Cleanor LabsCompetitions and hackathons
We enter data and AI competitions where our datasets and tooling give us a real shot, and we publish what we learn. This page tracks what we entered, what it produced, and the field notes we wish someone had handed us before our first leaderboard.
What we entered
Every entry is first hand, under the Cleanor Labs account. Active competitions carry no scores here: a leaderboard moves daily and a page should not pretend otherwise.
Kaggle · red-teaming competitionAI Agent Security Challenge
Competing now
A $50,000 red-teaming challenge: craft inputs that make defended AI agents leak what they should not, scored against live guardrails. We are competing under the Cleanor Labs account. The full write-up lands here after the competition closes — publishing tactics while it runs would break both the sharing rules and our own standing.
Our Kaggle profile →
Kaggle · open data programKaggle Datasets — Bronze medal
🥉 Bronze medal, July 2026
The Kaggle mirror of the Cleanor Search Index (monthly Google search demand for AI tools across 20 countries) earned a Bronze medal from the community. The dataset is the same open pipeline that powers our Search Index studies.
The medal dataset → · The studies built on it →
Field notes for participants
Distilled from our own submission logs. Process lessons only — tactics for competitions still running stay off this page until they close.
Read the prize split and the sharing rules before writing code
Half the strategy is in the rules PDF nobody reads: how the prize pool splits, what counts as a team, what you may publish and when. We plan the work only after we know what a win is actually worth and what we are allowed to say about it.
Bank a score early — the leaderboard keeps your best
Your best submission is permanent, so the first decent result should be on the board on day one, not after a week of polishing. From then on every submission is an experiment against a known baseline instead of a leap of faith.
Treat daily submissions as a budget, not a retry button
Most competitions cap submissions per day. Burning the whole allowance tuning one idea is the classic mistake: you learn one thing five times. Spread the budget across different hypotheses and keep a log of what each submission was testing.
Do not overfit the public leaderboard
The public score is computed on a slice of the data; the final ranking comes from a private slice you never see. A configuration tuned to the public split can drop hard on the private one. Trust your own validation more than the daily rank.
Keep every attempt reproducible
One self-contained script or notebook per submission, versioned, with the score written next to it. When something finally works you need to know exactly what it was — a directory of near-identical untitled notebooks is how winning configurations get lost.
Publish datasets, not just entries
Our Bronze medal came from open data, not from a contest entry. Publishing a clean, documented dataset compounds: it earns community recognition, citations, and search visibility long after any single competition ends.
Frequently asked questions
How does Cleanor Labs pick which competitions to enter?
Three filters: the task must overlap with what we already build (data pipelines, storage, AI tooling), the entry cost must be near zero beyond our time, and the result must leave a public artifact — a ranking, a medal, a dataset, or a write-up — that outlives the event.
Are Kaggle dataset medals worth pursuing?
Yes, if the dataset is genuinely useful. A medal is community recognition that compounds into citations and discoverability. Chasing medals with low-effort scraped data does not work: the votes come from people who actually used the data.
When will the AI Agent Security write-up be published?
After the competition closes. Kaggle competitions restrict sharing working solutions while the leaderboard is live, and a red-teaming challenge deserves extra care: the write-up will cover approach and process, not a recipe of working attacks.
Where does the competition data on this page come from?
First hand. Every entry listed here is one we entered ourselves under the Cleanor Labs account, and the field notes are distilled from our own submission logs, not aggregated from other write-ups.