Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Studio: keep Deep Research tables intact when a cited title has a pipe (#12990)

Nilay683f0417760a
diff --git a/studio/backend/core/research/citations.py b/studio/backend/core/research/citations.py
index a4213f68c23..fd575d33ca3 100644
--- a/studio/backend/core/research/citations.py
+++ b/studio/backend/core/research/citations.py
@@ -245,7 +245,7 @@ def _validate_masked_sources(report: str, sources: list[dict], placeholders: dic
source = source_by_url.get(url)
if source is None:
return None
- title = _citation_title(source, url)
+ title = re.sub(r"(\\*)\|", r"\1\1\\|", _citation_title(source, url))
token = _placeholder("research-citation", len(placeholders))
placeholders[token] = f"[{title}]({_escape_link_destination(url)})"
return token
diff --git a/studio/backend/tests/test_research_runs_hardening.py b/studio/backend/tests/test_research_runs_hardening.py
index ec59c841001..2e8e7e83116 100644
--- a/studio/backend/tests/test_research_runs_hardening.py
+++ b/studio/backend/tests/test_research_runs_hardening.py
@@ -18,6 +18,7 @@ from types import SimpleNamespace
import httpx
import pytest
from fastapi import HTTPException
+from markdown_it import MarkdownIt
from core import research_runs
from core.research.citations import (
@@ -257,9 +258,7 @@ def test_document_citation_regex_does_not_backtrack_catastrophically():
def test_citation_title_strips_brackets_for_catalog_and_citation():
- # Search titles routinely carry a bracketed prefix ("[PDF] ..."), and the prompt tells the
- # model to copy the catalog title verbatim into the link label, where a bracket makes the
- # citation unmatchable. Catalog and citation writer share this helper so they agree.
+ # brackets make verbatim link labels unmatchable; catalog and citation paths share this helper.
assert (
_citation_title({"title": "[PDF] Annual Report 2024"}, "https://x/a")
== "PDF Annual Report 2024"
@@ -268,10 +267,28 @@ def test_citation_title_strips_brackets_for_catalog_and_citation():
assert _citation_title({}, "https://x/a") == "https://x/a"
+def test_citation_title_with_a_pipe_keeps_its_table_row_intact():
+ report = "| Model | Context |\n|---|---|\n| Qwen3 [blog](https://q.example/) | 128K |"
+ sources = [{"url": "https://q.example/", "title": "Qwen3: Think Deeper | Qwen"}]
+ tokens = MarkdownIt("commonmark").enable("table").parse(_validate_report(report, sources, []))
+ body = tokens[[token.type for token in tokens].index("tbody_open") :]
+ row = [token.children for token in body if token.type == "inline"]
+ assert len(row) == 2
+ assert [child.type for child in row[0]] == ["text", "link_open", "text", "link_close"]
+ assert row[0][1].attrGet("href") == "https://q.example/"
+ assert row[0][2].content == "Qwen3: Think Deeper | Qwen"
+ assert row[1][0].content == "128K"
+
+
+def test_citation_title_with_an_escaped_pipe_keeps_its_backslash_and_its_row():
+ report = "| Tool | Note |\n|---|---|\n| grep [doc](https://g.example/) | alternation |"
+ sources = [{"url": "https://g.example/", "title": r"grep a\|b | Docs"}]
+ validated = _validate_report(report, sources, [])
+ assert r"[grep a\\\|b \| Docs](https://g.example/)" in validated
+
+
def test_prompt_budget_counts_the_whole_prompt(monkeypatch):
- # Budgeting only the evidence cannot prevent an overflow: at a small context the
- # untrimmable scaffolding (system prompt, plan, source catalogs) is already several times
- # the window, and the old floor added 1500 chars on top of that.
+ # fixed scaffolding can exceed a small context before any trimmable evidence is added.
monkeypatch.setattr(research_runs, "_loaded_context_length", lambda _inf = None: None)
assert research_runs._prompt_char_budget(4096) is None
assert research_runs._trimmable_budget(None, 99_999, 500) == 500
@@ -279,7 +296,6 @@ def test_prompt_budget_counts_the_whole_prompt(monkeypatch):
monkeypatch.setattr(research_runs, "_loaded_context_length", lambda _inf = None: 16384)
total = research_runs._prompt_char_budget(4096)
assert total == int((16384 - 4096) * research_runs._SYNTHESIS_EVIDENCE_CHARS_PER_TOKEN)
- # A trimmable section never exceeds what is left, and never goes negative.
assert research_runs._trimmable_budget(total, 0, 1_000) == 1_000
assert research_runs._trimmable_budget(total, total - 10, 1_000) == 10
assert research_runs._trimmable_budget(total, total + 5_000, 1_000) == 0