diff --git a/eval/grade-swebench.mjs b/eval/grade-swebench.mjs
index d1b4008..230c65f 100755
--- a/eval/grade-swebench.mjs
+++ b/eval/grade-swebench.mjs
@@ -23,7 +23,8 @@ if (!/^[a-zA-Z0-9_-]+$/.test(id ?? "") || !dataset || !process.env.EVAL_PATCH_PA
   process.exit(2);
 }
 const contents = readFileSync(dataset);
-const digestFile = ["django__django-14631", "pytest-dev__pytest-5787"].includes(id) ? "swebench-astra.sha256" :
+const digestFile = ["sympy__sympy-13878", "sphinx-doc__sphinx-7590", "scikit-learn__scikit-learn-25102", "sphinx-doc__sphinx-11510", "pytest-dev__pytest-6197"].includes(id) ? "swebench-boundary.sha256" :
+  ["django__django-14631", "pytest-dev__pytest-5787"].includes(id) ? "swebench-astra.sha256" :
   id.startsWith("astropy__astropy-") ? "swebench-candidates.sha256" :
   id === "pydata__xarray-6992" ? "swebench-xarray.sha256" : "swebench-verified-pilot.sha256";
 const expected = readFileSync(process.env.EVAL_DATASET_DIGEST_FILE ?? join(root, digestFile), "utf8").trim();
diff --git a/eval/results/astra-boundary-2026-09-24.md b/eval/results/astra-boundary-2026-09-24.md
new file mode 100644
index 0000000..b5c669e
--- /dev/null
+++ b/eval/results/astra-boundary-2026-09-24.md
@@ -0,0 +1,151 @@
+# Astra effort-boundary search
+
+## Outcome
+
+The five-candidate search did **not** establish the requested reliable medium-versus-higher-effort boundary. There were 20 screening attempts and 20 fresh confirmation attempts, with no replacements. All five candidates passed base-fails/reference-passes preflight. One SymPy xhigh screen timed out; the other 39 attempts were graded with valid usage evidence.
+
+The exploratory Sphinx #7590 confirmation resolved medium **0/5**, high **0/5**, xhigh **0/5**, and strict Jev **2/5**. Every Jev attempt had zero fallback events and zero terminal classification errors. The two router successes are interesting, but five attempts do not establish a router advantage, and the fixed high/xhigh baselines were not reliably successful. This is not evidence of equal-quality savings.
+
+Across all five attempts per arm, Jev used **6.2% more output tokens and 11.2% more time than high**; versus xhigh, it used **36.1% fewer output tokens and 5.1% less time**. These compare different quality outcomes, not equivalent successful solutions. Full input/cache/output means and all attempt IDs follow. The search stopped at the five-candidate budget; no further workers are queued.
+
+## Protocol
+
+### Observed Jev efforts
+
+The five Sphinx confirmation runs recorded **164 generation decisions: high 142, medium 7, low 15, xhigh 0, max 0**. There were 20 additional classification attempts and zero fallback events. These counts come from the saved decision logs, not the provider's reported response effort.
+
+For comparison, the five strict pytest #5787 runs in the preceding PR recorded high 84, medium 33, low 8, xhigh 0, max 0. Across these ten fallback-free Jev runs, **Jev never selected xhigh**. The fixed-xhigh screening and confirmation arms were explicit baselines, not Jev selections. This search therefore did not demonstrate router escalation to xhigh.
+
+Run set: `astra-boundary-2026-09-24`. Dataset revision: `78f471bf655a3137b2e8a75af1501690ec009ec3`. Candidate order: sympy__sympy-13878, sphinx-doc__sphinx-7590, scikit-learn__scikit-learn-25102, sphinx-doc__sphinx-11510, pytest-dev__pytest-6197.
+
+Predeclared screen: medium 0/2 and xhigh 2/2, with all four runs graded and evidence-valid. Qualifying cases receive five fresh attempts per arm. Confirmation: medium at most 1/5, and high or xhigh at least 4/5, with all 20 runs graded and evidence-valid. Stop after the first confirmed fixed-effort boundary, independent of whether Jev wins, or after five candidates. Failed preflight consumes a candidate slot.
+
+At most two concurrent attempts, 15-minute agent cap. Screen orders: medium/xhigh, xhigh/medium. Full orders rotate across rounds. Strict Jev: three additional retries, 10-second total deadline, fallback disabled. No replacements; screen attempts are excluded from confirmation. Independent base/reference preflight is required. Raw prompts, patches and logs remain ignored.
