🧭 Sketch your architecture to find gaps in understanding


Oct 18, 2026 | Read time: 11 mins.

Kevin Donovan · Architects' Hub
​
Forwarded this email? Subscribe to Architects’ Hub for the next issue.

Hey Reader welcome to this weekend's edition!

Here’s what I heard. Where is it wrong?

You know that feeling after a useful discovery meeting?

The right people were there. Their explanations made sense. You have several pages of notes and a fairly clear picture of the architecture.

Then you try to draw it.

One arrow in, you need to know whether the caller waits for a response. Nobody mentioned that. A little further along, 2 systems both seem to own the customer record. The nightly batch has consumers you can’t name.

The confidence you left with starts to look a little optimistic.

​Dan Hughes described a version of this recently. After meeting subject matter experts about a new architecture, he sat down the next morning to diagram what he’d learned. Drawing was how he explored his understanding. The document came out of that work.

I liked his example because it described something I keep coming back to: structured thinking clarifies the picture for us and gives other people something they can correct.

Drawing does it. Writing can do it too. The useful part is having to make your understanding specific enough to question.

This week we explored that through sketches, corrections, and a short routine. Today, let’s put it to work.

In this edition

  • Why a sketch reveals gaps that notes can carry.
  • How to invite useful corrections.
  • A copyable Draw-It-Back worksheet.
  • One worked example, with 6 questions and what their answers change.
  • When to write instead, and what to do with silence.
  • Community Corner.

Let’s get into it.👇

________

🧭

________

​

The picture has to join up

Each expert can give you an accurate account of their part while leaving the connections between those parts unresolved.

That becomes more obvious when you draw.

Every line needs 2 ends. “Integrates with payments” may be enough to get through your notes. On the page, you have to choose a caller, a receiver, and a direction. Then you can ask what happens if the receiver is unavailable.

Every box needs a boundary. “Customer” could mean the current customer profile, a billing account, or an address captured when an order was placed. Put those meanings in the same box and you may hide a distinction the implementation depends on.

Every flow needs a route through the system. “The batch feeds the warehouse” is a useful start. Following the flow may reveal a reconciliation job and a regulatory report that also depend on it.

A diagram can still hide things. A clean arrow labeled “integration” can be just as vague as the sentence it replaced.

The test is whether drawing made you state something you can now check.

When a gap appears, you may have forgotten an answer. You may have misunderstood it. Or the meeting may never have established it. Those need different follow-ups, which is why checking your notes comes before sending your questions.

Drawn to explain. Drawn to find out.

Both are useful jobs.

When I draw to explain, I choose the detail for the reader. I remove distractions, use consistent terms, and make the route through the picture easy to follow. The question is: can someone else understand the point?

When I draw to find out, I leave uncertainty visible. A box may have 2 possible owners beside it. An arrow may stop at a question mark. The first audience is me, followed by the people who can correct it. The question is: what have I assumed?

The same drawing can move between those jobs. Explaining it to someone can reveal a new gap too.

The trouble comes when we make it look settled before we’ve checked what it says. A clean template gives an assumption a surprisingly respectable appearance.

Keep the rough version. It records the questions the tidy one should be able to answer.

________

🧭

________

​

Give people something specific to correct

“Any questions?” asks people to find the uncertainty for themselves.

“Does Payment call Order here, or have I drawn that backward?” gives them a particular claim to check.

Imagine sending a rough sketch to the people from your discovery meeting. One reverses an arrow. Another tells you that a batch has been retired. A third adds the system that actually handles exceptions.

You now have corrections to a shared picture, rather than 3 more accounts you have to reconcile from memory.

The invitation matters:

“Here’s what I heard. Where is it wrong?”

You’re making your understanding available for review. That is different from asking people to approve a finished design.

A few habits help:

✅ Share it while it’s still easy to change. A photograph of paper or a whiteboard can be enough.

✅ Leave the questions visible. Point to the uncertain connection instead of apologizing for the whole drawing.

✅ Ask one specific question at a time. Ownership and failure handling often need an explanation, not a yes or no.

✅ Call it v0.1. Make its provisional status clear, without assuming the label alone will make people comfortable disagreeing.

✅ Thank the person who finds the consequential error. Do it where the others can see. You’re showing them what useful participation looks like.

Then update the picture and return it to the people who helped. They should be able to see what their correction changed.

________

🧭

________

​

Your Draw-It-Back worksheet

Use this after your next discovery conversation. The fields below are deliberately stacked so you can copy them into a note or email without wrestling with a table.

You need a blank page, your meeting notes, and a 20-minute timer.

If you have time before the meeting, spend an optional 10 minutes sketching what you expect. Dan described doing this too. Label it as an expectation so familiarity doesn’t turn into assumed fact.

Set up the page

Topic or flow:​
Write the outcome you’re trying to understand.

Scope:​
Write the part of the system you’re looking at.

People who can check it:​
Name the people who operate, build, or own that flow.

