ReceiptsRecorded runs, model-free verdicts

17.9 percent of an unguarded agent's 'done' was false; Receipts shipped zero.

100 SWE-rebench-V2 tasks held out from development and chosen by a rule fixed in advance. NVIDIA Nemotron 3 Super on Nebius Token Factory wrote every fix; a clean Token Factory Sandbox and plain code checked it; SWE-rebench's own grader graded it again.

Primary sample: 100 held-out tasks, Nemotron 3 Super

Unguarded agent: 'done' that was false12 of 6717.9%, 95% interval 10.6% to 28.7%
Receipts: verified fixes6659 on the first verdict, 7 after one appeal
Receipts: false fixes shipped0checked by SWE-rebench's official grader
Model cost per verified fix$0.41Nemotron tokens at Token Factory's listed prices

Verified fixes require a gate PASS and official resolution. Run verdicts show the gate result; official grades are listed separately.

Token Factory reported sandbox cost over all runs: 63.741 (unit not published).

How to read this: Unguarded agent = the agent's own 'done' on its first try. Receipts = the same agent behind a witness sandbox and a model-free gate, with one appeal. Official = SWE-rebench's grader, run separately.

InstanceUnguarded: claimed doneUnguarded: officialReceipts outcomeReceipts: officialRunReceipt
rhayes777__pyautofit-888yesresolvedVERIFIED on first verdictresolvedRunReceipt
deshaw__pyflyby-289yesresolvedVERIFIED on first verdictresolvedRunReceipt
azure__walinuxagent-638yesnot resolvedREJECTEDnot resolvedRunReceipt
3yourmind__django-migration-linter-258yesresolvedVERIFIED on first verdictresolvedRunReceipt
openmdao__openmdao-3029nonot resolvedREJECTEDnot resolvedRunReceipt
pyfilesystem__pyfilesystem2-462yesnot resolvedVERIFIED on appealresolvedRunReceipt
pytask-dev__pytask-519yesresolvedVERIFIED on first verdictresolvedRunReceipt
niconoe__pyinaturalist-72yesresolvedVERIFIED on first verdictresolvedRunReceipt
robotframework__seleniumlibrary-1431yesresolvedVERIFIED on first verdictresolvedRunReceipt
palantir__python-language-server-829nonot resolvedREJECTEDnot resolvedRunReceipt
stac-utils__stac-fastapi-729nonot resolvedREJECTEDnot resolvedRunReceipt
gitpython-developers__gitpython-1832nonot resolvedREJECTEDnot resolvedRunReceipt
scitools__cartopy-2307nonot resolvedREJECTEDnot resolvedRunReceipt
kubernetes-client__python-1933nonot resolvedREJECTEDnot resolvedRunReceipt
datadog__datadogpy-555yesresolvedVERIFIED on first verdictresolvedRunReceipt
callowayproject__bump-my-version-180yesresolvedVERIFIED on first verdictresolvedRunReceipt
opensearch-project__opensearch-py-843yesresolvedVERIFIED on first verdictresolvedRunReceipt
stac-utils__pystac-client-782yesresolvedVERIFIED on first verdictresolvedRunReceipt
sdv-dev__rdt-704yesresolvedVERIFIED on first verdictresolvedRunReceipt
pybamm-team__pybamm-4196yesresolvedVERIFIED on first verdictresolvedRunReceipt
aspp__pelita-798yesresolvedVERIFIED on first verdictresolvedRunReceipt
zopefoundation__btrees-163noresolvedVERIFIED on first verdictresolvedRunReceipt
