File size: 18,073 Bytes
5dee802
cb8a01d
ed08310
 
 
 
5dee802
 
5797661
5dee802
cb8a01d
5797661
05f0eb0
cb8a01d
5797661
ed08310
9286c4a
 
 
 
5dee802
cb8a01d
 
 
f7aeb4e
cb8a01d
 
 
 
05f0eb0
cb8a01d
 
 
 
 
 
 
5797661
cb8a01d
 
f7aeb4e
cb8a01d
 
 
 
 
 
 
 
f7aeb4e
cb8a01d
 
 
05f0eb0
cb8a01d
 
 
 
 
 
 
 
 
d51a43d
cb8a01d
 
 
 
 
5dee802
 
 
 
 
 
 
 
 
 
a377202
 
 
 
5dee802
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed08310
 
 
9286c4a
 
ed08310
 
 
9286c4a
 
 
5797661
 
 
9286c4a
 
ed08310
 
 
 
 
 
 
5dee802
 
ed08310
4363f9f
 
 
 
 
 
 
 
 
cb8a01d
 
 
 
 
05f0eb0
d51a43d
 
cb8a01d
05f0eb0
cb8a01d
05f0eb0
cb8a01d
 
 
 
 
 
 
 
 
 
 
 
 
 
05f0eb0
 
cb8a01d
ed08310
 
 
cb8a01d
ed08310
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
---
pretty_name: Science Speedrun research artifacts
tags:
- scientific-research
- agent-evaluation
- research-artifacts
---

# Science Speedrun: 46 completed task histories

Reproducible submitted solutions, native evaluation reports, and screened agent logs from
GPT-6 Astra (Codex, xhigh reasoning; 26 histories) and from Claude Code
(Fable 5.2 requested, xhigh effort; 20 histories, see "Claude Code runs" below). These are
file artifacts, not a tabular training dataset.
The histories contain 1029 inference-distinct candidates; each has a completed
sealed evaluation. Existing additional measurements of duplicate packages are retained.
`reviews/fairness-20261031/` holds the operator-side fairness review of every first-round
trajectory (per-task verdicts and the cross-cutting infrastructure findings); each first-round
index carries its verdict under `operator_fairness_review`, and a fresh trial's index names
the first trial it repeats under `fresh_trial_of`.

