Prove It Day 03: Prove It Arena

Archived event. The Arena ran once, on Sep 19, 2026, and can’t be run now: it needed the environments, team codes and AI judge provided on the day. This page records how the day worked. You can still reuse the claim labels and report skeletons, and check your work against the sample graded report.

Day 03 closed Prove It with two scored rounds: a read-only review worth 30 points and an incident investigation worth 70. Each team wrote one Markdown report per round, and an AI judge scored it.

The scenario

You get two unfamiliar cloud environments, a clock and a judge that only sees what you write. The Day 01 and Day 02 rules still hold:

Read-only. The agent can inspect resources, configuration, telemetry and logs. It can’t change anything or read secrets.
No credentials. You join with a team code, and the environments are operated for you.
Nothing executes. Every change in your report is a recommendation with a named owner.
One to four people per team. One laptop is enough.
Redact before you share. Remove account IDs, ARNs, endpoints, hostnames and customer data.

#	Block	Minutes	What happens
1	Leadership intro	10	Opens the day.
2	Briefing	10	Rules, clock and board.
3	Community share	30	Three teams present a Day 01 or Day 02 report, one per city. Join the Arena during this block.
4	Round 1: Find the Gaps	25	Read-only review across five Well-Architected pillars. 30 points, board public.
5	Break	10	Round 1 closes.
6	Round 2: Find the Cause	45	An incident, traced to its cause. 70 points, board hidden.
7	Reveal and close	20	Scores, the answers, and each region’s Prove It Champion and Runner-up.

Scoring rules

Rule	What it means for your team
One report per round	Markdown. Click Load .md or type in the editor.
Versions are free	Click Save version as often as you like. Each version keeps the time you saved it.
Finals are graded	Mark final & grade sends one version to the judge and uses one attempt.
Three attempts per round	The header shows what’s left, starting at Attempts 3 of 3 left.
Three-minute cooldown	Between graded attempts. The header shows Cooldown ready when you can submit again.
Best graded final counts	Your highest-scoring final in a round is your score for that round.
No final, no score	A round with no marked final scores zero.
Tie break	The earliest final mark wins.
No hints	Nothing is deducted. The only cost is a graded attempt.

The judge is an AI agent. It grades each final three times and takes the median, then returns a total and a short note so your next attempt can target what the last one missed. It never reveals the answer and doesn’t look at the environment for you. In Round 2 you don’t see your own score: the board only shows that you submitted.

Mark a first final early: by T+10 in Round 1 and by T+25 in Round 2. The judge’s note is worth more than ten extra minutes of writing, and a team that submits at the last minute gets no feedback to work from.

How to label every claim

Each line in a report is worth what backs it. Label it the way you did on Day 01:

Label	Use it when	Example
Verified	Raw output from a named call proves it, in a stated scope, at a stated time.	”Bucket allows public read. Source: get-bucket-policy-status, output attached. Read-only role, checked 10:14.”
Inferred	You reasoned it from a verified fact.	”Likely served to the public, because the bucket policy above allows it.”
Assumed	Nothing backs it yet.	”Probably unused; no access data checked.”
Blocked	The role couldn’t look.	”IAM credential report: access denied for this role.”

Findings aren’t capped. One proved finding beats five guessed ones, and a report that names its own gap scores better than one that hides it.

Walkthrough

Join your team during the community share

At the event, teams opened arena.cloudthinker.io (the Arena can’t be run now). Under Join your team, each team typed the join code from its team card or the slide, then clicked Enter the arena.

Prove It Arena home page showing the two round cards, the rules of the day and the Join your team code field

Success state: the Team console opens with your team name and a Live badge in the header. To change the name, click Rename in the header. A name another team already uses is refused.

Check the board and your round page

The board on the big screen lists every team with an R1 · 30% column, an R2 · 70% column and a Total. Each round column shows that round’s share of the 100-point total. The R2 column shows a lock until the reveal.

Live leaderboard showing every team with a column per round and a running total, both rounds not started

Your round page has the report.md editor, the Versions panel and the attempt counter.

Team round page with the Markdown editor, the Load .md button, the versions panel and the attempts left

Success state: the round shows Not started, the header reads Attempts 3 of 3 left, Cooldown ready and Judge idle, and a banner says you can draft but saving is disabled until the round opens. Nothing sent early counts.

Round 1: read the environment, then agree the list (T+00 to T+05)

Ask the agent what’s unusual before you write anything. Use five of the six Well-Architected pillars (Cost Optimization, Security, Performance Efficiency, Reliability, Operational Excellence) as a frame, not a quota.

