Submission: zubalr

[zubalr]

MIB Doc Challenge Submission

Links

• Solution repository (public, contains a Dockerfile): https://github.com/zubalr/mib-intake
• Submission form completed: 2026-07-29

Checklist

• I filled out the submission form linked above
• This PR only adds submissions/zubalr/predictions.jsonl, MEMO.md, and SUBMISSION.md
• predictions.jsonl passes scripts/validate_submission.py against data/validation_manifest.csv
• My solution repository is public and includes a Dockerfile
• My Docker image runs offline (--network none) and accepts <input_pdf_dir> <output_predictions_path>
• My submitted runtime uses no LLMs, VLMs, cloud OCR, or network services
• Model artifacts fit the size limits in DOCKER_SUBMISSION.md
• No hardcoded validation answers and no manual per-case edits
• My memo describes my approach, failure modes, and what I would improve with another week

Result

Section	Training set	Out of fold
Extraction	44.16 / 50	44.15 / 50
Classification	72.69 / 80	67.88 / 80
Calibration	17.07 / 20	16.50 / 20
Total	133.93 / 150	128.53 / 150, SE 0.16

Training-set score is the shipped image scored by scripts/evaluate.py on all
1,000 labeled packets, and is in-sample. The out-of-fold figure averages ten
fold assignments and is the estimate for unseen packets.

Validation run: 5,000 predictions, 0 missing, 2h55m on 4 vCPU, 2.11s per PDF.

[zubalr]

Superseded by the updated submission: #78

The new PR contains the final predictions.jsonl, MEMO.md, and SUBMISSION.md generated from the latest mib-intake release, with a reproducible public-training score of 137.89/150. Please use the new PR for evaluation and disregard this earlier submission.