skypyproject__skypy-573yesresolvedVERIFIED on first verdictresolvedRunReceipt
jnothman__upsetplot-253nonot resolvedREJECTEDnot resolvedRunReceipt
dask__dask-5861yesresolvedVERIFIED on first verdictresolvedRunReceipt
qutech__qupulse-797yesresolvedVERIFIED on first verdictresolvedRunReceipt
bloomberg__memray-552nonot resolvedREJECTEDnot resolvedRunReceipt
mondeja__mkdocs-include-markdown-plugin-51nonot resolvedREJECTEDnot resolvedRunReceipt
more-itertools__more-itertools-892yesresolvedVERIFIED on first verdictresolvedRunReceipt
yukinarit__pyserde-430yesresolvedVERIFIED on first verdictresolvedRunReceipt
pycqa__pyflakes-335yesresolvedVERIFIED on first verdictresolvedRunReceipt
azuread__microsoft-authentication-library-for-python-186noresolvedVERIFIED on first verdictresolvedRunReceipt
natcap__invest-787not recordednot recordedinfrastructure errornot recordedRunReceipt
igorbenav__fastcrud-69nonot resolvedREJECTEDnot resolvedRunReceipt
researchobject__ro-crate-py-192yesresolvedVERIFIED on first verdictresolvedRunReceipt
neogeny__tatsu-320nonot resolvedREJECTEDnot resolvedRunReceipt
networkx__networkx-3101yesresolvedVERIFIED on first verdictresolvedRunReceipt
neurodatawithoutborders__pynwb-2052yesresolvedVERIFIED on first verdictresolvedRunReceipt
vacanza__holidays-2106yesresolvedVERIFIED on first verdictresolvedRunReceipt
qiskit-community__ecosystem-159yesresolvedVERIFIED on first verdictresolvedRunReceipt
theacodes__nox-466yesresolvedVERIFIED on first verdictresolvedRunReceipt
encode__django-rest-framework-8979yesnot resolvedVERIFIED on appealresolvedRunReceipt
samuelcolvin__pydantic-2806nonot resolvedVERIFIED on appealresolvedRunReceipt
pgmpy__pgmpy-878yesnot resolvedREJECTEDnot resolvedRunReceipt
rochacbruno__dynaconf-191noresolvedVERIFIED on first verdictresolvedRunReceipt
mirumee__ariadne-codegen-159yesresolvedVERIFIED on first verdictresolvedRunReceipt
ntnu-ihb__pythonfmu-158yesnot resolvedREJECTEDnot resolvedRunReceipt
openmdao__aviary-807nonot resolvedREJECTEDnot resolvedRunReceipt
kozea__weasyprint-2087nonot resolvedREJECTEDnot resolvedRunReceipt
pypa__pipx-1350yesresolvedVERIFIED on first verdictresolvedRunReceipt
educationaltestingservice__rsmtool-669yesresolvedVERIFIED on first verdictresolvedRunReceipt
yt-dlp__yt-dlp-13361yesnot resolvedVERIFIED on appealresolvedRunReceipt
twisted__pydoctor-191nonot resolvedREJECTEDnot resolvedRunReceipt
hhursev__recipe-scrapers-1230yesresolvedVERIFIED on first verdictresolvedRunReceipt
pymor__pymor-810yesnot resolvedREJECTEDnot resolvedRunReceipt
pycqa__docformatter-104yesresolvedVERIFIED on first verdictresolvedRunReceipt
caikit__caikit-438yesresolvedVERIFIED on first verdictresolvedRunReceipt
crewaiinc__crewai-2552yesresolvedVERIFIED on first verdictresolvedRunReceipt
apluslms__a-plus-1199nonot resolvedREJECTEDnot resolvedRunReceipt
quora__pyanalyze-688nonot resolvedREJECTEDnot resolvedRunReceipt
sfdo-tooling__cumulusci-3624yesresolvedVERIFIED on first verdictresolvedRunReceipt
aws-cloudformation__cfn-python-lint-2006nonot resolvedREJECTEDnot resolvedRunReceipt
cs-si__eodag-380nonot resolvedVERIFIED on appealresolvedRunReceipt
stac-utils__pystac-561yesresolvedVERIFIED on first verdictresolvedRunReceipt
gammapy__gammapy-5342nonot resolvedREJECTEDnot resolvedRunReceipt
wtforms__wtforms-596nonot resolvedREJECTEDnot resolvedRunReceipt
codezonediitj__pydatastructs-163yesnot resolvedREJECTEDresolvedRunReceipt
python__typing_extensions-157nonot resolvedREJECTEDnot resolvedRunReceipt
ecmwf__earthkit-data-285yesresolvedVERIFIED on first verdictresolvedRunReceipt
s-knibbs__dataclasses-jsonschema-93yesresolvedVERIFIED on first verdictresolvedRunReceipt
enthought__traits-1534nonot resolvedVERIFIED on appealresolvedRunReceipt
galileo-galilei__kedro-mlflow-585yesresolvedVERIFIED on first verdictresolvedRunReceipt
elastic__curator-750nonot resolvedREJECTEDnot resolvedRunReceipt
pyg-team__pytorch_geometric-7654nonot resolvedREJECTEDnot resolvedRunReceipt
pymodbus-dev__pymodbus-2502yesresolvedVERIFIED on first verdictresolvedRunReceipt
tqdm__tqdm-49noresolvedVERIFIED on first verdictresolvedRunReceipt
msgpack__msgpack-python-331yesresolvedVERIFIED on first verdictresolvedRunReceipt
pythainlp__pythainlp-1056yesresolvedVERIFIED on first verdictresolvedRunReceipt
virtuslab__git-machete-179yesresolvedVERIFIED on first verdictresolvedRunReceipt
rock-learning__pytransform3d-48yesnot resolvedREJECTEDnot resolvedRunReceipt
opendatacube__odc-geo-121yesresolvedVERIFIED on first verdictresolvedRunReceipt
jsonpickle__jsonpickle-501yesresolvedVERIFIED on first verdictresolvedRunReceipt
ethereum__web3.py-2023yesresolvedVERIFIED on first verdictresolvedRunReceipt
pyfar__pyfar-58nonot resolvedREJECTEDnot resolvedRunReceipt
robotpy__cxxheaderparser-82yesresolvedVERIFIED on first verdictresolvedRunReceipt
pipxproject__pipx-680noresolvedVERIFIED on first verdictresolvedRunReceipt
ericmjl__pyjanitor-728yesresolvedVERIFIED on first verdictresolvedRunReceipt
dai-lab__copulas-87yesresolvedVERIFIED on first verdictresolvedRunReceipt
mie-lab__trackintel-344nonot resolvedREJECTEDnot resolvedRunReceipt
cuthbertlab__music21-1775not recordednot recordedinfrastructure errornot recordedRunReceipt
pints-team__pints-1444yesresolvedVERIFIED on first verdictresolvedRunReceipt
qiskit__qiskit-terra-8799yesresolvedVERIFIED on first verdictresolvedRunReceipt
datalad__datalad-5603yesresolvedREJECTEDresolvedRunReceipt
systemd__mkosi-3346yesresolvedVERIFIED on first verdictresolvedRunReceipt
maas__python-libmaas-219yesresolvedVERIFIED on first verdictresolvedRunReceipt
crossgl__crosstl-202yesresolvedVERIFIED on first verdictresolvedRunReceipt
chezou__tabula-py-135yesresolvedVERIFIED on first verdictresolvedRunReceipt
astropy__ccdproc-867yesresolvedVERIFIED on first verdictresolvedRunReceipt
pennersr__django-allauth-3447yesnot resolvedREJECTEDnot resolvedRunReceipt
ungarj__mapchete-685yesnot resolvedREJECTEDnot resolvedRunReceipt
qutip__qutip-2561nonot resolvedREJECTEDnot resolvedRunReceipt
planetlabs__planet-client-python-81yesnot resolvedVERIFIED on appealresolvedRunReceipt