+
+After all five screens failed the strict qualification rule, Sphinx #7590 (medium 0/2, xhigh 1/2) was selected for an exploratory full matrix as the strongest observed fixed-effort contrast. This is a disclosed post-screen extension on an existing candidate, not a sixth candidate or a qualified screen. The full confirmation threshold is unchanged.
+
+## sympy__sympy-13878
+
+Status: screen did not qualify
+
+### screen
+
+| Arm | Solved / initiated | Graded | Evidence valid | Mean seconds | Mean input | Mean cached input | Mean output |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| medium | 1/2 | 2 | 2 | 126.6 | 128789.0 | 112192.0 | 2620.5 |
+| xhigh | 1/2 | 1 | 1 | 874.6 | 1144966.0 | 1075968.0 | 9356.0 |
+
+| Run ID | Solved | Evidence valid | Timeout | Classification errors | Fallbacks |
+| --- | --- | --- | --- | ---: | ---: |
+| sympy__sympy-13878-gpt-6-astra-medium-1f04df1e-c41c-4a69-b7e2-eb279ae09802 | False | True | False | 0 | 0 |
+| sympy__sympy-13878-gpt-6-astra-xhigh-f37eb008-c39c-4ae9-9906-87cd4ad7a0c0 | None | False | True | 0 | 0 |
+| sympy__sympy-13878-gpt-6-astra-xhigh-c49ceb50-2bf1-4fb4-a653-4d7f48da26ee | True | True | False | 0 | 0 |
+| sympy__sympy-13878-gpt-6-astra-medium-a2fc61b0-e28f-400d-94b6-6e8c04124fac | True | True | False | 0 | 0 |
+
+## sphinx-doc__sphinx-7590
+
+Status: exploratory full matrix: inconclusive
+
+### screen
+
+| Arm | Solved / initiated | Graded | Evidence valid | Mean seconds | Mean input | Mean cached input | Mean output |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| medium | 0/2 | 2 | 2 | 99.5 | 138819.5 | 120832.0 | 1655.5 |
+| xhigh | 1/2 | 2 | 2 | 284.5 | 628557.5 | 572032.0 | 6852.0 |
+
+| Run ID | Solved | Evidence valid | Timeout | Classification errors | Fallbacks |
+| --- | --- | --- | --- | ---: | ---: |
+| sphinx-doc__sphinx-7590-gpt-6-astra-medium-6a36ed40-0e18-4d96-ab1e-4fe67ab8f75a | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-xhigh-e93ce56d-f313-4f5c-8cc2-3cdb74175cf2 | True | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-xhigh-c98e12d1-0c44-47c2-9fff-cb8574b23d21 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-medium-abbbc0c4-7862-45fa-a0b8-222140b36586 | False | True | False | 0 | 0 |
+
+### full
+
+| Arm | Solved / initiated | Graded | Evidence valid | Mean seconds | Mean input | Mean cached input | Mean output |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| medium | 0/5 | 5 | 5 | 99.4 | 112365.2 | 94336.0 | 1550.4 |
+| high | 0/5 | 5 | 5 | 271.7 | 600586.8 | 556467.2 | 4090.8 |
+| xhigh | 0/5 | 5 | 5 | 318.3 | 822392.8 | 759219.2 | 6804.2 |
+| jev | 2/5 | 5 | 5 | 302.1 | 631532.4 | 583910.4 | 4344.8 |
+
+| Run ID | Solved | Evidence valid | Timeout | Classification errors | Fallbacks |
+| --- | --- | --- | --- | ---: | ---: |
+| sphinx-doc__sphinx-7590-gpt-6-astra-medium-01236bdc-ff8c-4a41-b419-3c3e794a4449 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-high-4bb082c1-e364-4ed0-ac01-e58d32e6fcde | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-xhigh-20dd916f-70d8-4d61-b4bc-dccebc41feb8 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-jev-79c53a0c-9458-46ce-b6e6-c2450afd195b | True | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-jev-3826138c-a89c-4210-8f41-dfd364de6be4 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-xhigh-fa5f80e9-9bc7-411c-b3f6-fbe07440dd87 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-high-50c446fd-4f10-49ba-9fa4-e39e12ce745d | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-medium-29fcedb2-d4f0-47bc-b739-0db96ad1f638 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-high-390c670e-823d-4219-a3fd-a39ea97245b9 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-jev-96fe0d97-4be3-4cb3-b239-8d0f9dd8dc69 | True | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-medium-b29e2aad-636b-4224-8519-b60cc3637496 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-xhigh-e8da86ec-93d1-4cb1-993d-2724bdcbf35f | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-xhigh-e178a7af-8e3a-48c5-82ed-2db46128a6a8 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-medium-0c4f217f-02a4-4a64-beb5-2fea371a5f5a | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-jev-8696da9e-2fce-4ca0-9ed0-ec24418ebb8f | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-high-d8087def-c9a1-4904-9596-c2910dde6e22 | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-medium-0941ae27-fd84-47ac-91a5-628ec8727c2b | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-jev-02cb6ed7-e0da-4aed-a580-3801c58b1eac | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-high-a603b243-ea5b-4246-bc73-471ba124188e | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-7590-gpt-6-astra-xhigh-b60b37a7-f280-4e91-983f-b499ce5c9b78 | False | True | False | 0 | 0 |
+
+## scikit-learn__scikit-learn-25102
+
+Status: screen did not qualify
+
+### screen
+
+| Arm | Solved / initiated | Graded | Evidence valid | Mean seconds | Mean input | Mean cached input | Mean output |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| medium | 2/2 | 2 | 2 | 84.2 | 137409.0 | 115712.0 | 1357.5 |
+| xhigh | 1/2 | 2 | 2 | 249.9 | 430986.5 | 377472.0 | 4077.5 |
+
+| Run ID | Solved | Evidence valid | Timeout | Classification errors | Fallbacks |
+| --- | --- | --- | --- | ---: | ---: |
+| scikit-learn__scikit-learn-25102-gpt-6-astra-medium-20a11f78-7ee2-4bce-bf3c-ac49a7164b90 | True | True | False | 0 | 0 |
+| scikit-learn__scikit-learn-25102-gpt-6-astra-xhigh-033be741-d719-4812-aaf3-cae87eba0da0 | True | True | False | 0 | 0 |
+| scikit-learn__scikit-learn-25102-gpt-6-astra-xhigh-32effe56-f6ab-434d-81aa-6aeabf201b4b | False | True | False | 0 | 0 |
+| scikit-learn__scikit-learn-25102-gpt-6-astra-medium-a6e4b066-415c-4983-b61e-62ae54589201 | True | True | False | 0 | 0 |
+
+## sphinx-doc__sphinx-11510
+
+Status: screen did not qualify
+
+### screen
+
+| Arm | Solved / initiated | Graded | Evidence valid | Mean seconds | Mean input | Mean cached input | Mean output |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| medium | 0/2 | 2 | 2 | 97.5 | 149084.5 | 119232.0 | 1694.0 |
+| xhigh | 0/2 | 2 | 2 | 65.0 | 113180.5 | 83840.0 | 1082.0 |
+
+| Run ID | Solved | Evidence valid | Timeout | Classification errors | Fallbacks |
+| --- | --- | --- | --- | ---: | ---: |
+| sphinx-doc__sphinx-11510-gpt-6-astra-medium-3afedc7b-a848-4bc5-8f12-4a158e5ec39f | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-11510-gpt-6-astra-xhigh-8adf5760-31e9-4008-99a2-de3b4ef85c7b | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-11510-gpt-6-astra-xhigh-219f39c6-fac5-4e6a-9f61-9762f9e41eff | False | True | False | 0 | 0 |
+| sphinx-doc__sphinx-11510-gpt-6-astra-medium-52156959-21bc-4b48-80bc-ef37d22f3acc | False | True | False | 0 | 0 |
+
+## pytest-dev__pytest-6197
+
+Status: screen did not qualify
+
+### screen
+
+| Arm | Solved / initiated | Graded | Evidence valid | Mean seconds | Mean input | Mean cached input | Mean output |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| medium | 2/2 | 2 | 2 | 116.0 | 160802.0 | 139136.0 | 1099.5 |
+| xhigh | 2/2 | 2 | 2 | 352.4 | 946642.0 | 893312.0 | 5952.5 |
+
+| Run ID | Solved | Evidence valid | Timeout | Classification errors | Fallbacks |
+| --- | --- | --- | --- | ---: | ---: |
+| pytest-dev__pytest-6197-gpt-6-astra-medium-88f43744-08a0-4c0d-9209-7ff2974e1b1f | True | True | False | 0 | 0 |
+| pytest-dev__pytest-6197-gpt-6-astra-xhigh-0dbccb6b-84ea-4f35-900f-207b7b360ec5 | True | True | False | 0 | 0 |
+| pytest-dev__pytest-6197-gpt-6-astra-xhigh-40c3e16a-29fd-401e-b53a-2a87190cec48 | True | True | False | 0 | 0 |
+| pytest-dev__pytest-6197-gpt-6-astra-medium-85b1ff17-1fea-4e5f-87c7-c0c8751dd7ec | True | True | False | 0 | 0 |
+
+## Interpretation limits
+
+This deliberately selects effort-sensitive cases and cannot estimate broad workload savings. Five attempts per arm provide limited precision. Token means require valid evidence; elapsed time includes initiated failures. Cached input is part of input, not additional consumption. Output includes reasoning and auxiliary calls. Concurrent execution and shared upstream/host activity affect latency. A classification failure is an incomplete delivery, not a wrong patch.
