fix: cross-tier timestamp precision mismatch (client 0.4.3)

_utc_now_iso() produced 6-digit microsecond timestamps while SQL
DEFAULTs produced 3-digit milliseconds. The width difference broke
lexicographic sorting across tiers ('Z' 0x5A > '3' 0x33 at position
24). Truncated to 3-digit milliseconds matching SQLite output.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

sibyl-memory-client/CHANGELOG.md

@@ -4,6 +4,22 @@ All notable changes to `sibyl-memory-client` are recorded here. Format
 follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). Versioning
 follows [SemVer](https://semver.org/).
 
+## [0.4.3] - 2026-05-26
+
+### Fixed
+
+- **Cross-tier timestamp precision mismatch.** `_utc_now_iso()` produced
+  6-digit microsecond timestamps (`45.525358Z`) while every SQL DEFAULT
+  used SQLite's 3-digit milliseconds (`45.525Z`). The width difference
+  broke lexicographic sorting across tiers: `'Z'` (0x5A) > `'3'` (0x33)
+  at position 24, so a journal event written 0.358 ms after an entity
+  update would sort *before* it in any `ORDER BY ts` merge. Now truncated
+  to 3-digit milliseconds to match SQLite output. Affects journal_events,
+  revenue_events, error_events, learning_runs.completed_at, and
+  skill_proposals.reviewed_at. Existing rows retain their original
+  precision (cosmetic, sort-correct within their own tier). Reported by
+  external tester smoke test on sibyl-memory-mcp 0.1.2.
+
 ## [0.4.2] - 2026-05-22
 
 `_sanitize_fts5_query` default mode flipped from phrase-match to

sibyl-memory-client/README.md

@@ -47,13 +47,15 @@ Most agent-memory products store everything on someone else's servers, treat eve
 | Frozen things, kept but out of the way | ARCHIVE | `archive_entity(kind, name)` |
 | Search across everything | FTS5 | `search_entities(query)` |
 
-## What's in v0.2.x
+## What's in v0.4.x
 
 - The full five-tier memory model and the API surface above.
+- Cross-tier FTS5 search across entities, state, reference, and journal tiers.
 - Multi-tenant isolation: one machine can hold separate memory for separate identities.
 - Self-learning module (paid-tier): the agent watches your patterns and proposes reusable skills.
 - Memory linter (paid-tier): a health check on the local database.
 - Tier gating: free-tier callers get clear errors pointing at the upgrade page; paid-tier callers get full access.
+- Uniform millisecond timestamp precision across all tiers (v0.4.3).
 
 ## Tier model
 

sibyl-memory-client/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "sibyl-memory-client"
-version = "0.4.2"
+version = "0.4.3"
 description = "Local-first agentic memory SDK. SQLite-backed five-tier hierarchical schema, FTS5 search, multi-tenant, with self-learning skill detection and local memory linter. Foundation of the Sibyl Memory Plugin family."
 authors = [{ name = "SIBYL, Sibyl Labs LLC", email = "sibyl@sibyllabs.org" }]
 license = { text = "MIT" }

sibyl-memory-client/src/sibyl_memory_client/storage.py

@@ -40,13 +40,15 @@ _DB_SIDECAR_SUFFIXES = ("-wal", "-shm")
 
 
 def _utc_now_iso() -> str:
-    """Return current UTC time in ISO 8601 microsecond-precision format.
-
-    Higher precision than the schema's millisecond default (`%f` in
-    SQLite's strftime) reduces ts collisions when multiple writes land in
-    the same millisecond. Microsecond-precision strings sort correctly
-    lexically since the format is fixed-width."""
-    return datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%S.%fZ")
+    """Return current UTC time in ISO 8601 millisecond-precision format.
+
+    Matches the 3-digit precision of SQLite's ``strftime('%f')`` so that
+    timestamps produced by Python and by SQL DEFAULTs sort identically in
+    lexicographic comparisons.  Prior versions emitted 6-digit microseconds
+    which broke cross-tier ``ORDER BY ts`` merges ('Z' > '3' at position 24).
+    Fixed in 0.4.3."""
+    now = datetime.now(timezone.utc)
+    return now.strftime("%Y-%m-%dT%H:%M:%S.") + f"{now.microsecond // 1000:03d}Z"
 
 
 def new_id() -> str: