Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
cases: list<item: struct<case_id: string, class: string, expected: list<item: struct<class: string, edge: s (... 224 chars omitted)
child 0, item: struct<case_id: string, class: string, expected: list<item: struct<class: string, edge: string>>, fo (... 212 chars omitted)
child 0, case_id: string
child 1, class: string
child 2, expected: list<item: struct<class: string, edge: string>>
child 0, item: struct<class: string, edge: string>
child 0, class: string
child 1, edge: string
child 3, formats: struct<descriptor: struct<expressible: bool, path: string, reason: string>, rfc8693: struct<expressi (... 104 chars omitted)
child 0, descriptor: struct<expressible: bool, path: string, reason: string>
child 0, expressible: bool
child 1, path: string
child 2, reason: string
child 1, rfc8693: struct<expressible: bool, reason: string, path: string>
child 0, expressible: bool
child 1, reason: string
child 2, path: string
child 2, ucan: struct<expressible: bool, reason: string, path: string>
child 0, expressible: bool
child 1, reason: string
child 2, path: string
count: int64
counts: struct<descriptor: struct<expressible: int64, inexpressible: int64>, rfc8693: struct<expressible: in (... 83 chars omitted)
child 0, descriptor: struct<expressible: int64, inexpressible: int64>
child 0, expressible: int64
child 1, inexpressible: int64
child 1, rfc8693: struct<expressible: int64, inexpressible: int64>
child 0, expressible: int64
child 1, inexpressible: int64
child 2, ucan: struct<expressible: int64, inexpressible: int64>
child 0, expressible: int64
child 1, inexpressible: int64
formats: list<item: string>
child 0, item: string
selected_case_ids: list<item: string>
child 0, item: string
version: string
framework: string
scenario: string
source_refs: list<item: null>
child 0, item: null
generations: list<item: struct<actions: list<item: string>, caps: list<item: struct<limit: int64, name: string>>, (... 103 chars omitted)
child 0, item: struct<actions: list<item: string>, caps: list<item: struct<limit: int64, name: string>>, node: stri (... 91 chars omitted)
child 0, actions: list<item: string>
child 0, item: string
child 1, caps: list<item: struct<limit: int64, name: string>>
child 0, item: struct<limit: int64, name: string>
child 0, limit: int64
child 1, name: string
child 2, node: string
child 3, purposes: list<item: string>
child 0, item: string
child 4, resources: list<item: string>
child 0, item: string
child 5, tools: list<item: string>
child 0, item: string
child_credentials: string
handoff_evidence: string
to
{'child_credentials': Value('string'), 'framework': Value('string'), 'generations': List({'actions': List(Value('string')), 'caps': List({'limit': Value('int64'), 'name': Value('string')}), 'node': Value('string'), 'purposes': List(Value('string')), 'resources': List(Value('string')), 'tools': List(Value('string'))}), 'handoff_evidence': Value('string'), 'scenario': Value('string'), 'source_refs': List(Value('null')), 'version': Value('string')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
cases: list<item: struct<case_id: string, class: string, expected: list<item: struct<class: string, edge: s (... 224 chars omitted)
child 0, item: struct<case_id: string, class: string, expected: list<item: struct<class: string, edge: string>>, fo (... 212 chars omitted)
child 0, case_id: string
child 1, class: string
child 2, expected: list<item: struct<class: string, edge: string>>
child 0, item: struct<class: string, edge: string>
child 0, class: string
child 1, edge: string
child 3, formats: struct<descriptor: struct<expressible: bool, path: string, reason: string>, rfc8693: struct<expressi (... 104 chars omitted)
child 0, descriptor: struct<expressible: bool, path: string, reason: string>
child 0, expressible: bool
child 1, path: string
child 2, reason: string
child 1, rfc8693: struct<expressible: bool, reason: string, path: string>
child 0, expressible: bool
child 1, reason: string
child 2, path: string
child 2, ucan: struct<expressible: bool, reason: string, path: string>
child 0, expressible: bool
child 1, reason: string
child 2, path: string
count: int64
counts: struct<descriptor: struct<expressible: int64, inexpressible: int64>, rfc8693: struct<expressible: in (... 83 chars omitted)
child 0, descriptor: struct<expressible: int64, inexpressible: int64>
child 0, expressible: int64
child 1, inexpressible: int64
