From 16208b3a7a27b0bf946c1bf710278149938ce7b5 Mon Sep 17 00:00:00 2001
From: "google-labs-jules[bot]"
 <161369871+google-labs-jules[bot]@users.noreply.github.com>
Date: Tue, 28 Jul 2026 17:29:31 +0000
Subject: [PATCH] =?UTF-8?q?=E2=9A=A1=20Bolt:=20Cache=20deterministic=20con?=
 =?UTF-8?q?text=20lexicons?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Added `functools.lru_cache` to `_compiled_context_lexicon` in `openmed.clinical.context`.
- Prevents expensive regex compilation loops across span evaluations.
- Created learning log in `.jules/bolt.md`.

Co-authored-by: zrt219 <199104500+zrt219@users.noreply.github.com>
---
 .jules/bolt.md                      | 3 +++
 openmed/openmed/clinical/context.py | 4 ++++
 2 files changed, 7 insertions(+)
 create mode 100644 .jules/bolt.md

diff --git a/.jules/bolt.md b/.jules/bolt.md
new file mode 100644
index 0000000..c61fe04
--- /dev/null
+++ b/.jules/bolt.md
@@ -0,0 +1,3 @@
+## 2024-06-25 - Context Lexicon Regex Compilation Bottleneck
+**Learning:** Evaluated span assertions in the openmed NLP pipeline heavily depend on repeated deterministic cue compilations via `_compiled_context_lexicon`. This function constructs and formats numerous regexes for every evaluated axis (negation, temporality, certainty, etc). Doing this on every decision axis per text/span iteration caused massive redundant regex compilations and became a significant CPU bottleneck on the main thread.
+**Action:** Always memoize (e.g. `functools.lru_cache`) deterministic regex lexicons and configuration functions where inputs are purely hashable parameters like `language` strings.
diff --git a/openmed/openmed/clinical/context.py b/openmed/openmed/clinical/context.py
index 9fd11df..9546489 100644
--- a/openmed/openmed/clinical/context.py
+++ b/openmed/openmed/clinical/context.py
@@ -39,6 +39,7 @@
 from collections.abc import Iterable, Iterator, Mapping, Sequence
 from dataclasses import dataclass, replace
 from datetime import date
+from functools import lru_cache
 from typing import Any, Literal
 
 from openmed.clinical.lexicons import (
@@ -154,6 +155,9 @@ class _CompiledContextLexicon:
     backward_context_cues: frozenset[str]
 
 
+# ⚡ Bolt: Cache deterministic context lexicons to avoid compiling the same
+# regex patterns thousands of times during span/text evaluations.
+@lru_cache(maxsize=32)
 def _compiled_context_lexicon(language: str | None = None) -> _CompiledContextLexicon:
     lexicon = get_clinical_cue_lexicon(language)
     token_boundaries = lexicon.token_boundaries