bench(ranking): the position benchmark sets its own query ceiling; v2 partials in the brief

A pass over every region of pmpro (22,315 pairs) took 255 s without the join facts and past the scan's
300 s default with them, and a timed-out pass answers nothing -- reported once as tiers={} and the static
order. mwwpform with carried: 292 -> 30.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_01T7XD7x5KonMdmGRkp63EQG

.kiro/specs/flow-aware-ranking/brief.md

@@ -153,8 +153,14 @@ A phase that cannot show that is not an improvement, it is a story.
   this brief handed to phase 3. WP Statistics's `record()` and MW WP Form's `_delete_files()` are in
   tier 3, one below where their sources say they belong, because `carried` -- stage one of the two-stage
   join, which the arbiter computes anyway -- was not yet reported by the evidence branch. It is now
-  (`taint.sc` reports it only where a sink exists, so tier-0 pairs pay nothing); the re-measurement is
+  (`taint.sc` reports it only where a sink exists, so tier-0 pairs pay nothing).
-  in flight and is the number phase 2 stands or falls on.
+
+  **With `carried` (v2), as they land:** mwwpform **292 → 30** (CVE-2023-6559 at 29, among 37 tier-4
+  regions; evidence pass 68 → 84 s) -- the first repository under the gate. vampi unchanged at 16. pmpro's
+  v2 pass hit the scan's 300 s default query ceiling (v1 took 255 s; the join facts on its 1,040
+  sink-bearing pairs cost more than the 45 s of headroom) and answered nothing, which the benchmark
+  reported as `tiers={}` and the static order; the benchmark now sets its own ceiling and pmpro is
+  re-measured.
 
 - **Phase 3 — the LLM ranker.** Same evidence, ordered by a model. Gate: beats phase 2 leave-one-repository-out.
   If it does not, phase 2 ships and phase 3 does not.

benchmarks/ranking/position.py

@@ -16,6 +16,7 @@ from __future__ import annotations
 
 import argparse
 import json
+import os
 import sys
 from collections import Counter
 from pathlib import Path
@@ -48,6 +49,11 @@ def evidence_ordered_regions(root: Path, regions: list[ScanRegion], language: st
     from openultrasast.cpg.backend import resolve_cpg_backend
     from openultrasast.model.scan import _collect, _evidence_pass, _hook_callbacks, order_by_evidence
 
+    # A benchmark pass runs over EVERY region, not the scan's 500: pmpro's 22,315 pairs took 255 s without
+    # the join facts and past the scan's 300 s default with them, and a pass that times out answers
+    # nothing -- which read as "tiers={}" once and cost a run. The scan's per-push ceiling is not this
+    # measurement's; an operator who set one keeps it.
+    os.environ.setdefault("OPENULTRASAST_CPG_QUERY_TIMEOUT", "1800")
     backend = resolve_cpg_backend()
     started = time.monotonic()
     # The RECIPE's language, never the first region's: WP Statistics's first region is JavaScript, and a