Compared with a default environment, what is unusual here? Group it by cost optimization, security, performance efficiency, reliability and operational excellence.
Which of these is a real risk with evidence behind it, and which only looks alarming? Show the raw output for each.

Success state: a short list where every item points at a call and its raw output. If the agent hands you a long list, ask which single item it would fix first and what proves it. If it can’t read something, that’s a Blocked line for block 01.

Round 1: submit a first final by T+10

Paste this skeleton into the editor, fill it, then click Save version and Mark final & grade.

# Round 1: Find the Gaps

## 01 What we looked at
- Environment and scope reached:
- Blocked (role could not look):

## 02 What we would fix (biggest impact first)
1. Finding:
   - Pillar:
   - Evidence (source, scope, time):
   - Label: Verified / Inferred / Assumed

## 03 What needs a person
- Change that needs approval, and who approves it:
- Dropped on purpose, and why:

Success state: the version appears in Versions with its time, the counter drops to Attempts 2 of 3 left, the cooldown starts, and the judge returns a total with a short note. Your score shows in the R1 column on the public board.

Round 1: aim the second version at the note (T+18), stop at T+23

Fix the gap the note points to. Strengthen evidence before you add findings.

For the finding I am about to submit, give me the source, scope and time that prove it, and tell me what would make it wrong.

Success state: a second graded final before the clock runs out. The round closes on the clock, not on a save.

Round 2: frame the incident and two hypotheses (T+00 to T+12)

Read the symptom in the customer’s words before any dashboard. Write the incident question: what’s broken, for whom, since when. Healthy dashboards are part of the exercise.

Give me at least two explanations for this symptom that fit the evidence so far, and the reading that would rule each one out.
Which resource can you show actually changed, with a timestamp? Which ones are you only assuming changed?

Success state: two written hypotheses, each with the reading that would kill it. If one confident cause arrives at once, ask for the rival explanation first.

Round 2: submit a first final by T+25, a second by T+40

Use five blocks. Then aim the next version at the block you can defend least.

# Round 2: Find the Cause

## 01 Question and scope
What is broken, for whom, since when. What our access could not see.

## 02 Hypotheses, one ruled out
- H1:
- H2:
- Ruled out by (reading, time):

## 03 Evidence
Source, scope and time for each item, with raw output.
- Label: Verified / Inferred / Assumed / Blocked

## 04 Mechanism
Symptom -> ... -> changed resource. The edge we could not prove:

## 05 Root cause and response
- Resource whose configuration changed, and when:
- Conditions we cannot date (contributing factors):
- Recommended change, who approves it, and the check that proves it worked:

Success state: the attempt counter drops and the board shows your team submitted, but the R2 score stays locked. If you run out of time, submit what you have. A report that ends with “insufficient evidence” and shows the log is complete.

What made this work

A dated change. In Round 2 the root cause is the resource whose own configuration changed. Conditions you can’t date are contributing factors. See root cause analysis for how Resolve, CloudThinker’s Frontier Investigation Agent for production incidents, builds a causal chain.
A response with an owner. Say what to change, who approves it, and what you’d check afterwards. A response that only names the symptom reads as a guess. In your own workspace, Approvals is where that owner signs off.
Honest gaps. Blocked scope and unproven edges, written down, count in your favor.
An early first final. The judge’s note on attempt one is the only feedback you get. Teams that wait lose it.

At the reveal, both rounds’ scores open together, then the organisers walk through what was there to find. Teams that complete all three days keep three portfolio reports and the series skill certificates. Outstanding finalists may be selected for the three-month CloudThinker Ambassador Program.

Next steps

Prove It overview

See all three days and what each one builds.

Day 02: Investigate

Review triggers, root causes and contributing factors before Round 2.

Root cause analysis

See how Resolve traces a symptom to the change that caused it.

Approvals

Learn how CloudThinker gates changes behind a person.

Learn

Workshops

Scenarios

Prove It Day 03: Prove It Arena

The scenario

Scoring rules

How to label every claim

Walkthrough

What made this work

Further reading

Next steps

Prove It overview

Day 02: Investigate

Root cause analysis

Approvals

​The scenario

​Scoring rules

​How to label every claim

​Walkthrough

​What made this work

​Further reading

​Next steps

Prove It overview

Day 02: Investigate

Root cause analysis

Approvals

The scenario

Scoring rules

How to label every claim

Walkthrough

What made this work

Further reading

Next steps