diff --git a/eval/search-astra-boundary.py b/eval/search-astra-boundary.py
new file mode 100644
index 0000000..f3f3619
--- /dev/null
+++ b/eval/search-astra-boundary.py
@@ -0,0 +1,137 @@
+"""Bounded, predeclared search; prepare once, then run without replacing failures.
+
+python eval/search-astra-boundary.py prepare|run UNIQUE_RUN_SET
+Requires Node 24, built plugin, Docker and canonical eval credentials.
+"""
+import concurrent.futures
+import hashlib
+import json
+import os
+from pathlib import Path
+import statistics
+import subprocess
+import sys
+
+ROOT = Path(__file__).resolve().parent
+IDS = ['sympy__sympy-13878', 'sphinx-doc__sphinx-7590', 'scikit-learn__scikit-learn-25102', 'sphinx-doc__sphinx-11510', 'pytest-dev__pytest-6197']
+REVISION = '78f471bf655a3137b2e8a75af1501690ec009ec3'
+mode, name = sys.argv[1:]
+assert mode in ['prepare', 'run', 'supplementary'] and name and all(c.isalnum() or c in '-_' for c in name)
+directory = ROOT / 'runs' / name
+if mode == 'prepare':
+    from datasets import load_dataset
+    directory.mkdir()
+    source = load_dataset('SWE-bench/SWE-bench_Verified', split='test', revision=REVISION)
+    rows = [next(r for r in source if r['instance_id'] == id_) for id_ in IDS]
+    contents = json.dumps(rows, ensure_ascii=False) + '\n'
+    (directory / 'dataset.json').write_text(contents)
+    (ROOT / 'swebench-boundary.sha256').write_text(hashlib.sha256(contents.encode()).hexdigest() + '\n')
+    tasks = [{'id': r['instance_id'], 'repo': f"https://github.com/{r['repo']}.git", 'commit': r['base_commit'], 'grade': 'swebench', 'agentTimeoutMinutes': 15, 'prompt': r['problem_statement'].replace('\r\n', '\n')} for r in rows]
+    (directory / 'tasks.json').write_text(json.dumps({'tasks': tasks}))
+    print('Prepared pinned candidates:', ', '.join(IDS))
+    sys.exit(0)
+
+assert os.environ.get('JEV_ROUTER_API_KEY') and os.environ.get('CLIPROXY_KEY')
+with (directory / ('supplementary-started' if mode == 'supplementary' else 'started')).open('x') as file:
+    file.write('Do not restart: initiated attempts must not be replaced.\n')
+rows = json.loads((directory / 'dataset.json').read_text())
+env = {**os.environ, 'SWE_BENCH_DATASET_PATH': str(directory / 'dataset.json'), 'SWE_BENCH_PYTHON': sys.executable}
+outcomes = json.loads((directory / 'outcomes.json').read_text()) if mode == 'supplementary' else []
+
+def attempt(task, phase, round_, arm):
+    run_set = f'{name}-{task}-{phase}-r{round_}'
+    logpath = directory / f'{task}-{phase}-r{round_}-{arm}.log'
+    print(f'START {task} {phase} r{round_} {arm}', flush=True)
+    with logpath.open('x') as log:
+        completed = subprocess.run(['node', 'eval/run.mjs', '--manifest', str(directory / 'tasks.json'), '--task', task, '--model', 'gpt-6-astra', '--arm', arm], cwd=ROOT.parent, env={**env, 'EVAL_RUN_SET': run_set}, stdout=log, stderr=subprocess.STDOUT)
+    paths = []
+    for path in (ROOT / 'runs').glob(f'{task}-gpt-6-astra-{arm}-*/result.json'):
+        result = json.loads(path.read_text())
+        if result.get('run_set') == run_set:
+            paths.append((path, result))
+    assert len(paths) == 1, f'Incomplete/missing runner result: {logpath}'
+    path, result = paths[0]
+    result['runner_exit'] = completed.returncode
+    if arm == 'jev':
+        config = json.loads((path.parent / 'opencode.json').read_text())['plugin'][0][1]
+        assert (config['maxRetries'], config['fallbackMode'], config['jevTimeoutMs']) == (3, 'error', 10000)
+    print(f"DONE {task} {phase} r{round_} {arm} solved={result['grade_passed']} evidence={result.get('evidence_valid')}", flush=True)
+    return result
+
+def matrix(task, phase, orders):
+    with concurrent.futures.ThreadPoolExecutor(max_workers=2) as pool:
