scripts: wt-h2-count.py — worldtree-dev's U11b H2 gate check (verbatim); rehearsed on copies, 0 on both

scripts/wt-h2-count.py

@@ -0,0 +1,46 @@
+"""H2 re-check: count legacy chunks with scope == agent_self per agent store.
+
+Provenance: written by worldtree-dev against the v1.0.0b192 tag and delivered
+2026-09-30 (thread 01M3SGEQDRQD7DWBVT4K73FAHP) for the U11b step-5 gate. Run it
+VERBATIM, right before each instance's legacy-data deletion, inside that
+instance's worldtree-api container (it pins the chromadb that wrote the stores):
+
+    docker exec -i <worldtree-api container> python - < scripts/wt-h2-count.py ; echo "exit=$?"
+
+Exit 0 = global count 0, go. Exit 2 = non-zero: STOP, send the output to
+worldtree-dev before deleting anything. Attach the full stdout to the stamp.
+
+⚠ chromadb rewrites HNSW sidecars and the sqlite file ON OPEN: the store's
+bytes change even though no row does. Compare live sha256s against the archive
+manifest BEFORE running this, or run it against a `cp -a` copy (point
+WORLDTREE_DATA_PATH at the copy's root).
+
+Opens each <WORLDTREE_DATA_PATH>/agents/<agent>/memory/<agent>.chroma with
+chromadb and counts rows whose metadata scope is "agent_self" (the legacy
+reader's own where-clause). A missing store dir or collection counts as 0
+and says so. Exit 0 when the global total is 0, exit 2 otherwise.
+"""
+import os, sys
+from pathlib import Path
+import chromadb
+
+root = Path(os.environ["WORLDTREE_DATA_PATH"])
+agents = sys.argv[1:] or ["forseti", "lofn", "mimir"]
+total = 0
+for agent in agents:
+    d = root / "agents" / agent / "memory" / f"{agent}.chroma"
+    if not d.is_dir():
+        print(f"{agent}: store absent ({d}) -> agent_self=0")
+        continue
+    client = chromadb.PersistentClient(path=str(d))
+    try:
+        col = client.get_collection("memory_chunks")
+    except Exception as exc:
+        print(f"{agent}: no memory_chunks collection ({type(exc).__name__}) -> agent_self=0")
+        continue
+    n_all = col.count()
+    n_self = len(col.get(where={"scope": "agent_self"}, include=[])["ids"])
+    total += n_self
+    print(f"{agent}: rows={n_all} agent_self={n_self}")
+print(f"H2 global agent_self count: {total}")
+sys.exit(0 if total == 0 else 2)