# Colab Notebook Design
This crate now has two notebook entry points:
- `colab/dsfb_computer_graphics_demo.ipynb` for the legacy synthetic artifact
- `colab/dsfb_unreal_native_evidence.ipynb` for the strict Unreal-native evidence path
The Unreal-native notebook is the canonical notebook for real engine-native diligence.
The notebook exists to make the crate evaluable from a clean browser session. It does not reimplement the artifact in Python. Instead, it installs the small set of runtime dependencies needed for Colab, clones the repository, builds `dsfb-computer-graphics`, and drives the Rust CLI so the crate remains the source of truth for Demo A, Demo B, metrics, figures, and reports.
For the Unreal-native path, the notebook is a replay-and-review surface for an existing real Unreal-native dataset bundle. It does not pretend Colab is generating Unreal exports itself.
“The experiment is intended to demonstrate behavioral differences rather than establish optimal performance.”
## Design Goals
- keep the notebook honest about scope and limits
- make repeated runs safe by default with timestamped output directories
- generate a reviewer-ready bundle without requiring a local graphics toolchain
- display the major figures inline so a reviewer can inspect the artifact before downloading anything
- surface the expanded scenario suite, stronger baselines, and mixed-outcome cases without asking the reviewer to inspect JSON manually
- package the current run as both a PDF and a ZIP for archiving and external review
## Output Organization
The notebook writes under the crate-local root `output-dsfb-computer-graphics/`.
Each execution creates a timestamped subdirectory such as:
```text
output-dsfb-computer-graphics/
output-dsfb-computer-graphics-YYYYMMDD-HHMMSS/
```
The timestamped run directory holds the generated artifacts for that execution, including:
- `artifact_manifest.json`
- `scenario_suite_manifest.json`
- `figures/`
- `scenarios/`
- `metrics.json`
- `report.md`
- `reviewer_summary.md`
- `five_mentor_audit.md`
- `check_signing_blockers.md`
- `demo_b_decision_report.md`
- `completion_note.md`
- `artifacts_bundle.pdf`
- `demo_b/`
The ZIP archive is written one level above the run directory and uses the same timestamped run name. This layout prevents accidental overwrite and keeps each run self-describing.
## PDF and ZIP Bundling
The notebook calls the crate-local script `colab/build_artifact_bundle.py` after `cargo run -- run-all --output <run-dir>`.
The bundle script:
- reads `artifact_manifest.json` to discover the actual outputs from the run
- rasterizes the SVG figures for a clean PDF export
- assembles a real PDF bundle containing the main metrics, figures, and scope statement
- creates a ZIP archive of the entire timestamped artifact directory
This is intended to support review convenience and archival traceability, not to imply production packaging completeness.
## Assumptions
- the notebook runs in a Colab-like Linux environment with shell access
- the environment can install `librsvg2-bin`, `zip`, and the small Python dependencies used for display and bundling
- the repository URL and branch configured near the top of the notebook may need to be updated by the user for a fork or a different default branch
## What The Notebook Does Not Claim
- It does not claim deployment readiness.
- It does not claim optimal TAA tuning or optimal adaptive sampling.
- It does not claim measured production GPU performance.
- It does not claim superiority over full commercial temporal reconstruction stacks.
The legacy synthetic notebook is a polished access path to a bounded synthetic artifact, not a substitute for broader engine integration or field validation.
For the strict engine-native path, use `colab/dsfb_unreal_native_evidence.ipynb` and keep the provenance label `unreal_native` honest.