+        futures = [pool.submit(attempt, task, phase, i+1, arm) for i, order in enumerate(orders) for arm in order.split()]
+        return [f.result() for f in futures]
+
+def report():
+    (directory / 'outcomes.json').write_text(json.dumps(outcomes, indent=2) + '\n')
+    lines = ['# Astra effort-boundary search', '', f'Run set: `{name}`. Dataset revision: `{REVISION}`. Candidate order: ' + ', '.join(IDS) + '.', '',
+             'Predeclared screen: medium 0/2 and xhigh 2/2, with all four runs graded and evidence-valid. Qualifying cases receive five fresh attempts per arm. Confirmation: medium at most 1/5, and high or xhigh at least 4/5, with all 20 runs graded and evidence-valid. Stop after the first confirmed fixed-effort boundary, independent of whether Jev wins, or after five candidates. Failed preflight consumes a candidate slot.', '',
+             'At most two concurrent attempts, 15-minute agent cap. Screen orders: medium/xhigh, xhigh/medium. Full orders rotate across rounds. Strict Jev: three additional retries, 10-second total deadline, fallback disabled. No replacements; screen attempts are excluded from confirmation. Independent base/reference preflight is required. Raw prompts, patches and logs remain ignored.', '']
+    if mode == 'supplementary':
+        lines += ['After all five screens failed the strict qualification rule, Sphinx #7590 (medium 0/2, xhigh 1/2) was selected for an exploratory full matrix as the strongest observed fixed-effort contrast. This is a disclosed post-screen extension on an existing candidate, not a sixth candidate or a qualified screen. The full confirmation threshold is unchanged.', '']
+    for outcome in outcomes:
+        lines += [f"## {outcome['task']}", '', f"Status: {outcome['status']}", '']
+        for phase in ['screen', 'full']:
+            results = outcome.get(phase, [])
+            if not results:
+                continue
+            lines += [f'### {phase}', '', '| Arm | Solved / initiated | Graded | Evidence valid | Mean seconds | Mean input | Mean cached input | Mean output |', '| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |']
+            for arm in ['medium', 'high', 'xhigh', 'jev']:
+                group = [r for r in results if r['arm'] == arm]
+                if not group:
+                    continue
+                def mean(field):
+                    values = [r[field] for r in group if r.get(field) is not None and (field == 'elapsed_ms' or r.get('evidence_valid'))]
+                    return round(statistics.mean(values)/(1000 if field == 'elapsed_ms' else 1), 1) if values else '—'
+                lines.append('| ' + ' | '.join(str(x) for x in [arm, f"{sum(r['grade_passed'] is True for r in group)}/{len(group)}", sum(r['grade_passed'] is not None for r in group), sum(bool(r.get('evidence_valid')) for r in group), mean('elapsed_ms'), mean('input_tokens'), mean('cached_input_tokens'), mean('output_tokens')]) + ' |')
+            lines += ['', '| Run ID | Solved | Evidence valid | Timeout | Classification errors | Fallbacks |', '| --- | --- | --- | --- | ---: | ---: |']
+            for r in results:
+                lines.append('| ' + ' | '.join(str(r.get(k)) for k in ['run_id', 'grade_passed', 'evidence_valid', 'timed_out', 'classification_errors', 'fallbacks']) + ' |')
+            lines.append('')
+    lines += ['## Interpretation limits', '', 'This deliberately selects effort-sensitive cases and cannot estimate broad workload savings. Five attempts per arm provide limited precision. Token means require valid evidence; elapsed time includes initiated failures. Cached input is part of input, not additional consumption. Output includes reasoning and auxiliary calls. Concurrent execution and shared upstream/host activity affect latency. A classification failure is an incomplete delivery, not a wrong patch.']
+    (ROOT / 'results' / f'{name}.md').write_text('\n'.join(lines) + '\n')
+
+for row in rows:
+    task = row['instance_id']
+    if mode == 'supplementary':
+        if task != 'sphinx-doc__sphinx-7590':
+            continue
+        outcome = next(o for o in outcomes if o['task'] == task)
+        full = outcome['full'] = matrix(task, 'full', ['medium high xhigh jev', 'jev xhigh high medium', 'high jev medium xhigh', 'xhigh medium jev high', 'medium jev high xhigh'])
+        assert all(not r.get('fallbacks') for r in full if r['arm'] == 'jev')
+        counts = {arm: sum(r['grade_passed'] is True for r in full if r['arm'] == arm) for arm in ['medium', 'high', 'xhigh']}
+        confirmed = all(r['grade_passed'] is not None and r.get('evidence_valid') for r in full) and counts['medium'] <= 1 and max(counts['high'], counts['xhigh']) >= 4
+        outcome['status'] = 'exploratory full matrix: confirmed fixed-effort boundary' if confirmed else 'exploratory full matrix: inconclusive'
+        report()
+        continue
+    outcome = {'task': task, 'status': 'preflight'}
+    outcomes.append(outcome)
+    ok = True
+    for label, patch, expected in [('base', 'diff --git a/jev-preflight-marker.txt b/jev-preflight-marker.txt\nnew file mode 100644\n--- /dev/null\n+++ b/jev-preflight-marker.txt\n@@ -0,0 +1 @@\n+baseline grader check\n', 1), ('gold', row['patch'], 0)]:
+        target = directory / task / label
+        target.mkdir(parents=True)
+        (target / 'patch.diff').write_text(patch)
+        with (target / 'grader.log').open('w') as log:
+            grade = subprocess.run(['node', str(ROOT / 'grade-swebench.mjs'), task], cwd=target, env={**env, 'EVAL_TASK_COMMIT': row['base_commit'], 'EVAL_PATCH_PATH': str(target / 'patch.diff')}, stdout=log, stderr=subprocess.STDOUT)
+        ok = ok and grade.returncode == expected
+        print(f'PREFLIGHT {task} {label} exit={grade.returncode}', flush=True)
+    if not ok:
+        outcome['status'] = 'preflight failed; no agent attempts'
+        report()
+        continue
+    screen = outcome['screen'] = matrix(task, 'screen', ['medium xhigh', 'xhigh medium'])
+    qualifies = all(r['grade_passed'] is not None and r.get('evidence_valid') for r in screen) and all(r['grade_passed'] == (r['arm'] == 'xhigh') for r in screen)
+    outcome['status'] = 'screen did not qualify'
+    report()
+    if not qualifies:
+        continue
+    full = outcome['full'] = matrix(task, 'full', ['medium high xhigh jev', 'jev xhigh high medium', 'high jev medium xhigh', 'xhigh medium jev high', 'medium jev high xhigh'])
+    assert all(not r.get('fallbacks') for r in full if r['arm'] == 'jev')
+    counts = {arm: sum(r['grade_passed'] is True for r in full if r['arm'] == arm) for arm in ['medium', 'high', 'xhigh']}
+    confirmed = all(r['grade_passed'] is not None and r.get('evidence_valid') for r in full) and counts['medium'] <= 1 and max(counts['high'], counts['xhigh']) >= 4
+    outcome['status'] = 'confirmed fixed-effort boundary' if confirmed else 'full confirmation inconclusive'
+    report()
+    if confirmed:
+        break
+print('COMPLETE', flush=True)
diff --git a/eval/swebench-boundary.sha256 b/eval/swebench-boundary.sha256
new file mode 100644
index 0000000..6490a7e
--- /dev/null
+++ b/eval/swebench-boundary.sha256
@@ -0,0 +1 @@
+08858dfb08717e633a802e3ff7613eaa3d6f206b000f26b311b40521064d03ee