How I Passed the OpenTelemetry Certified Associate — 13 of 16 certs for Golden Kubestronaut

I just passed the OpenTelemetry Certified Associate (OTCA) on 27 August 2026 — my ninth multiple-choice CNCF cert on the way to Golden Kubestronaut.

No real exam questions here: you have to sign a confidentiality agreement with the Linux Foundation, and I'm keeping to it.

What the OTCA actually weights

Nearly half the exam is one domain, and it isn't the half people prepare for. "API and SDK" doesn't mean writing instrumentation code — it means knowing what the SDK's parts do once configured: instrument selection, span processors and their queue behaviour, samplers and how a decision propagates downstream. The Collector's 26% is almost entirely YAML — read a pipeline, say what's wrong with it. Fundamentals at 18% is narrower than "what is observability": resource identity, semantic conventions, schema_url. The last 10% is small and unforgiving — you've either watched a Collector drop data or you haven't.

How I studied

• I used KodeKloud's Prep Course — OpenTelemetry Certified Associate (OTCA) as the spine.
• Writing the questions, and doing the mock exams again and again.

Writing the questions is the study

The cheap version needs no tool and is the part worth stealing: for every topic you study, write three exam questions about it, including the wrong answers. If you can't produce a convincing wrong answer, you've found your gap.

I ran that habit at an unreasonable scale. Alongside the last several certs I've been building my own mock exam simulator — https://www.goldenkubestronaut.app

The OTCA bank meant three 60-question mocks weighted to the official split, every question with a defensible answer, three plausible distractors, and an explanation tied back to official documentation — all written from the published curriculum, and finished before I sat the exam. You cannot write a good distractor for a topic you only half understand. The distractor is the test, not the answer.

One of my own 180:

An incoming HTTP request carries traceparent with the sampled flag clear. The SDK is configured with OTEL_TRACES_SAMPLER=parentbased_traceidratio and OTEL_TRACES_SAMPLER_ARG=0.5. What sampling decision is made for the resulting server span?

1. It is sampled with 50% probability based on its trace ID, ignoring the parent flag
2. It is always sampled because parentbased_traceidratio defaults to keeping all traces with a parent
3. It is dropped; parentbased_* honors the parent's sampled flag when one exists
4. It is sampled only if the parent's tracestate contains an otel vendor entry

Answer

3. parentbased_traceidratio follows the upstream traceparent sampled flag whenever there is one; the 0.5 ratio is the fallback used only at the root. A sampled-clear parent therefore drops the child, which is what keeps a trace consistent end to end. (1) inverts that precedence, (2) invents a default that doesn't exist, and (4) confuses tracestate with the sampled flag.

Then I sat the mocks cold — same format, same 75% threshold, timed. The gaps that surfaced were real gaps, and I had a week to close them. The exam itself felt familiar: not because I'd seen the questions, but because I'd already had to think in the same shapes.

Try the application yourself

If you're studying for KCNA or CGOA, both have a free mock exam once you log in — the easiest way to judge whether the questions are any good before spending anything.

goldenkubestronaut.app

Here's what an OTCA question looks like in practice mode.

Practical notes

• 90 minutes, 60 questions, 75% to pass, remote-proctored — that's what I sat on 27 August 2026. The Linux Foundation OTCA page is the authority if it moves.
• Test your webcam and clear your desk well before your slot. Flag and move rather than staring — multiple choice rewards a second pass.

What's left

Three to go for Golden Kubestronaut: CNPA, then LFCS and CNPE hands-on.