fix(query): reach the not-indexed marker on the not_found path

[tirth8205]

Follow-up to #884, found by smoke-testing the merged feature against this repository's own graph rather than trusting the tests.

The gap

query_graph returns early at query.py:375 with status: "not_found" for every pattern except consumers_of and file_summary. An unresolved target therefore never reached the empty-result path where #884 attaches the marker. The feature's most valuable case, telling an agent that a zero means "I never indexed this", was unreachable for 14 of the 16 query patterns in production. The original test did not catch it because it exercised file_summary, one of the two patterns that skip that branch.

The fix

Attach the marker to the not_found response, and check for an empty or stale graph before declaring a target simply unindexed. Priority now matches what semantic_search_nodes already did: empty graph, then stale graph, then genuinely not indexed.

A stale graph explains the miss and has a remedy, so it outranks wording that reads like a permanent parser limitation.

Verified against a real stale graph

This repository's checked-in graph was built on 2026-05-16 at commit 52cf3bc, with HEAD at 455a85a. Querying a function that plainly exists in the source:

Before:

status: not_found, confidence: (absent)

After #884 alone:

target not indexed: no node matching 'attach_context_savings', so this 0 is not evidence that none exist

After this change:

graph is stale: no node matching 'attach_context_savings'; the graph predates HEAD, so run `code-review-graph update` first

The first is silent, the second blames the parser, the third names the actual cause and the fix.

Tests

9 new tests: the marker is present on not_found across five patterns (the reachability regression), stale beats not-indexed, current graph still says not-indexed, empty graph says build, and a resolved target with results still carries no marker so the token guarantee holds on this path too.

Full suite 2768 passed, 5 skipped, 2 xpassed in a clean checkout. ruff and mypy clean.

[github-actions]

code-review-graph review

Overall risk: 0.75 (HIGH) — 8 changed function(s)/class(es), 0 affected flow(s), 1 test gap(s)

Risk-scored changes

Risk	Level	Symbol	Location	Tested
0.75	high	code_review_graph/tools/query.py::query_graph	code_review_graph/tools/query.py:249	yes
0.35	low	code_review_graph/uncertainty.py::unresolved_stale_note	code_review_graph/uncertainty.py:248	no
0.35	low	code_review_graph/uncertainty.py::empty_query_confidence	code_review_graph/uncertainty.py:343	yes
0.05	low	tests/test_uncertainty.py::test_not_found_response_carries_a_marker	tests/test_uncertainty.py:485	(test)
0.05	low	tests/test_uncertainty.py::test_unresolved_target_on_a_stale_graph_says_stale_not_unindexed	tests/test_uncertainty.py:496	(test)
0.05	low	tests/test_uncertainty.py::test_unresolved_target_on_a_current_graph_says_not_indexed	tests/test_uncertainty.py:516	(test)
0.05	low	tests/test_uncertainty.py::test_unresolved_target_on_an_empty_graph_says_build	tests/test_uncertainty.py:531	(test)
0.05	low	tests/test_uncertainty.py::test_resolved_target_with_results_still_has_no_marker	tests/test_uncertainty.py:546	(test)

Test gaps

• code_review_graph/uncertainty.py::unresolved_stale_note (code_review_graph/uncertainty.py:248)

Token savings: this graph-backed report used ~17,804 fewer tokens (~91%) than reading every changed file in full (estimated, chars/4 approximation).

---

Powered by code-review-graph — local-first analysis; no code leaves the CI runner.