Submission: LukaJurisic — attributed V6 baseline

[LukaJurisic]

MIB Doc Challenge Submission

Links

• Solution repository: https://github.com/LukaJurisic/mib-doc-challenge-solution
• Generation receipt: https://github.com/LukaJurisic/mib-doc-challenge-solution/blob/9e140db7f2ca88d3af569d129df869164e2efdfa/docs/V6_S4_BASELINE_GENERATION_RECEIPT.json
• Submission form: completed on 2026-08-03

Artifact

• 5,000 independently generated validation predictions
• SHA-256: 08ba6fb615129dccd78ce29025d8d3945b7bea1b1e5bec4a0093fb8e7ec8e71f
• Official validator: 5,000 valid records; zero missing expected IDs; --require-complete passed
• No participant prediction file was downloaded or copied. Matching public bytes arise from independently reproducing the attributed deterministic implementation.

The solution preserves MIT attribution to Calling Moonshots, its handemanai baseline, and Tyler's public reproduction evidence. Luka's baseline contribution is prediction-neutral build pinning, deterministic distributed generation, receipt-bound timeout recovery, immutable merge/finalization, validation, and release packaging.

Runtime disclosure: distributed generation proves artifact completeness, not official single-container runtime eligibility. Luka's local monolithic 5,000-case run did not complete within 30,000 seconds; organizer-host performance remains unresolved and is disclosed in the repository.

Checklist

• I filled out the submission form linked by the organizers
• This PR only adds submissions/LukaJurisic/predictions.jsonl, MEMO.md, and SUBMISSION.md
• predictions.jsonl passes scripts/validate_submission.py --require-complete 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 documented limits
• No hardcoded validation answers and no manual per-case edits
• The memo describes the approach, failure modes, and what I would improve with another week

[LukaJurisic]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 46a6a20ab5

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".

[LukaJurisic]

Submission form completed on 2026-08-03. The independently generated baseline remains the deadline-insurance artifact while the single bounded M14 candidate is evaluated.