts gets written. Follow that one artifact and you can see it cross from the output rail of one step into the control rail of another.

That's the pattern the notation makes visible here, and it requires no new shapes to represent it. The standard already says outputs can become controls downstream, and boundary objects already cross workflow edges. What this example surfaces is that a workflow's outputs divide into two kinds. Operational outputs leave the workflow and become inputs somewhere else: the approved email, the updated record. Governance outputs turn back and strengthen the controls that govern the workflow's own future runs: the voice file, the pricing guide, the objection bank. The second kind is what makes a workflow improve rather than merely repeat. The organization isn't accumulating information. It's converting human judgment into reusable controls, one cycle at a time.

This example also sharpens what a human judgment checkpoint is. A3 is not quality inspection. The humans aren't validating AI output like an inspector checking parts. They're thinking, with the relational AI as a partner in the thinking, and then committing to a course of action. The conversation isn't the checkpoint. The checkpoint is the moment of commitment the conversation leads to. That distinction matters for the notation: the relational AI collaboration and the human judgment checkpoint are adjacent but structurally separate, the mechanism of arriving at a decision on one side, the accountable act of deciding on the other.

And this is how Overlap defines its checkpoint. Another organization would define the same shape differently: legal approval, clinical judgment, pricing authority, sign-off of record. That's not a weakness in the notation, it's the core principle doing its job. Shape tells you the structural role, the point where human accountability is required before work proceeds. Label tells you the specific type, which each organization sets for itself. A board deciding what a human judgment checkpoint means in its own workflows is doing governance work, and the notation gives it the shape to hang that decision on.

The last thing this example reveals sits above the single workflow. A workflow is one closed transformation. An organization is a network of workflows connected by boundary objects. The relationship loop is one loop among several. A thinking loop that researches and publishes produces content that generates engagement signals, and those signals feed the relationship loop as inputs. The relationship loop surfaces objections and emerging demand, and those feed the thinking loop as inputs. The output of one loop is the input of another, and neither flow is one-way. Drawn at the organizational level, this isn't one enormous decomposition. It's a map of named loops exchanging boundary objects, each loop internally complete with its own ICOM structure, the organization emerging from how intelligence circulates between them. That's a shift of perspective, not a change to the notation. The same boundary object that marks the edge of a single workflow marks the coupling between workflows when you zoom out.

What the four examples demonstrate together

Across these examples, the notation does something neither flowcharts nor narrative descriptions can: it forces precision about the distinction between what governs a step and what executes it. In every case, the controls are where the governance intelligence lives: the prompts, the templates, the strategy frames, the facilitation questions. And in every case, the controls are the thing most likely to be undocumented, held in expert judgment rather than transferable artifacts.

That's no accident. It reflects something true about how organizations currently use AI. The mechanisms are visible: people can name the AI tool they used. The outputs are visible: there's a slide, a report, a document. But the controls that governed how the AI produced that output are often invisible, implicit, and unreplicable. HAWD makes them visible by structural necessity. You can't draw the diagram without naming the controls. You can't name the controls without building them. And you can't build them without deciding, explicitly, what governs each step.

The fourth example adds a dimension to that argument. Controls aren't only documented once and then applied. In a mature workflow, the workflow produces its own controls: human judgment, captured and written back into the stores and guides that govern the next cycle. That's a sharper account of organizational learning than "the AI gets better." The AI isn't training itself. The workflow is improving because its controls are improving, and the controls are improving because the organization is systematically converting judgment into reusable governance. HAWD can represent that directly, because a control is just an artifact, and an artifact can be the output of one step and the governing control of another.

That's the governance argument for the methodology. Documenting workflows is a general recommendation that produces general results. The sharper argument is that a notation system which forces explicit representation of controls creates a specific organizational discipline: knowing, for every AI-assisted step in every workflow, what's governing the AI, whether that control exists in a form that can be verified, replicated, and improved, and where the organization's own judgment is being converted into the controls that govern its future work.

Section 7 applies the notation directly to the governance requirements named by major AI governance frameworks: showing how HAWD provides the method layer those frameworks assume but don't supply.