diff --git a/program.md b/program.md
index dea9bcc01..f5d2f0507 100644
--- a/program.md
+++ b/program.md
@@ -11,7 +11,7 @@ To set up a new experiment, work with the user to:
 3. **Read the in-scope files**: The repo is small. Read these files for full context:
    - `README.md` — repository context.
    - `prepare.py` — fixed constants, data prep, tokenizer, dataloader, evaluation. Do not modify.
-   - `train.py` — the file you modify. Model architecture, optimizer, training loop.
+   - `train.py` — the source file you modify. Model architecture, optimizer, training loop.
 4. **Verify data exists**: Check that `~/.cache/autoresearch/` contains data shards and a tokenizer. If not, tell the human to run `uv run prepare.py`.
 5. **Initialize results.tsv**: Create `results.tsv` with just the header row. The baseline will be recorded after the first run.
 6. **Confirm and go**: Confirm setup looks good.
@@ -23,7 +23,14 @@ Once you get confirmation, kick off the experimentation.
 Each experiment runs on a single GPU. The training script runs for a **fixed time budget of 5 minutes** (wall clock training time, excluding startup/compilation). You launch it simply as: `uv run train.py`.
 
 **What you CAN do:**
-- Modify `train.py` — this is the only file you edit. Everything is fair game: model architecture, optimizer, hyperparameters, training loop, batch size, model size, etc.
+- Modify `train.py` — this is the only source file you edit. Everything is fair game: model architecture, optimizer, hyperparameters, training loop, batch size, model size, etc.
+- Do not modify `prepare.py` or any other source file.
+
+Experiment artifacts and bookkeeping:
+- You must append exactly one new row to `results.tsv` after every experiment, including `keep`, `discard`, and `crash`.
+- You must append the row before any `git reset`, checkout, or revert.
+- Updating `results.tsv` is required bookkeeping and is not a source-code modification.
+- You may write or overwrite `run.log` for each run.
 
 **What you CANNOT do:**
 - Modify `prepare.py`. It is read-only. It contains the fixed evaluation, data loading, tokenizer, and training constants (time budget, sequence length, etc).