Make the material-memory example a lightweight reproducible starter

MEMORY_EXAMPLE.md

@@ -10,19 +10,30 @@ parameters cannot distinguish them.
 ## Run it
 
 ```sh
-pip install numpy h5py matplotlib huggingface_hub
+pip install numpy matplotlib huggingface_hub
 python show_material_memory.py
 ```
 
-Download [the script](show_material_memory.py) first. It fetches one pinned HDF5
-shard (not the entire dataset), checks its SHA-256, and produces a PNG, a CSV
+Download [the script](show_material_memory.py) first. It fetches a **4.5 KB training
+excerpt**, checks its SHA-256, and produces a PNG, a CSV
 containing all 121 states, and a JSON evidence record. The figure displays
 states 0–60 so the two returns are easy to follow. No solver, GPU, model weights
 or paid service is required.
 
-Already have the data?
+The excerpt contains the four unchanged arrays used here, including the complete
+strain tensors. It is a lossless copy from the pinned original shard, not a new
+simulation or an extra dataset sample. Source provenance stays in the output.
+
+For offline use, also download [the small excerpt](example/training-history-00252.npz):
+
+```sh
+python show_material_memory.py --sample training-history-00252.npz
+```
+
+Already have the original data? Install `h5py` to read its HDF5 shard:
 
 ```sh
+pip install h5py
 python show_material_memory.py --shard data/t2_material_loading_memory_v1/shards/part-00002.h5
 ```
 
@@ -51,7 +62,8 @@ or stress of the opposite sign? Then inspect the CSV to trace how it got there.
 不足以确定此刻的应力，还需要加载历史或内部状态。
 
 这是合成材料点数据中的一个训练样本，不是实测数据，也不是 DENIM 的预测结果。
-运行脚本即可生成图、完整历程表和校验记录。图中使用有符号等效应力；加载步
+运行脚本只需下载 4.5 KB 的训练样本摘录，即可生成图、完整历程表和校验记录。
+摘录与原数据逐项一致，不新增样本。图中使用有符号等效应力；加载步
 不是实际时间。没有读取封存测试集，也没有修改训练、验证或测试划分。
 
 [Explore the interactive lab](https://lab.haoming-luo.com/material-memory/) ·

show_material_memory.py

@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 """Show two zero-strain states with different stresses in published train data.
 
-Install: pip install numpy h5py matplotlib huggingface_hub
+Install: pip install numpy matplotlib huggingface_hub
 Run: python show_material_memory.py
 Offline: python show_material_memory.py --shard /path/to/part-00002.h5
 Code: Apache-2.0; source data: CC BY 4.0, Haoming Luo / AgentFEM.
@@ -12,7 +12,6 @@ import hashlib
 import json
 from pathlib import Path
 
-import h5py
 import numpy as np
 
 REPO = "HaomingLuo/AgentFEM-Material-Loading-Memory"
@@ -20,20 +19,32 @@ REVISION = "cc9644aa5b51ac4f0ff5cccc9a42452de0d220a3"
 FILE = "data/t2_material_loading_memory_v1/shards/part-00002.h5"
 SHA256 = "952b90eed1ce3ac288487cf31180384fc12f79aa03d95e4885bfd73b85c9a46d"
 SAMPLE = "00252"  # First training trajectory in the symmetric-cyclic J2 cohort.
+EXCERPT_REVISION = "62cedc709b159c9faa652e0aff34ff48771b3f07"
+EXCERPT_FILE = "example/training-history-00252.npz"
+EXCERPT_SHA256 = "c16bd5da346dd6926d32c730ad9ad040c5bd9a51dccba92b1f8c4ae1dce8717f"
+ARRAY_NAMES = ("total_strain", "signed_equivalent_strain",
+               "signed_equivalent_stress_pa", "equivalent_plastic_strain")
 
 
-def read_sample(path):
+def read_sample(path, excerpt=False):
     digest = hashlib.sha256(Path(path).read_bytes()).hexdigest()
-    if digest != SHA256:
-        raise ValueError("Shard checksum mismatch; use the pinned published file.")
-    with h5py.File(path, "r") as f:
-        g = f[SAMPLE]
-        if str(g.attrs["split"]) != "train":
-            raise ValueError("Illustration must use training data only.")
-        arrays = {name: g[name][:] for name in (
-            "total_strain", "signed_equivalent_strain",
-            "signed_equivalent_stress_pa", "equivalent_plastic_strain")}
-        case_id = str(g.attrs["case_id"])
+    if digest != (EXCERPT_SHA256 if excerpt else SHA256):
+        raise ValueError("Input checksum mismatch; use the pinned published file.")
+    if excerpt:
+        with np.load(path, allow_pickle=False) as packed:
+            if (str(packed["split"]) != "train" or str(packed["sample_id"]) != SAMPLE
+                    or str(packed["source_sha256"]) != SHA256):
+                raise ValueError("Unexpected excerpt provenance.")
+            arrays = {name: packed[name] for name in ARRAY_NAMES}
+            case_id = str(packed["case_id"])
+    else:
+        import h5py  # Only needed when inspecting the complete source shard.
+        with h5py.File(path, "r") as f:
+            g = f[SAMPLE]
+            if str(g.attrs["split"]) != "train":
+                raise ValueError("Illustration must use training data only.")
+            arrays = {name: g[name][:] for name in ARRAY_NAMES}
+            case_id = str(g.attrs["case_id"])
     if not all(np.isfinite(a).all() for a in arrays.values()):
         raise ValueError("Non-finite trajectory.")
     # First two returns to zero after the virgin state, not maximum-gap selection.
@@ -49,16 +60,22 @@ def read_sample(path):
 
 def main():
     parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument("--shard", type=Path)
+    inputs = parser.add_mutually_exclusive_group()
+    inputs.add_argument("--shard", type=Path, help="Original HDF5 shard (requires h5py)")
+    inputs.add_argument("--sample", type=Path, help="Local 4.5 KB NPZ training excerpt")
     parser.add_argument("--out", type=Path, default=Path("memory-example"))
     args = parser.parse_args()
-    if args.shard is None:
+    if args.shard is None and args.sample is None:
         from huggingface_hub import hf_hub_download
-        args.shard = Path(hf_hub_download(REPO, FILE, repo_type="dataset", revision=REVISION))
-    d, a, b, case_id = read_sample(args.shard)
+        args.sample = Path(hf_hub_download(REPO, EXCERPT_FILE, repo_type="dataset", revision=EXCERPT_REVISION))
+    using_excerpt = args.sample is not None
+    d, a, b, case_id = read_sample(args.sample if using_excerpt else args.shard, excerpt=using_excerpt)
     strain = d["signed_equivalent_strain"] * 100
     stress = d["signed_equivalent_stress_pa"] / 1e6
     report = {"repository": REPO, "revision": REVISION, "shard_sha256": SHA256,
+              "input_kind": "lossless_training_excerpt" if using_excerpt else "original_shard",
+              "input_sha256": EXCERPT_SHA256 if using_excerpt else SHA256,
+              "excerpt_revision": EXCERPT_REVISION if using_excerpt else None,
               "sample_id": SAMPLE, "case_id": case_id, "split": "train",
               "selection": "First two noninitial returns to zero strain; no interpolation",
               "states": [{"index": i, "signed_equivalent_strain": float(strain[i]/100),