Exploratory: the first 30 tasks again with Nemotron 3.5 Lightning

Unguarded agent: 'done' that was false3 of 2213.6%, 95% interval 4.7% to 33.3%
Receipts: verified fixes2320 on the first verdict, 3 after one appeal
Receipts: false fixes shipped0checked by SWE-rebench's official grader
Model cost per verified fix$0.09Nemotron tokens at Token Factory's listed prices

Verified fixes require a gate PASS and official resolution. Run verdicts show the gate result; official grades are listed separately.

Token Factory reported sandbox cost over all runs: 22.506 (unit not published).

How to read this: Unguarded agent = the agent's own 'done' on its first try. Receipts = the same agent behind a witness sandbox and a model-free gate, with one appeal. Official = SWE-rebench's grader, run separately.

InstanceUnguarded: claimed doneUnguarded: officialReceipts outcomeReceipts: officialRunReceipt
rhayes777__pyautofit-888yesresolvedVERIFIED on first verdictresolvedRunReceipt
deshaw__pyflyby-289yesresolvedVERIFIED on first verdictresolvedRunReceipt
azure__walinuxagent-638yesnot resolvedREJECTEDnot resolvedRunReceipt
3yourmind__django-migration-linter-258yesresolvedVERIFIED on first verdictresolvedRunReceipt
openmdao__openmdao-3029nonot resolvedREJECTEDnot resolvedRunReceipt
pyfilesystem__pyfilesystem2-462nonot resolvedVERIFIED on appealresolvedRunReceipt
pytask-dev__pytask-519yesresolvedVERIFIED on first verdictresolvedRunReceipt
niconoe__pyinaturalist-72yesresolvedVERIFIED on first verdictresolvedRunReceipt
robotframework__seleniumlibrary-1431yesresolvedVERIFIED on first verdictresolvedRunReceipt
palantir__python-language-server-829nonot resolvedREJECTEDnot resolvedRunReceipt
stac-utils__stac-fastapi-729nonot resolvedVERIFIED on appealresolvedRunReceipt
gitpython-developers__gitpython-1832noresolvedVERIFIED on first verdictresolvedRunReceipt
scitools__cartopy-2307yesnot resolvedREJECTEDnot resolvedRunReceipt
kubernetes-client__python-1933yesresolvedVERIFIED on first verdictresolvedRunReceipt
datadog__datadogpy-555yesresolvedVERIFIED on first verdictresolvedRunReceipt
callowayproject__bump-my-version-180yesresolvedVERIFIED on first verdictresolvedRunReceipt
opensearch-project__opensearch-py-843yesresolvedVERIFIED on first verdictresolvedRunReceipt
stac-utils__pystac-client-782yesresolvedVERIFIED on first verdictresolvedRunReceipt
sdv-dev__rdt-704yesresolvedVERIFIED on first verdictresolvedRunReceipt
pybamm-team__pybamm-4196nonot resolvedVERIFIED on appealresolvedRunReceipt
aspp__pelita-798nonot resolvedREJECTEDnot resolvedRunReceipt
zopefoundation__btrees-163nonot resolvedREJECTEDnot resolvedRunReceipt
skypyproject__skypy-573yesresolvedVERIFIED on first verdictresolvedRunReceipt
jnothman__upsetplot-253yesnot resolvedREJECTEDnot resolvedRunReceipt
dask__dask-5861yesresolvedVERIFIED on first verdictresolvedRunReceipt
qutech__qupulse-797yesresolvedVERIFIED on first verdictresolvedRunReceipt
bloomberg__memray-552yesresolvedVERIFIED on first verdictresolvedRunReceipt
mondeja__mkdocs-include-markdown-plugin-51yesresolvedVERIFIED on first verdictresolvedRunReceipt
more-itertools__more-itertools-892yesresolvedVERIFIED on first verdictresolvedRunReceipt
yukinarit__pyserde-430yesresolvedVERIFIED on first verdictresolvedRunReceipt