# Startup Build v2.1 reproducibility notes

This public-safe bundle accompanies Future Shock’s paper “Building Is Not Shipping: Launch Standards in Multi-Agent AI Teams.”

Included here:

- `startup-build-v2-1.pdf` — paper PDF
- `startup-build/launch-standard-ruleset.yaml` — safe excerpt of the interaction condition and exact final-ballot frame guidance
- `startup-build/artifact-validation-manifest.json` — static validation manifest for the feasibility baseline and ballot-frame replication artifacts

Not included here yet:

- raw model outputs
- full prompt files
- provider request/response metadata
- private local paths
- secrets or credentials
- full generated artifact HTML files

Why not include raw traces immediately?

The Mac Mini evidence package contains raw outputs and prompts, and v2.1 uses them internally for quote verification. Before public release, raw traces need a privacy/security scrub for credentials, request IDs, private paths, provider metadata, and any system prompts not intended for publication.

Minimum reproducibility target:

1. Publish exact launch-standard rules and final-ballot guidance.
2. Publish a run manifest with seeds, model map, context mode, strike mode, provider route, ballot mechanics, and outcome fields.
3. Publish artifact hashes and static validation status.
4. Keep raw outputs internally hashed and available for audit until a scrubbed public trace package is ready.

Recommended future release:

A frozen evidence zip with:

- run summaries (`final_summary.json`)
- exact prompt files
- raw model outputs after scrub
- generated artifact HTML
- validation scripts
- checksums
- README mapping paper claims to files