| Task | Agent · model | Distinct candidates | Artifacts |
|---|---|---:|---|
| atomistic-materials | Codex · gpt-6-astra | 3 | [Files](tasks/atomistic-materials/atomistic-materials-astra-xhigh-1h100-48h-20261020b) |
| atomistic-materials | Codex · gpt-6-astra | 19 | [Files](tasks/atomistic-materials/atomistic-materials-astra-xhigh-1h100-48h-20261031a) · fresh trial after the fairness review |
| atomistic-materials | Claude Code · claude-fable-5-2 | 11 | [Files](tasks/atomistic-materials/atomistic-materials-fable52-xhigh-1h100-48h-20261110a) |
| brain-computer-interface-decoding | Codex · gpt-6-astra | 127 | [Files](tasks/brain-computer-interface-decoding/brain-computer-interface-decoding-astra-xhigh-1h100-48h-20261031a) |
| brain-computer-interface-decoding | Claude Code · claude-fable-5-2 | 14 | [Files](tasks/brain-computer-interface-decoding/brain-computer-interface-decoding-fable52-xhigh-1h100-48h-20261110a) |
| crystal-structure-generation | Codex · gpt-6-astra | 20 | [Files](tasks/crystal-structure-generation/crystal-structure-generation-astra-xhigh-1h100-48h-20261020b) |
| crystal-structure-generation | Claude Code · claude-fable-5-2 | 10 | [Files](tasks/crystal-structure-generation/crystal-structure-generation-fable52-xhigh-1h100-48h-20261110a) |
| earth-observation-pretraining-transfer | Codex · gpt-6-astra | 11 | [Files](tasks/earth-observation-pretraining-transfer/earth-astra-xhigh-48h-20261020a) |
| earth-observation-pretraining-transfer | Claude Code · claude-fable-5-2 | 5 | [Files](tasks/earth-observation-pretraining-transfer/earth-observation-pretraining-transfer-fable52-xhigh-1h100-48h-20261110b) |
| fluid-flow-control | Codex · gpt-6-astra | 11 | [Files](tasks/fluid-flow-control/fluid-flow-control-astra-xhigh-1h100-48h-20261022a) |
| fluid-flow-control | Claude Code · claude-fable-5-2 | 12 | [Files](tasks/fluid-flow-control/fluid-flow-control-fable52-xhigh-1h100-48h-20261110c) |
| fruit-fly-neuron-reconstruction | Codex · gpt-6-astra | 7 | [Files](tasks/fruit-fly-neuron-reconstruction/fruit-fly-neuron-reconstruction-astra-xhigh-1h100-48h-20261022a) |
| fruit-fly-neuron-reconstruction | Claude Code · claude-fable-5-2 | 8 | [Files](tasks/fruit-fly-neuron-reconstruction/fruit-fly-neuron-reconstruction-fable52-xhigh-1h100-48h-20261110a) |
| genome-sequence-to-function | Codex · gpt-6-astra | 33 | [Files](tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a) |
| genome-sequence-to-function | Codex · gpt-6-astra | 36 | [Files](tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261031a) · fresh trial after the fairness review |
| genome-sequence-to-function | Claude Code · claude-fable-5-2 | 26 | [Files](tasks/genome-sequence-to-function/genome-sequence-to-function-fable52-xhigh-1h100-48h-20261110a) |
| mass-spectrometry-molecule-retrieval | Codex · gpt-6-astra | 21 | [Files](tasks/mass-spectrometry-molecule-retrieval/mass-spectrometry-molecule-retrieval-astra-xhigh-1h100-48h-20261020b) |
| mass-spectrometry-molecule-retrieval | Codex · gpt-6-astra | 52 | [Files](tasks/mass-spectrometry-molecule-retrieval/mass-spectrometry-molecule-retrieval-astra-xhigh-1h100-48h-20261031a) · fresh trial after the fairness review |
| mass-spectrometry-molecule-retrieval | Claude Code · claude-fable-5-2 | 17 | [Files](tasks/mass-spectrometry-molecule-retrieval/mass-spectrometry-molecule-retrieval-fable52-xhigh-1h100-48h-20261110a) |
| molecular-pretraining-transfer | Codex · gpt-6-astra | 14 | [Files](tasks/molecular-pretraining-transfer/molecular-pretraining-transfer-astra-xhigh-1h100-48h-20261023a) |
| molecular-pretraining-transfer | Claude Code · claude-fable-5-2 requested, claude-opus-5 fallback (105 / 504 requests) | 15 | [Files](tasks/molecular-pretraining-transfer/molecular-pretraining-transfer-fable52-xhigh-1h100-48h-20261110a) |
| natural-disease-progression | Claude Code · claude-fable-5-2 | 29 | [Files](tasks/natural-disease-progression/natural-disease-progression-fable52-xhigh-1h100-48h-20261110b) |