First 10 minutes: notes closed

Draw these 4 things, in order:

1. Boxes​
Draw the systems and teams. Put an owner beside each. Use a ? where you don’t know.

2. Lines​
Connect the boxes. Show who calls whom and which way information travels. Mark any uncertain connection.

3. Boundary​
Outline the part your team is responsible for. Mark the edges you can’t explain.

4. One flow​
Trace an outcome end to end: order placed to order fulfilled, or claim filed to claim paid. Include a failure point.

Don’t fill a gap with the arrangement from your last org. Mark it.

Memory is the starting point here, not evidence that something is true.

Remaining 10 minutes: check and question

Open the notes. Replace question marks only where the notes support an answer.

For each consequential gap, copy this small block:

Gap:​
What is uncertain?

Question:​
What exactly do you need to establish?

Ask:​
Who is best placed to answer?

Consequence:​
What would the answer change in your drawing or decision?

Repeat the block for the next gap. Start with the 3 that most affect the main flow.

At 20 minutes, stop. Send the rough version with the invitation above. Keep it beside v0.2 when the answers arrive.

Read the questions, not just the count

On Thursday I suggested question-count bands as prompts. They aren’t a measurement of understanding, and I wouldn’t turn them into a score.

Few questions might mean prior knowledge, a simple scope, or assumptions you haven’t noticed. Many might mean a broad discovery session.

One unknown about recovery after payment can matter more than 10 minor naming questions.

Ask: which missing answer could change the design, an owner’s commitment, or what happens when the system fails?

That is where to begin.

________

🧭

________

​

One sketch, 6 useful questions

Let’s work through a fictional discovery. These answers are illustrative, not a client story.

You leave with 4 systems in mind: Order, Payment, Customer, and Fulfillment. The initial picture is simple: Order requests payment authorization, then sends work to Fulfillment. Customer supplies an address. Someone also mentioned a warehouse extract.

Here is what the first sketch makes you ask.

1. Does Order wait for Payment, and who owns a retry?

The answer: Authorization is synchronous. Order owns retries and reuses the original idempotency key.

What changes: Label the connection and its retry owner. A generic “payment integration” arrow now carries a specific behavior. You can ask whether the implementation actually prevents duplicate requests from producing duplicate effects.

2. Which customer record supplies the delivery address?

The answer: Customer holds the current address. Order saves a snapshot at checkout.

What changes: Draw the snapshot as part of Order. Changing the customer profile later doesn’t silently change the delivery address on an existing order.

Two copies of an address are no longer automatically a conflict. They have different purposes.

3. Does the warehouse read a batch or events?

The answer: The warehouse consumes order events. A nightly batch remains for reconciliation.

What changes: Draw 2 paths and name their purposes. “We’ve moved to events” does not establish that the batch can be removed.

4. Who still consumes the batch?

The answer: Finance reconciliation and a regulatory-reporting job.

What changes: Add those consumers outside the original 4-system picture. Any retirement plan now needs their owners involved.

This is why the first boundary is provisional. Following the flow can change the scope of the conversation.

5. Who owns the vendor adapter?

The answer: Platform owns the API gateway. Fulfillment owns the adapter beyond it and its support rotation.

What changes: Put the ownership boundary between gateway and adapter. The next incident has a more useful starting point than “it’s somewhere in the platform.”

6. What happens after authorization if fulfillment fails?

The answer: In this example, payment has been authorized but not captured. Order records a pending state. An operations queue resolves the exception or requests a void of the authorization.

What changes: Add the exception path and queue owner. Mark authorization separately from capture. A failure after capture would need its own answer.

The useful change isn’t simply that v0.2 has more boxes. It now shows behavior, ownership, and dependencies that were missing from the first account.

Some answers will need checking against code, configuration, or operational evidence. A confident reply is a lead to follow, not automatic proof.

________

🧭

________

​

Sometimes the right shape is a page of text

A drawing is good at showing relationships. It is less good at explaining why you accepted a particular downside.

For a decision, try a short statement with these fields:

Choice: What are we proposing?

Alternatives: What else did we consider?

Accepted downside: What are we prepared to carry?

Owner: Who agrees to carry it?

Revisit condition: What evidence would make us reopen the choice?

Then ask which statement is wrong, incomplete, or unsupported.

For a quality requirement, write the condition and the response you need. “Fast” is hard to correct. A named operation, workload, and response-time target give people something they can test and negotiate.

The same discipline applies: make the claim clear enough that someone can improve it.

And sometimes nobody replies

Silence can be frustrating, especially when you’ve deliberately sent something unfinished.

Start with the ordinary explanations. Did people have time? Was the request clear? Did they know you needed an answer?

Then narrow the ask.

If the picture is too broad: point to one arrow and ask one named person about it.

If the people who know are missing: include someone who operates or supports the flow, not only someone who designed it.

If ownership is unclear: ask who can establish it. Keep the gap visible until that happens.

If people hesitate to disagree: offer a short walkthrough and correct your own assumptions openly.

You don’t yet know which explanation applies. Avoid turning a quiet thread into either approval or evidence that nobody cares.

Record what was confirmed, by whom, and what remains uncertain.

