Define leakage-resistant delay-model contract - #3
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4c642918d5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - route identifier | ||
| - route distance band | ||
|
|
||
| Historical aggregates are permitted only when calculated from rows strictly earlier than the prediction row: |
There was a problem hiding this comment.
For a T-24h prediction, requiring only that an aggregate use rows earlier than the current row still permits future outcomes. For example, when scoring an 18:00 flight at 18:00 the previous day, a same-day 10:00 flight sorts earlier but its ArrDel15 is not yet known. Implementing the contract literally would therefore leak labels into every listed prior-delay feature and invalidate validation/test metrics; require each contributing outcome to have become available before the current flight's T-24h timestamp, including the prior used for smoothing.
Useful? React with 👍 / 👎.
1 participant
What changed
Why
The live v0.1 API currently exposes a transparent rules baseline. Before adding a trained model, the project needs an auditable contract that prevents target leakage, retrospective metric inflation, and unsafe artifact handling.
Impact
Issue #2 now has a concrete foundation for preprocessing and training. Future model results can be evaluated against a fixed question and acceptance standard instead of changing the task to fit the output.
Validation
python -m pytest— 6 passedgit diff --check