submitting-results

name	submitting-results
description	Submit competition results to the leaderboard for this project using make submit and Kaggle CLI. Use when the user asks to submit a model, run make submit, or send the current submission.csv to the leaderboard.

Submitting Results (AI Cup 2026 Performance)

When to use this skill

Use this skill whenever the user wants to:

• Submit a new model to the AI Cup 2026 | Performance Track leaderboard.
• Run make submit or otherwise send the current submission.csv to Kaggle.
• Ensure submissions are associated with a clear, informative message.

This skill is project-specific to the bird-prediction repo.

---

Prerequisites

• The project root is /Users/nomomon/Desktop/Projects/bird-prediction.
• submission.csv in the project root is the file to submit (already generated and tested).
• The working tree must be clean (no uncommitted changes), because the Makefile enforces this.
• The Kaggle CLI is already configured and authenticated for the user.

You must not try to bypass the Makefile’s cleanliness check. If the tree is dirty, guide the user to commit or stash changes first.

---

Submission workflow

1. Verify git status

From the project root, run:

cd /Users/nomomon/Desktop/Projects/bird-prediction
git status

• If git status reports “nothing to commit, working tree clean”, you can proceed.
• If there are uncommitted changes, ask the user whether to:
  • Commit them (preferred for changes that should be kept), or
  • Stash/discard them (only if explicitly requested).

Do not run make submit until the working tree is clean.

2. Ensure the correct submission file is in place

By convention in this project:

• The best current model’s predictions are copied or written to submission.csv in the project root.
• If multiple candidate submissions exist (e.g. submission_tuned_calibrated_targetenc.csv), first copy the chosen one:

cd /Users/nomomon/Desktop/Projects/bird-prediction
cp submission_tuned_calibrated_targetenc.csv submission.csv

Only after this step should you submit.

3. Understand how make submit works

The Makefile contains:

submit:
	@if [ -n "$$(git status --porcelain)" ]; then \
		echo "commit all changes before submission"; \
		exit 1; \
	fi; \
	hash=$$(git rev-parse --short HEAD); \
	read -p "Submission message: " msg; \
	kaggle competitions submit -c ai-cup-2026-performance -f submission.csv -m "[$$hash] $$msg"; \
	echo "Waiting 20s for submission to process..."; \
	sleep 20; \
	kaggle competitions submissions -c ai-cup-2026-performance

Key points:

• It refuses to run if there are uncommitted changes.
• It computes the current git commit short hash and prefixes it in the Kaggle message like [abcd123].
• It reads a single-line submission message from stdin (via read -p "Submission message: " msg).
• It submits submission.csv to the ai-cup-2026-performance competition.
• It waits 20 seconds, then prints the recent submissions table with public scores.

Because read expects input from stdin, you must pipe the message when driving this non-interactively.

4. Running make submit with a piped message

To submit non-interactively (as the agent), use this pattern:

cd /Users/nomomon/Desktop/Projects/bird-prediction
printf 'Your submission message here\n' | make submit

Guidelines for the message:

• Include the phase or experiment description and OOF score (and any notable settings).
• Keep it to a single concise line; the Makefile automatically prepends the commit hash.

Examples:

printf 'Phase 1: tuned+calibrated histgb with target-encoding (OOF 0.5555)\n' | make submit

printf 'Phase 2: stronger regularization search, tuned histgb + target-encoding (OOF 0.56x)\n' | make submit

printf 'Phase 3: per-class sample weights for weak classes (Cormorants/Ducks/Waders)\n' | make submit

After the command finishes, inspect the printed submissions table to see:

• The fileName (should be submission.csv),
• The description (including commit hash + your message),
• The publicScore for the new submission.

---

Agent usage checklist

When the user asks to “submit”, “use make submit”, or “send this to the leaderboard”:

1. Confirm best file is in submission.csv
   • If needed, copy the best candidate file over submission.csv.
2. Ensure clean git state
   • Run git status; if dirty, coordinate with the user to commit/stash before proceeding.
3. Construct a clear message
   • Include: phase name, key settings (tuned/ensemble/target-encoding/etc.), OOF mAP, and any notable class behavior.
4. Run printf 'message\n' | make submit
   • Use the exact project root path and command pattern above.
5. Report back to the user
   • Summarize: which file was submitted, the message used, and the resulting public leaderboard score from the printed table.

Follow this workflow every time you submit to keep submissions traceable and consistent.