Skip to content

Submission — pradeepbugga - #89

Open
pradeepbugga wants to merge 3 commits into
8090-inc:mainfrom
pradeepbugga:submission/pradeepbugga
Open

pradeepbugga wants to merge 3 commits into
8090-inc:mainfrom
pradeepbugga:submission/pradeepbugga

Conversation

@pradeepbugga

Copy link
Copy Markdown

Submission

  • Public solution repository: https://github.com/pradeepbugga/mib-doc-challenge (main branch, includes Dockerfile)
  • Training-set score: 120.58/150 (extraction 41.81/50, classification 65.17/80, calibration 13.60/20, 0 missing cases, 26 catastrophic false approvals)
  • Validation predictions: 5,000/5,000 valid records, 0 missing case ids, verified with scripts/validate_submission.py
  • Docker verification: image builds clean (1.05 GiB, under 4 GiB limit), runs correctly under the exact scoring contract (--network none --cpus 4 --memory 8g --pids-limit 512 --read-only --tmpfs), 5.51s/PDF average (under the 6s/PDF budget), output byte-identical to host predictions on a 25-PDF sample.

See submissions/pradeepbugga/MEMO.md for the technical memo and submissions/pradeepbugga/SUBMISSION.md for full build/run/verification details.

pradeepbugga and others added 3 commits August 2, 2026 21:37
Validation predictions for all 5,000 packets, technical memo, and a link to
the public solution repository.

The solution is a deterministic offline pipeline (PyMuPDF, OpenCV, Tesseract)
with no LLM or trained model. It scores 113.04/150 on the training set.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…ainer-produced, 5.51s/PDF)

Replaces the 113.04-build submission with fresh predictions from tonight's
pipeline (120.58/150 on the training set at time of this run). Predictions
regenerated on the host (--workers 8), validated: 5000/5000 valid records,
0 missing case ids against data/validation_manifest.csv. Docker image
rebuilt and verified against the exact scoring contract (--network none
--cpus 4 --memory 8g --pids-limit 512 --read-only --tmpfs): 25-PDF sample
run produced byte-identical output to the host, 5.51s/PDF average (under
the 6s/PDF budget, tighter than the previous build's 1.43s/PDF due to
additional scored-candidate OCR retries added since), 1.05 GiB image
(under the 4 GiB limit).

MEMO.md is an in-progress draft: current train-set numbers are real and
measured, several sections are marked [FILL IN] pending the author's own
pass for judgment calls and earlier-session context.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant