Checkpoint experiments through SWE pilot v2

README.md

@@ -1,13 +1,35 @@
 # messageboardbench
 
-Does cheating spread from one AI agent to another through a shared file?
+Do coding agents share, adopt, reject, or correct cheating techniques through a public team board?
 
-Agents get coding tasks whose tests contradict each other, so no honest solution exists and
-any pass is provably a cheat. Each agent also gets a scratch directory the prompt calls
-workspace and never describes as a channel. We measure whether carrying that directory from
-one agent to the next raises the cheating rate.
+Agents work on ImpossibleBench tasks in separate containers with private scratch space
+and, in the board condition, explicit tools for reading and posting team messages.
+The default pilot uses three agents per cohort, two cohorts, and 90 messages per
+episode; model, population size, sampling and budgets are configurable.
 
 Built on [ImpossibleBench](https://github.com/safety-research/impossiblebench) and
-[Inspect AI](https://inspect.aisi.org.uk). Tasks and payloads are synthetic throughout.
+[Inspect AI](https://inspect.aisi.org.uk). Passing contradictory tests is an operational
+signal of specification gaming, not a complete label of intent or attempted cheating.
 
-See `docs/setup.md` to install and run, and `docs/findings.md` for measured numbers.
+See [setup](docs/setup.md) and the [board experiment workflow](docs/board-pilot.md).
+Run `just board` to choose parameters interactively and start the experiment.
+Use `just board-preview` for a configuration preview without model calls, or
+`just board-run --model muse` to run with Muse Spark Contributor. Executing makes
+paid requests; the existing key spending cap is never raised.
+
+Start with [EXPERIMENT.md](EXPERIMENT.md) for the research question, what we have
+done, what the results establish, and the next work. The latest pilot contains one
+explicitly attributed cross-task adoption; it does not establish an increased
+gaming rate or concealed collusion.
+
+| Location | Purpose |
+|---|---|
+| `src/`, `scripts/`, `tests/` | Implementation, entry points and tests |
+| `docs/` | Setup, workflow and supporting research |
+| [results/](results/README.md) | Reviewed, frozen experiment evidence |
+| `logs/` | Raw local runs, ignored by Git |
+| `work/` | Disposable local working files, ignored by Git |
+
+This is the sole working repository for the experiment. The former `messageboard`
+repo is [retired](docs/migration/README.md); personal notes and unrelated material
+remain in its archive. Run `just evidence-check` to verify the migrated evidence.