________

🧭

________

​

Community Corner

​Dan Hughes gave this week its starting point: using a diagram to think through a new architecture, with documentation emerging from the work. His habit of sketching before a meeting, when time allows, also adds a useful first step to the routine.

​Daniel Douglas raised the related question of what it takes for a platform to deliver value. His post challenged the assumption that an unsuccessful platform must have started without anyone asking what problem it should solve. That question may have been asked repeatedly. Getting to value still takes leadership, technical work, and change.

My reply focused on use: technology that isn’t used, or is underused, can’t deliver its intended value. The connection to this week is practical. A drawing earns its place when it helps someone understand, question, or act.

Thank you both for giving us useful things to work with.

Dan joined Tuesday’s discussion with the reminder that the conversation a diagram helps you have can begin inside your own head.

The replies also improved the routine:

​Jordan Corn: explain what happens along the arrows.
In Thursday’s discussion, Jordan described preparing something he can share as a way to find where he understands and where he’s guessing. His addition is now in the worksheet. On Tuesday, he also described the exploratory drawing gradually becoming a useful document as more people correct it.

​Erin Allen: make the second pass a habit.
In that same Thursday thread, Erin described carrying a paper notepad and debriefing daily from paper into digital form. She said she was adding the pre-meeting sketch. I like the deliberate second encounter with the material, rather than leaving the notes where they landed.

​Rene Hjortskov Nielsen: the sketch may need a guide.
His Wednesday response pushed back on email as the default. A walkthrough can establish the semantics before anyone can usefully challenge the drawing. I’ve made that distinction explicit above.

​Grant Ecker: the document comes out of collaboration.
In Friday’s discussion, Grant contrasted a document produced through collaboration with one used to tell everyone else how to build. That is a useful test of how the artifact was made, as well as what it contains.

​Anders W. Tell: keep the knowledge that needs to travel.
Also on Friday, Anders challenged a reading that undervalues diagrams as records. They communicate across people, agents, and time. His correction belongs alongside the week’s emphasis on thinking: use the drawing to learn, then leave a record someone else can understand.

Thank you, Dan, Daniel, Jordan, Erin, Rene, Grant, and Anders. This is what I want the Hub to make room for: a useful starting point, then other practitioners making it more complete.

________

🧭

________

​

Try it at your next discovery meeting

Give yourself 20 minutes afterward.

Draw the boxes, lines, boundary, and one flow. Check your notes. Choose the questions whose answers could change what you do.

Then send it:

“Here’s what I heard. Where is it wrong?”

Keep the first version. When the corrections come back, you will have a useful record of how your understanding changed.

Hit reply and tell me: what was the most consequential gap your sketch revealed?

I’d like to hear what the page made you notice.

Until next Sunday,

Kevin

________

________

Kevin Donovan​
I write Architects' Hub every Sunday for senior architects and technology leaders.
You're getting this because you subscribed at architects-hub.kit.com.
Know someone who’d find this useful? Forward it to them.

​

Korsvejen 9, Viborg, Viborg 8800
​Unsubscribe · Preferences​

Architects' Hub

A Sunday newsletter for senior and enterprise architects. One decision, one practical model, one tool you can use Monday. Written by Kevin Donovan. Join 2,500+ architects. One email every Sunday.

Read more from Architects' Hub
Three doors slightly ajar in a row. Through them, a telephone, a stack of documents with a magnifying glass, and an arrow curving back on itself. A single coin rests on the floor in front of the doors.

Nov. 22 2026 | Read time: 11 mins. Kevin Donovan · Architects' HubForwarded this email? Subscribe to Architects’ Hub for the next issue. The room went quiet Hey , Two Sundays ago, our hypothetical pilot case got its commitment record. Last Sunday it got a walkthrough, and the operating owner signed with 3 tests on record. The AI assistant now drafts delayed-order replies for 3 teams, and the drafts are good. This week someone proposed letting it do more than just draft… In this edition The...

Five simple figures in a ring around a small central box, each holding one object: a magnifying glass, a signpost, a wrench, a mop, a headset. One figure carries a leaning stack of boxes.

Nov 15, 2026 | Read time: 11 mins. Kevin Donovan · Architects' HubForwarded this email? Subscribe to Architects’ Hub for the next issue. Tuesday, 10:40 am. Hey ,Last Sunday we left a hypothetical customer-service assistant with a commitment record and unanswered questions about capacity, data and support. This week, move the story forward 3 weeks. The team has agreed to a bounded trial on 1 queue. Agents still review every draft. Wider rollout is waiting. I want to spend this edition with the...

A working terminal with four loose connections labeled Funding, Operations, Data and Authority.

November 8, 2026 | Read time: 9 mins. Kevin Donovan · Architects' HubForwarded this email? Subscribe to Architects’ Hub for the next issue. Hey ,In this edition of Architects' Hub: What the service lead heard when the sponsor said, “We should roll this out.” A commitment record filled in for one example customer-service queue, PLUS a blank copy for yours. How to take a conditional decision back to the sponsor who wants a date. Community Corner: Santiago Jesus Lasso Valenciano on the work that...