child 1, rfc8693: struct<expressible: int64, inexpressible: int64>
child 0, expressible: int64
child 1, inexpressible: int64
child 2, ucan: struct<expressible: int64, inexpressible: int64>
child 0, expressible: int64
child 1, inexpressible: int64
formats: list<item: string>
child 0, item: string
selected_case_ids: list<item: string>
child 0, item: string
version: string
framework: string
scenario: string
source_refs: list<item: null>
child 0, item: null
generations: list<item: struct<actions: list<item: string>, caps: list<item: struct<limit: int64, name: string>>, (... 103 chars omitted)
child 0, item: struct<actions: list<item: string>, caps: list<item: struct<limit: int64, name: string>>, node: stri (... 91 chars omitted)
child 0, actions: list<item: string>
child 0, item: string
child 1, caps: list<item: struct<limit: int64, name: string>>
child 0, item: struct<limit: int64, name: string>
child 0, limit: int64
child 1, name: string
child 2, node: string
child 3, purposes: list<item: string>
child 0, item: string
child 4, resources: list<item: string>
child 0, item: string
child 5, tools: list<item: string>
child 0, item: string
child_credentials: string
handoff_evidence: string
to
{'child_credentials': Value('string'), 'framework': Value('string'), 'generations': List({'actions': List(Value('string')), 'caps': List({'limit': Value('int64'), 'name': Value('string')}), 'node': Value('string'), 'purposes': List(Value('string')), 'resources': List(Value('string')), 'tools': List(Value('string'))}), 'handoff_evidence': Value('string'), 'scenario': Value('string'), 'source_refs': List(Value('null')), 'version': Value('string')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
SubGrant Benchmark (dev split)
A benchmark of agent delegation chains for evaluating verifiers that check whether authority only ever narrows when one agent delegates to another. This is the public dev split of the SubGrant benchmark (120 cases). SubGrant itself is an offline, format-neutral, deterministic verifier — see the SubGrant repository for the code and reports.
Files
cases.jsonl— the 120 dev cases. Each line is a JSON object with acase_id, agraph(adgraph/1delegation graph), and anexpectedlist of findings{class, edge}(empty for clean cases).dev_ext/— 24 of the dev cases re-expressed in three source formats (ucan/,rfc8693/,descriptor/) plusmanifest.json, which records, per (case, format) pair, whether the format can express the case and — when it cannot — the reason.
Composition (dev, 120 cases)
| Class | Count | Meaning |
|---|---|---|
| D1 | 8 | amplification (child holds more than parent) |
| D2 | 6 | depth breach |
| D3 | 6 | re-delegation breach |
| D4 | 6 | lifetime (window) breach |
| D5 | 8 | fan-out recombination |
| D6 | 8 | caveat drop / weakening |
| D7 | 6 | confused-deputy reachability |
| D8 | 4 | unbound delegation (presence-only) |
| clean | 68 | valid chains that must raise no finding |
| total | 120 |
Cases are generated deterministically (no RNG); the case-file digest is anchored in the source repository's manifest.
Scoring rule
A planted case is counted detected only when a finding matches on both the check class and the edge. Clean cases must raise no D1–D7 finding. Extra findings on planted cases are counted separately. This is the rule used for the headline numbers below.
Headline numbers (audited)
- Dev bench (this split): recall D1–D7 = 1.000 each; clean false positives 0/68; D8 4/4 (GATE 1 PASS).
- Baselines on dev: a naive per-edge checker scores 0.000 on D5 and D7 (and 0.167 on D2); SubGrant scores 1.000 across D1–D7.
- Framework-default audit (8 frameworks, 2026-11-11): none checks child ⊆ parent by default, and none records a delegation-authority binding on the handoff.
Sealed holdout — not included
A second, sealed 120-case holdout was authored independently, with the verifier's predictions pre-registered in git before the holdout was written. SubGrant passed GATE 2 on it (recall D1–D7 = 1.000, clean FP 0/68, D8 4/4, 0 extra findings). The holdout is not published and is not part of this dataset — only this dev split and its aggregate holdout result are public.
License & contact
All Rights Reserved (license: other). These materials are made public
solely so they can be read and evaluated; no license to use, copy, modify, or
create derivative works is granted. Licensing or transfer inquiries:
https://nagayu-portfolio.nagayu.workers.dev/en
- Downloads last month
- 104