Skip to content

DueCare Kaggle Kernel Guide

kaggle/_INDEX.md and docs/current_kaggle_notebook_state.md are the authoritative active-kernel inventory. This guide is the human-readable purpose map generated from the same root kaggle/*/kernel-metadata.json script-kernel folders, plus scripts/kaggle_live_slug_map.json for public live status.

Notebook artifact policy

Do not create .ipynb notebooks for the judge-facing submission by default. Treat kernel.py plus the folder README as the source of truth. Historical notebook wrappers live under _archive/kaggle-notebook-previews-2026-05-11/; do not recreate them in active kaggle/*/ folders unless Taylor explicitly asks.

Every judge-facing Kaggle bundle must make its own bootstrap path explicit: print required Kaggle settings, fail fast on missing GPU/secrets/datasets/model sources, install DueCare from attached Kaggle wheels first, then pinned PyPI, then immutable GitHub release assets or commit-pinned archives only as a fallback, and print the resolved DueCare version before loading Gemma 4. Never rely on _reference/, local .venv, root-level legacy mirrors, untracked files, or a moving GitHub branch such as main.

Review order

  • Active script kernels: 3
  • Public-live active kernels in kaggle_live_slug_map.json: 0
  • Active kernels without a live slug: 3
Priority Notebooks / modules Why review in depth
P0 01-duecare-exploration-workbench Broad reviewer workbench: chat, harness comparison, search, knowledge extraction, bulk review, and trace inspection.
P0 02-live-demo Focused live demo and video path for the current judging story.
P1 A-00-omni-experiment-workbench Quantitative proof path: baseline, harnessed, synthetic-data, fine-tuning, judging, and report artifacts.
Optional 03-universal-llm-benchmark + 04-kaggle-community-benchmark Benchmark surfaces for proof runs; useful, but not required for the primary recording path.
P1 kaggle/kernels/* generated mirrors Reference-only generated mirror material. Do not treat it as the active submission path.

Active kernel purpose map

ID Title Status Kaggle URL Purpose
01-duecare-exploration-workbench DueCare App Tracked draft Pending publication Broad reviewer workbench for chat, harness comparison, bulk review, knowledge extraction, search, sharing, traces, and activity logs.
02-live-demo DueCare Live Demo Tracked draft Pending publication Focused live demo path for judges and video capture, using the shared Gemma 4 runtime and live-demo surface.
A-00-omni-experiment-workbench DueCare Fine-tuning and Evaluation Tracked draft Pending publication Quantitative control plane for baseline, harnessed, synthetic-data, fine-tuning, judging, checkpointing, and report evidence bundles.

Optional benchmark surfaces

These root Kaggle kernels are maintained public proof surfaces, but they are not required for the primary recording path.

ID Title Status Kaggle URL Purpose
03-universal-llm-benchmark DueCare Universal LLM Benchmark Tracked draft Pending publication Tracked DueCare notebook for DueCare Universal LLM Benchmark; review builder, setup, outputs, and publication status before relying on it in the demo.
04-kaggle-community-benchmark DueCare Kaggle Community Benchmark Tracked draft Pending publication Tracked DueCare notebook for DueCare Kaggle Community Benchmark; review builder, setup, outputs, and publication status before relying on it in the demo.

Active module deep-review queue

  1. Exploration workbench - kaggle/01-duecare-exploration-workbench/kernel.py, packages/duecare-llm-chat/src/duecare/chat/app.py, and the registered harness pages.
  2. Live demo - kaggle/02-live-demo/kernel.py, packages/duecare-llm-server/src/duecare/server, and the Cloudflare launch path.
  3. Active A-00 experiment pipeline - kaggle/A-00-omni-experiment-workbench/kernel.py, checkpointing, activity artifacts, reports, and judge options.
  4. Shared runtime and harnesses - gemma4_runtime.py, harness/__init__.py, harnesses/base.py, and harnesses/model_interface.py.
  5. Docs and contract tests - harness trinity docs, model-loading trace, A-00 parity tests, workbench UI tests, and active Kaggle state docs.

Generated by python scripts/generate_notebook_guide.py.