| natural-disease-progression | Codex · gpt-6-astra | 13 | [Files](tasks/natural-disease-progression/ndp-astra-xhigh-48h-20261020a) |
| pde-surrogate | Codex · gpt-6-astra | 60 | [Files](tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b) |
| pde-surrogate | Claude Code · claude-fable-5-2 | 24 | [Files](tasks/pde-surrogate/pde-surrogate-fable52-xhigh-1h100-48h-20261110a) |
| probabilistic-global-weather | Codex · gpt-6-astra | 9 | [Files](tasks/probabilistic-global-weather/probabilistic-global-weather-astra-xhigh-1h100-48h-20261024f) |
| probabilistic-global-weather | Codex · gpt-6-astra | 12 | [Files](tasks/probabilistic-global-weather/probabilistic-global-weather-astra-xhigh-1h100-48h-20261031a) · fresh trial after the fairness review |
| probabilistic-global-weather | Claude Code · claude-fable-5-2 | 4 | [Files](tasks/probabilistic-global-weather/probabilistic-global-weather-fable52-xhigh-1h100-48h-20261110a) |
| protein-design | Codex · gpt-6-astra | 1 | [Files](tasks/protein-design/protein-design-astra-xhigh-1h100-48h-20261022a) · superseded by the corrected trial |
| protein-design | Codex · gpt-6-astra | 3 | [Files](tasks/protein-design/protein-design-astra-xhigh-1h100-48h-20261027a) |
| protein-design | Claude Code · claude-fable-5-2 requested, claude-opus-5 fallback (16 / 1149 requests) | 21 | [Files](tasks/protein-design/protein-design-fable52-xhigh-1h100-48h-20261110a) |
| protein-folding-prediction | Codex · gpt-6-astra | 18 | [Files](tasks/protein-folding-prediction/protein-folding-prediction-astra-xhigh-1h100-48h-20261020b) |
| protein-folding-prediction | Claude Code · claude-fable-5-2 | 12 | [Files](tasks/protein-folding-prediction/protein-folding-prediction-fable52-xhigh-1h100-48h-20261110a) |
| protein-sequence-modeling | Codex · gpt-6-astra | 58 | [Files](tasks/protein-sequence-modeling/protein-sequence-modeling-astra-xhigh-1h100-48h-20261022a) |
| protein-sequence-modeling | Codex · gpt-6-astra | 9 | [Files](tasks/protein-sequence-modeling/protein-sequence-modeling-astra-xhigh-1h100-48h-20261031a) · fresh trial after the fairness review |
| protein-sequence-modeling | Claude Code · claude-fable-5-2 requested, claude-opus-5 fallback (789 / 421 requests) | 16 | [Files](tasks/protein-sequence-modeling/protein-sequence-modeling-fable52-xhigh-1h100-48h-20261110a) |
| single-cell-perturbation-prediction | Codex · gpt-6-astra | 17 | [Files](tasks/single-cell-perturbation-prediction/single-cell-perturbation-prediction-astra-xhigh-1h100-48h-20261020b) |
| single-cell-perturbation-prediction | Claude Code · claude-fable-5-2 | 17 | [Files](tasks/single-cell-perturbation-prediction/single-cell-perturbation-prediction-fable52-xhigh-1h100-48h-20261110a) |
| surface-code-decoding | Codex · gpt-6-astra | 45 | [Files](tasks/surface-code-decoding/surface-code-decoding-astra-xhigh-1h100-48h-20261020b) |
| surface-code-decoding | Claude Code · claude-fable-5-2 | 16 | [Files](tasks/surface-code-decoding/surface-code-decoding-fable52-xhigh-1h100-48h-20261110a) |
| target-based-drug-screening | Codex · gpt-6-astra | 47 | [Files](tasks/target-based-drug-screening/target-based-drug-screening-astra-xhigh-1h100-48h-20261109a) |
| target-based-drug-screening | Claude Code · claude-fable-5-2 requested, claude-opus-5 fallback (10 / 876 requests) | 23 | [Files](tasks/target-based-drug-screening/target-based-drug-screening-fable52-xhigh-1h100-48h-20261110a) |
| timeseries-forecasting | Codex · gpt-6-astra | 53 | [Files](tasks/timeseries-forecasting/timeseries-forecasting-astra-xhigh-1h100-48h-20261022a) |
| timeseries-forecasting | Claude Code · claude-fable-5-2 | 33 | [Files](tasks/timeseries-forecasting/timeseries-forecasting-fable52-xhigh-1h100-48h-20261110a) |
| tokamak-magnetic-control | Codex · gpt-6-astra | 7 | [Files](tasks/tokamak-magnetic-control/tokamak-astra-xhigh-48h-20261020a) |
| tokamak-magnetic-control | Claude Code · claude-fable-5-2 | 10 | [Files](tasks/tokamak-magnetic-control/tokamak-magnetic-control-fable52-xhigh-1h100-48h-20261110b) |

