Milestone map
Milestone map
3 milestones
Complete a Getting Started competition end to end
1–2 weeks
Before competing for placement, work through a Kaggle Getting Started competition (Titanic, House Prices, or Digit Recogniser) entirely from scratch: exploratory data analysis, feature engineering, a baseline model, and a final submission. The goal is not placement — it is to build the mechanical fluency you need to iterate quickly in real competitions, where time spent on setup is time not spent on features.
Proof required
Share the public URL of your Kaggle profile or notebook showing a completed Getting Started competition submission — the submission history must show at least one scored submission, and the notebook or write-up must include your EDA, feature engineering decisions, and final model choice with a brief rationale for why you chose that approach over alternatives.
What gets checked
- Notebook includes EDA with at least 3 observations about the data that informed feature decisions — 'I looked at the data and built a model' is not EDA
- Feature engineering section explains at least one feature created or transformed and the reasoning — a model trained on raw columns without any feature decisions is a missed learning
- Model choice is justified with a comparison to at least one alternative — 'I used Random Forest because it works' is less credible than 'I tried Logistic Regression first; Random Forest improved CV score from 0.78 to 0.83'
Common mistakes
- Copying a public notebook without understanding it — the proof will break at M2 when the competition format changes and the same notebook no longer applies
- Skipping EDA and going straight to modelling — most competition performance gains come from data understanding and feature engineering, not model choice
- Not establishing a local cross-validation baseline before submitting — submissions without CV scores make it impossible to know whether a change improved things
Resources
Foundationstart here
Depthgo deeper
What a verifier looks for
- Ask for the public Kaggle profile or notebook URL — the submission history on the Kaggle platform is independently visible and shows actual submission dates and scores
- Ask the submitter to explain one feature they engineered and whether it improved or hurt their CV score — this separates genuine experimentation from copying a public notebook
- Check whether the notebook includes a validation strategy — submitting to the public leaderboard without local CV is a red flag for later competition attempts
You'll sign in first, then come straight back here.
Build a competitive pipeline for a Featured competition
4–8 weeks (while competition is active)
Move from Getting Started to a Featured competition — one with prize money or a meaningful ranking component. Study the top public notebooks and discussion threads for the competition to understand what the strong competitors are doing, then build your own pipeline that incorporates at least two techniques you learned from that research. The goal is a pipeline that can be iterated quickly and that you understand well enough to explain every decision.
Proof required
Share a Kaggle notebook or GitHub repository showing your pipeline for a Featured competition, including: (1) your cross-validation strategy and a table of CV scores across at least 3 model variants or feature configurations you tested, (2) a note on two techniques you learned from top public notebooks and whether they improved your CV score, and (3) your current leaderboard position at the time of submission.
What gets checked
- CV table shows genuine experimentation — at least 3 different configurations tested, not just a single model submitted and resubmitted with minor tweaks
- Two named techniques from top notebooks are described: what they are, why they were expected to help, and what actually happened to CV score when applied
- Leaderboard position is visible in the Kaggle submission history — not self-reported
Common mistakes
- Joining a competition in its final week — not enough time to iterate meaningfully; join a competition at least 4 weeks before close
- Using a public notebook verbatim without adapting it — the proof requires showing your own experimentation, not mirroring the best public solution
- Optimising leaderboard score at the expense of understanding — a shaken-up ensemble that scores well but you cannot explain will not transfer to the next competition
Resources
Foundationstart here
Depthgo deeper
What a verifier looks for
- Check the Kaggle leaderboard for the competition and find the submitter's username — their position should be independently visible in the public leaderboard
- Ask the submitter to explain their cross-validation strategy in detail — a candidate who genuinely built their own pipeline can explain why they chose their fold structure and whether the CV score correlated with leaderboard movement
- Ask about one technique they tried that did NOT improve CV score — genuine iteration always produces some failures; only reproducing public notebooks produces uniformly 'successful' experiments
You'll sign in first, then come straight back here.
Achieve top-25% placement in a Featured or Research competition
Ongoing during an active competition; the milestone is completed at the competition's close
Top-25% in a Kaggle Featured or Research competition places you above the majority of submitters who entered with serious intent. It is a verifiable, externally set threshold that requires genuine data science competence — not just running a public notebook. At this stage your pipeline should reflect iterated experimentation: feature engineering decisions driven by CV evidence, at minimum a tuned gradient boosting model or neural network, and a validation strategy that predicts leaderboard movement reliably.
Proof required
Share the public Kaggle competition URL and your username. The leaderboard must show your team or username placing in the top 25% of all teams when the competition closes or at the final submission cutoff. Also share the notebook or repo used for your final submission, noting one change from M2 that most improved your CV score.
What gets checked
- Top-25% placement is confirmed by the public final leaderboard — not the public leaderboard during the competition, which can shift on private test data
- Final submission notebook or repo is linked and the submitter can explain the most impactful single change from M2's pipeline
- Competition is a Featured or Research competition with a real prize or ranking component — Getting Started competitions do not count for this milestone
Common mistakes
- Leaderboard shakeup: finishing top-25% on the public leaderboard but dropping significantly on the private — this indicates overfitting to the public set; genuine top-25% should survive the private evaluation
- Joining only very small competitions where top-25% is reached with a baseline model — the proof should be from a competition with at least 200 teams to have meaningful signal
- Submitting only once and achieving the result by luck — while valid as a competition result, the milestone is most credible when the progression from M2 is visible in the submission history
Resources
Masteryfor the dedicated
What a verifier looks for
- Verify the placement directly on the Kaggle competition leaderboard using the submitter's username — the final leaderboard is public and permanent after competition close
- Confirm the competition is Featured or Research with at least 200 participating teams — Sprint competitions and Getting Started competitions do not meet the bar
- Ask the submitter what the leaderboard shakeup was: the difference between public leaderboard rank and final private leaderboard rank — a large shakeup (e.g. top-10% public but top-40% private) suggests the M3 threshold was barely met and the result warrants discussion
You'll sign in first, then come straight back here.