docs: opt-in W&B/Weave runbook for AppSprout Crashlanded benchmarks (#68)

* docs: document opt-in W&B/Weave for AppSprout Crashlanded runs

Keep open-source RLE wandb-optional: AppSprout benchmark recipe only,
no --wandb hook implied for run_scenario.py.

Co-authored-by: Jason <jkbennitt@users.noreply.github.com>

* docs: use a stable heading slug for the W&B runbook

Avoid '&' in the README heading so the CONTRIBUTING anchor is unambiguous.

Co-authored-by: Jason <jkbennitt@users.noreply.github.com>

* docs: note Weave is optional and not in the tracking extra

uv sync --extra tracking installs wandb + huggingface-hub only.
weave is a separate import, not locked, and stays a no-op if missing.

Co-authored-by: Jason <jkbennitt@users.noreply.github.com>

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Jason <jkbennitt@users.noreply.github.com>

Author: jkbennitt

CONTRIBUTING.md

@@ -203,7 +203,7 @@ docker/                    # HeadlessRim Dockerfile, compose, entrypoint
 - [felix-agent-sdk](https://github.com/AppSprout-dev/felix-agent-sdk) >= 0.3.0
 - [RIMAPI](https://github.com/IlyaChichkov/RIMAPI) C# mod
 - httpx, pydantic >= 2.0, pyyaml
-- Optional: wandb, huggingface-hub (`uv sync --extra tracking`)
+- Optional: wandb, huggingface-hub (`uv sync --extra tracking`). W&B stays opt-in (`--wandb` on `run_benchmark.py` only). Weave is a separate optional import (not in the extra or lockfile); see [README — Weights and Biases (opt-in)](README.md#weights-and-biases-opt-in).
 
 ## Questions?
 

README.md

@@ -228,6 +228,36 @@ python scripts/visualize_results.py results/ --all
 python scripts/analyze_spread.py --spread-dir results/spread
 ```
 
+### Weights and Biases (opt-in)
+
+W&B is **strictly opt-in**. Open-source RLE does not require `wandb`, does not enable it by default, and degrades gracefully if the package is missing or unauthenticated. Third parties who omit `--wandb` need no W&B setup.
+
+AppSprout Crashlanded **benchmark** runs should use W&B / Weave. That is an AppSprout ops policy, not a core default and not a required dependency.
+
+The optional `tracking` extra installs `wandb` and `huggingface-hub` only. `uv sync --extra tracking` enables W&B. Weave tracing is a separate optional import in `wandb_logger.py` and is not in `uv.lock`; install it yourself (`uv pip install weave`) if you want traces, or expand the extra later. Missing `weave` is a no-op.
+
+`scripts/run_scenario.py` has **no `--wandb` hook** today. Do not assume a single-scenario run will log to W&B unless that hook is added later. This recipe is for `run_benchmark.py` only.
+
+```bash
+# Optional extra — not part of the default install (wandb + huggingface-hub)
+uv sync --extra tracking
+
+# Weave traces only if weave is installed separately (not in the tracking extra)
+# uv pip install weave
+
+# Authenticate only if this machine is not already logged in.
+# Never commit a W&B key. WANDB_API_KEY may be set in the environment
+# (do not print it).
+wandb login
+
+python scripts/run_benchmark.py \
+  --wandb \
+  --no-think --no-pause \
+  --ticks 10 --tick-interval 30 --seed 42 \
+  --output results/crashlanded \
+  --harness felix --harness baseline
+```
+
 ## Benchmark Results
 
 **Live board — `spread-2026-09-06`, scoring 1.2, harness × model** — Crashlanded, 10 ticks, seed 42, N=1. **Content-first — not statistically valid (no confidence intervals). Winners advance to N=4; N=4 is not published.** Ranked by mean composite across the run. Featured numbers live on [rle.appsprout.dev](https://rle.appsprout.dev) and the [HF card](https://huggingface.co/datasets/AppSprout/rle-benchmarks) (`runs/spread-2026-09-06`).