# Layer2C Labs Design Challenge · Reference Experience
> Canonical page: https://layer2c.com/design-challenge
> Publisher: The CTO Advisor LLC (DBA The Advisor Bench)
> Machine-readable record updated August 30, 2026

## The question

Which decisions are we trying to accelerate, and who owns them today?

The Design Challenge carries one operating question through four acts: a working conversation, a challenged architecture, a measured lab, and a published record the field can cite. It starts with decision friction rather than a product claim and treats authority, accountability, evidence, and limits as part of the architecture.

## How to read this reference

Each artifact below is published work, chosen to show what that act looks like in practice. The lab runs whether or not a vendor pays for it: 18 of 19 published labs are unsponsored. Commissioned Design Challenges publish at their own destinations; the first is underway with Google Cloud. The selection and the argument are editorial, not any vendor's message. Sponsorship and evidentiary limits are stated with each artifact.

## Act I · Conversation

Purpose: Establish the operating question with builders and operators before reducing it to an architecture diagram.

Engagement: HPE sponsored this portion of the CTO Advisor Road Trip. Town of Vail field engagement, published in 2026.

1. Inside a Real AI Smart City Project
   - Subject: How HPE infrastructure, NVIDIA accelerated computing, and Kamiwaza orchestration support Town of Vail municipal workloads.
   - Video and transcript: https://thectoadvisor.com/ctodose/inside-a-real-ai-smart-city-project-town-of-vail-hpe-x8drokuf8uk/
   - YouTube: https://www.youtube.com/watch?v=X8Drokuf8uk

2. How AI Turned 80 Years of “Dirty Data” into a Smart City Platform
   - Subject: How difficult municipal records became context for practical AI services, and why data work matters as much as models.
   - Video and transcript: https://thectoadvisor.com/ctodose/how-ai-turned-80-years-of-dirty-data-into-a-smart-city-platf-o6pvqthpfdk/
   - YouTube: https://www.youtube.com/watch?v=O6PVqThpFDk

## Act II · Challenge

Purpose: Put an architecture answer on the record and challenge its assumptions while the field watches.

Format:
1. Keith brings the problem.
2. The vendor works the solution.
3. Keith challenges the design.
4. Layer2C Labs validates.

Reference artifact: Why RAG Breaks at Enterprise Scale, and What Comes After
- Engagement: Articul8 sponsored content, recorded before GTC. This was not a Design Challenge; it demonstrates the same challenge beat inside an editorial interview.
- Video and transcript: https://thectoadvisor.com/ctodose/why-rag-breaks-at-enterprise-scale-and-what-comes-after-arti-6pdslqgsoi/
- YouTube: https://www.youtube.com/watch?v=-6PdsLqgsOI

## Act III · Lab

Purpose: Put the design on real infrastructure so controls, seams, and failure paths become observable evidence.

Funding: 18 of 19 published Layer2C Labs are unsponsored. The lab leading this act is one of them.

Reference artifact: Layer2C Lab 003 · The validator determines done, not the loop
- Engagement: Nobody commissioned this lab. Editorial, unsponsored.
- Lab ruling: https://labs.layer2c.com/labs/loop-control
- Finding: A local bug-fix agent was supposed to need a frontier tier to escalate to. Gated by a deterministic harness with every call metered, the same-tier repair loop returned nothing.

Measured results:
- Pass rate 25% before the same-tier repair loop and 25% after it. No gain.
- Roughly 5% of tasks are where same-tier repair actually pays, once measured.
- Roughly $4.60 in total metered cost across 156 API calls, published with the result.

Limits:
- The task set is small: eight tasks in the culminating run, 26 calibrated.
- Good for the architecture verdict, not a benchmark of pass rates.

Second reference artifact: Layer2C Lab 015 · Inherit the boundary. Own the gate.
- Engagement: Kamiwaza sponsored this lab. Kamiwaza did not choose the assessed vendors or see the production assessment instrument, and the lab record states the findings include the ones they would not have chosen.
- Lab ruling: https://labs.layer2c.com/labs/evidence-authority
- Video: https://youtu.be/E-_zPBFsSdo

Measured results:
- 241 retrieval hits under adversarial cross-vendor querying; none crossed a workroom.
- 42 of 42 claims grounded in the asking workroom's own corpus, checked by a deterministic resolver rather than by reading.
- 0 citations reached into the other analyst's vendor material.

Limits:
- This is not a security result.
- Every account was authorized and acting in good faith.
- Scope: one deployment, one release, two workrooms, and one assessment workload.

The two artifacts are the point of this act: the same instrument, the same discipline about stated limits, run once with a funder and once without.

## Act IV · Publish and activate

Purpose: Turn the findings into a record a person or assistant can retrieve and into grounded prompts the field can use in customer engagements.

The standing corpus:
- 27 independent vendor assessments scored against the 4+1 Layer AI Infrastructure Model. None commissioned.
- 19 lab records, each carrying its finding, numbers, limits, and funder.
- Every page also published as .md and .json for retrieval, indexed by the router at https://labs.layer2c.com/llms.txt

A sponsored publication, as an example of what one looks like:

Reference artifact: Designing Decision Systems for Smart Cities
- HPE-sponsored paper, January 2026.
- Paper: https://layer2c.com/designing-decision-systems-for-smart-cities.pdf
- Subject: A repeatable CTO framework that starts with decision friction, designs the human and AI loop, builds shared capability, targets mundane cross-department decisions, and orchestrates an ecosystem rather than a product stack.


Field enablement prompts:

1. Architecture discovery
   - Prompt: Map this customer's current AI architecture across the 4+1 layers. Identify which decisions are Retained, Delegated, or Ceded, and give me the five questions needed to validate the map.
   - Use: First meeting; architecture workshop.

2. Competitive positioning
   - Prompt: Compare our platform with the customer's incumbent for this workload. Use the Layer2C assessments and labs to distinguish capability from authority, cite the evidence, and name what remains unproven.
   - Use: Account planning; technical validation.

3. Customer follow-through
   - Prompt: Turn the findings from this conversation into a Design Challenge brief: one testable architecture question, the authority boundary, success metrics, evidence required, and the next workshop agenda.
   - Use: Solution design; next-step activation.

Each grounded response should carry the relevant Layer2C assessment, lab ruling, transcript, evidence limits, and disclosure. The AEO/GEO claim is corpus presence, never ranking or placement.

## Vendor outcome

- A sharper story tied to an enterprise architecture decision and answered in the open.
- A credible proof set spanning video, measured lab work, and editorial analysis.
- An account executive who cites evidence instead of claiming, and can hold the architecture conversation without a sales engineer in the room.

## Related machine-readable sources

- Layer2C assessment index: https://layer2c.com/llms.txt
- Complete Layer2C assessment corpus: https://layer2c.com/llms-full.txt
- Layer2C Labs corpus router: https://labs.layer2c.com/llms.txt
- Layer2C Labs structured index: https://labs.layer2c.com/labs/index.json

## Contact

Keith Townsend · keith@thectoadvisor.com