fix(serve): suppress query truncation banner when no nodes were cut (#2601)

[ousamabenyounes]

Why

Fix #2601

query printed the truncation banner even when zero whole nodes were cut. _subgraph_to_text renders every NODE line before any EDGE line, so a char-budget overflow whose cut lands past the last node drops only trailing edges — cut_count is 0, yet the banner still fired:

[!] TRUNCATED: showing 91 of 91 nodes (~6000-token budget). The answer may be among the 0 cut nodes ...

The banner's audience is usually an agent, and this false warning is actively harmful: "the answer may be among the 0 cut nodes" teaches the reader to distrust a complete answer, and an agent that follows the banner's own advice issues follow-up context_filter / get_node calls to recover nodes that were never cut — pure wasted round-trips.

What changed

graphify/serve.py — in _subgraph_to_text, when cut_count == 0 return the full output (the complete node set plus all its edges) with no banner. All nodes are shown and every edge connects shown nodes, so it is a complete answer and renders as one — no misleading notice, and no silently-dropped edges (which would repeat the #BUG2 "silence reads as absence" mistake). When a whole node is genuinely cut (cut_count > 0) the banner still fires exactly as before.

The sibling _cut_lines_to_budget is line-oriented and cannot produce cut_count == 0 (an over-budget cut always drops at least one line), so it is intentionally left untouched.

Test verification (RED → GREEN)

New test test_subgraph_to_text_no_banner_when_only_edges_overflow builds a 4-node K4 and picks token_budget=60 so every NODE line fits but the trailing EDGE lines overflow. It asserts all 4 nodes are shown, no TRUNCATED/cut nodes banner, and all 6 edges survive (so a banner-suppressed-but-edge-truncated implementation would still fail).

RED — unmodified v8, new test only:

>       assert "TRUNCATED" not in text and "truncated" not in text
E       AssertionError: assert ('TRUNCATED' not in '[!] TRUNCAT...ific symbol)'
E         'TRUNCATED' is contained here:
E           [!] TRUNCATED: showing 4 of 4 nodes (~60-token budget). The answer may be among the 0 cut nodes ...
FAILED tests/test_serve.py::test_subgraph_to_text_no_banner_when_only_edges_overflow
1 failed

GREEN — with the fix:

tests/test_serve.py::test_subgraph_to_text_no_banner_when_only_edges_overflow PASSED
1 passed

Full local suite

• pytest tests/ (Python 3.12 & 3.10): 4306 passed, 3 skipped. Two failures are pre-existing on unmodified v8 and unrelated to this change (test_collect_files_skips_hidden picks up local .venv/hidden dirs in the checkout; test_property_normalize_id_agrees_with_its_own_caseless_form is a hypothesis Unicode combining-mark falsifying example). Both reproduce identically on v8 with this diff stashed — iso-baseline, no new failures.
• tests/test_serve.py: 140 passed. The existing truncation guards (test_subgraph_to_text_truncates, test_subgraph_to_text_seed_survives_truncation, test_subgraph_to_text_truncation_notice_at_top) still pass — the banner is unchanged whenever a node is actually cut.
• ruff check graphify tests clean; skillgen --check OK (no generated files touched).
• Minimal diff: one guard in serve.py, one new test, one CHANGELOG entry.

[graphify-labs]

Graphify reviewed this change.

Worth a look — the grounded gate found no coupling regressions or blocking issues, but 1 advisory finding(s) below merit a look before merge.

Formal verification. 1 change(s) tested, no difference found (not proven).

---

Graphify review — findings

This PR modifies the _subgraph_to_text function in graphify/serve.py to add an early return that skips the truncation banner when a character-budget overflow only trims trailing edges and cuts zero whole nodes. When cut_count == 0, the function now returns the full output without the "showing N of N nodes … among the 0 cut nodes" notice. A corresponding test and a CHANGELOG entry (#2601) are added. The changed-symbol list is large, but the actual diff only touches one code branch in serve.py, one new test, and one changelog line.

Worth a look

• cut_count==0 returns full output including edges past char budget — graphify/serve.py:1092 · Escalate · medium
  • agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 699 functions depend on the 496 functions this change touches.

Health — grade A; 10 existing hotspot(s) in the area this change touches (pre-existing, not introduced here):

• dispatch_command() — 2 callers, 117 callees (high)
• _query_graph_text() — 18 callers, 8 callees (high)
• _score_query() — 15 callers, 5 callees (high)
• _query_terms() — 17 callers, 3 callees (high)
• run_benchmark() — 16 callers, 3 callees (high)
• _build_server() — 2 callers, 16 callees (high)
• _load_graph() — 9 callers, 3 callees (medium)
• _shortest_path_text() — 5 callers, 4 callees (medium)
• …and 2 more

Verification — 699 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 549 function(s) in the blast radius were not formally verified this run

Formal verification

No difference found (not proven): No behavior difference found in \_subgraph\_to\_text (not a proof).

The verifier ran both versions of \_subgraph\_to\_text on many inputs and saw identical behavior every time. Strong evidence the change is safe, but evidence, not a proof.

Guarantee: Empirical: differential testing (both versions run on many generated inputs). A divergence on an untested input remains possible, so this is 'no counterexample found', not 'proven equivalent'.

Note: An input the sampler did not try could still differ.

[safishamsi]

Shipped in v0.9.41 (graphifyy==0.9.41 on PyPI) (#2601). Credited in the release notes. Thanks @ousamabenyounes!