Post-audit fixes from launch review

HONEST_RESULTS.md "Files of record": drop the stale "trained checkpoints
not shipped" claim — they ARE shipped now (the 944K Atome blob and PyTorch
source, plus the vanilla FP32 baseline used in the reversal A/B). The 60K
ab_sweep stays unshipped because it's 24 throwaway runs reproducible in
~20 min of CPU.

scripts/demo.py + scripts/evaluate.py docstrings: point at the bundled
atome_1m_v1.pt as the example checkpoint, so a fresh clone can run them
without training first. (train_demo.py still defaults to atome_demo.pt
because that's the train-from-scratch flow.)

Author: Atome LM Team

HONEST_RESULTS.md

@@ -103,14 +103,20 @@ row (negligible % overhead at 944K).
 
 ## Files of record
 
-Trained checkpoints (`*.pt`, `checkpoints/ab_sweep/`) are **not shipped** in
-this public repository — Atome LM is train-from-scratch (see README). The
-training *logs* below ship, so every reported number stays auditable
-step-by-step:
-
-- `checkpoints/atome_1m_v1.train.json` — 944K Atome, every-1000-step log.
-- `checkpoints/vanilla_1m_v1.train.json` — 944K vanilla FP32, same.
+The trained 944K checkpoints and their training logs ship with the kit, so
+every reported number is auditable step-by-step *and* re-evaluable directly:
+
+- `checkpoints/atome_944k.bin` — packed C-engine blob (ATOME01 format).
+- `checkpoints/atome_1m_v1.pt` — 944K Atome PyTorch source.
+- `checkpoints/vanilla_1m_v1.pt` — 944K vanilla FP32 baseline (for the
+  reversal A/B above).
+- `checkpoints/atome_1m_v1.train.json` — every-1000-step training log.
+- `checkpoints/vanilla_1m_v1.train.json` — same for the vanilla baseline.
 - `ab_results.json` — exact 60K A/B result row.
 - `FRONTIER.md` — frontier writeup with full 944K disclosure.
 - `PAPER.md` — architecture writeup.
 - `tests/` — 146 green tests.
+
+The 60K sweep itself (`checkpoints/ab_sweep/`) is **not** shipped — that
+was 24 throwaway training runs. Reproducing the sweep takes ~20 minutes
+of CPU using the included `scripts/`.

scripts/demo.py

@@ -11,7 +11,7 @@
 
 Usage
 -----
-    python scripts/demo.py --checkpoint checkpoints/atome_demo.pt
+    python scripts/demo.py --checkpoint checkpoints/atome_1m_v1.pt
     python scripts/demo.py --random                    # random-init,
                                                        # for plumbing checks
     python scripts/demo.py --random --temperature 1.0 --top-p 0.9

scripts/evaluate.py

@@ -11,8 +11,8 @@
 
 Usage:
     python scripts/evaluate.py \\
-        --checkpoint checkpoints/atome_demo.pt \\
-        --data data/tinystories.txt \\
+        --checkpoint checkpoints/atome_1m_v1.pt \\
+        --data data/sample.txt \\
         --max-bytes 100000
 """
 from __future__ import annotations