topos-node 1.3.33__tar.gz → 1.3.35__tar.gz

{topos_node-1.3.33/topos_node.egg-info → topos_node-1.3.35}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: topos-node
-Version: 1.3.33
+Version: 1.3.35
 Summary: Topos personal AI engine (FastAPI): local data, sync, control plane WebSocket client
 License-Expression: Apache-2.0
 Project-URL: Homepage, https://topos.dialogues.ai

{topos_node-1.3.33 → topos_node-1.3.35}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "topos-node"
-version = "1.3.33"
+version = "1.3.35"
 description = "Topos personal AI engine (FastAPI): local data, sync, control plane WebSocket client"
 readme = "README.md"
 license = "Apache-2.0"

{topos_node-1.3.33 → topos_node-1.3.35}/topos/__version__.py

@@ -1,6 +1,6 @@
 """Topos Control Plane Version Information."""
 
-__version__ = "1.3.33"
-__version_info__ = (1, 3, 33)
+__version__ = "1.3.35"
+__version_info__ = (1, 3, 35)
 
 __all__ = ["__version__", "__version_info__"]

topos_node-1.3.35/topos/analytics/dataset_resolution.py

@@ -0,0 +1,67 @@
+"""Which dataset holds the owner's messaging, when the caller cannot say.
+
+`/v1/ingestion/datasets` returns ZERO rows on a node whose messages arrived by sync rather
+than upload — measured on the live node 2026-08-27. Every messaging screen resolves its
+dataset from that list, so a database holding 7,668 messages, 180 dyads and 4,048 entities
+rendered as an empty Social page and an empty Luck page at the same time. The engine is one
+GROUP BY away from the answer.
+
+Lives on its own so the luck read and the messenger-analytics handlers share ONE rule; two
+copies would drift, and the two screens would then disagree about which dataset the node is.
+"""
+
+from __future__ import annotations
+
+import sqlite3
+from typing import Any, Tuple
+
+
+def has_messaging_substrate(conn: Any, dataset_id: str) -> bool:
+    """Does THIS dataset carry the messages the messaging screens read?
+
+    Cheap existence probes, not counts: the only job is to tell "nothing happened" apart
+    from "this dataset cannot answer".
+    """
+    if not str(dataset_id or "").strip():
+        return False
+    for sql in (
+        "SELECT 1 FROM messenger_dyad_stats WHERE dataset_id=? LIMIT 1",
+        "SELECT 1 FROM conversation_messages WHERE dataset_id=? AND is_from_self=1 LIMIT 1",
+    ):
+        try:
+            if conn.execute(sql, (dataset_id,)).fetchone():
+                return True
+        except sqlite3.Error:
+            continue
+    return False
+
+
+def resolve_primary_dataset(conn: Any) -> str:
+    """The dataset with the most conversation messages, or empty if there are none."""
+    try:
+        row = conn.execute(
+            "SELECT dataset_id, COUNT(*) n FROM conversation_messages"
+            " WHERE dataset_id IS NOT NULL GROUP BY dataset_id ORDER BY n DESC LIMIT 1"
+        ).fetchone()
+    except sqlite3.Error:
+        return ""
+    return str(row[0]) if row and row[0] else ""
+
+
+def resolve_messaging_dataset(conn: Any, requested: str) -> Tuple[str, bool]:
+    """(dataset_to_use, engine_resolved_it).
+
+    Falls back when the caller names nothing, or names a dataset with no messages — the
+    second case matters because answering it literally reports real work with "who has heard
+    is unknown", which is true of the id and useless to the person.
+
+    Callers MUST surface `engine_resolved_it`; a silent substitution would show one dataset's
+    data under another's name.
+    """
+    requested = str(requested or "").strip()
+    if requested and has_messaging_substrate(conn, requested):
+        return requested, False
+    primary = resolve_primary_dataset(conn)
+    if primary and primary != requested:
+        return primary, True
+    return requested, False