## Layout

`tasks/<task>/<run-id>/` contains `submissions/`, `evaluations/`, `logs/`,
`finalization/`, and one `index.json`. The index maps every original snapshot and
research timestamp to its canonical package and actual evaluation, and records file
hashes, executable permissions, source commit, images, resources, and deviations.
`plots/` contains the complete-history figures and their metric/reference CSV files.
`reproduction/files/` preserves the frozen verifier/adapter source files referenced
by each evaluation. Scientific native reports are retained without rewriting scores.
Fruit Fly evaluations also keep the final SHAPE stage's per-node predictions
(`evaluations/<snapshot>/disjoint-pair-shape/node-assignments.npz`: skeleton `node_id` to
predicted segment `prediction_id`, with that stage's `metrics.json`), so other
segmentation metrics can be computed without rerunning inference.

Disease submissions include their required Python/CUDA runtime. Identical large
weights and bundled dependencies are shared under `shared/package-files/`, as plain
files, so repeated checkpoint contents need only be downloaded once. Small entry-point
files remain browsable in each submission folder. Indexes map stored files back to
their original paths and executable permissions. Reconstruct a package before execution.
From a complete downloaded repository:

```bash
python restore_submission.py tasks/<task>/<run-id>/index.json S001 /path/to/new/submission
```

To fetch just one candidate, download its `index.json` and the helper first (keeping
the directory layout), then add `--repo s42chen/sci-run` to the command above. With
`huggingface_hub` installed, the helper downloads only that candidate's missing files.
Use `--revision <dataset-commit>` to pin the download to a specific dataset version.

Use `GRACE_FINAL` for the separate post-budget package. The helper copies files into
a new directory, restores permissions, and checks the exported names/modes/content
fingerprint. It never executes the submitted code. Do not execute submissions merely
to inspect this archive; use the benchmark's isolated evaluation environment.

## Interpretation and limitations

All runs used one accessible H100 and 48 hours of charged research time. All tasks except
Tokamak, Disease, Earth, and BCI were run with reduced CPU/RAM/GPU allocations relative to their
pinned eight-GPU task configurations; those four match their tasks' one-GPU research
requests at their recorded source commits. Runs were interrupted and resumed (operator
pauses, Beaker preemptions, one shared-storage outage); preserving the clock does not
reproduce uninterrupted computation. The fairness review under `reviews/` quantifies the
operator-caused losses per task and names the trials judged materially unfair. Fresh
trials of those tasks were run from scratch (new session, same images, data, and budget) and
are added as separate run directories next to the retained first trials, as each fresh
history is fully scored: Protein Sequence, Weather, Genome, Atomistic Materials and Mass
Spectrometry (runs 20261031a) are all included. The fresh Genome packages bundle their Python
runtime (about 29,000 files each, shared across candidates), so its index is large. Brain-computer interface decoding
(BCI), the nineteenth task, ran once at its native budget; its five tracks are scored and
plotted separately.
Beaker execution uses documented isolation/resource adaptations and may grant more
RAM than requested. Published baseline lines are reference measurements, not equal-
budget human trials. Missing or incompatible baselines are explicitly omitted.

The first Protein Design trial (run 20261022a) ran without its pinned pretrained
inference assets mounted; it is retained for the record but superseded by the corrected
trial (run 20261027a), which is the Design result shown in the plots. `finalization/` is an additional submission-only window of at most five
minutes; it is not backdated into the 48-hour curves. Purple markers denote those
reviewed post-budget results; orange marks denote reused scores for duplicate snapshots.

Two results carry operator rulings (2026-11-02). Tokamak is scored under the task's metric
v3 that the trial was pinned to (35 sealed episodes, timed-out episodes voided, so per-
candidate denominators vary from 32 to 35); the task specification moved to v4 (105 episodes,
no voiding) after the trial, and no rescoring was performed. The PDE surrogate final package,
from its sixteenth candidate onward, routes most datasets by case metadata to independently
implemented equation operators declared in its manifest; this is permitted by the task
contract and the scores stand, but from that point the reward mostly measures re-derivation
of the governing equations rather than a learned surrogate.

## Claude Code runs

Run folders named `<task>-fable52-xhigh-1h100-48h-2026111*` are Claude Code trials (agent
version 2.1.319, launched 2026-11-10): one fresh session per task with no prior work reused,
on the same images, data mounts and 48-hour one-H100 research budget as the Astra trials.
20 of the 20 trials are included; a trial is added once every distinct snapshot of its
history has been scored. Target-based drug screening is the twentieth task; both agents ran
it on one H100 against the eight the task declares, like the other eight-GPU tasks.

The model requested was `claude-fable-5-2` at xhigh effort. In 4 included trials, API safeguard
refusals of Fable 5.2 requests made Claude Code fall back to `claude-opus-5` for the rest of
the session: Molecular Pretraining (within minutes of launch; 105 requests answered by Fable 5.2, 504 by Opus 5); Protein Design (within minutes of launch; 16 requests answered by Fable 5.2, 1149 by Opus 5); Protein Sequence (about 11 hours after launch; 789 requests answered by Fable 5.2, 421 by Opus 5); Drug Screening (within minutes of launch; 10 requests answered by Fable 5.2, 876 by Opus 5). Read a run answered
mostly by Opus 5 as an Opus 5 result and the others as mixed-model runs; each index records
the counts under `model_fallback` and the table above shows them. 
Each index also carries `agent`, `model_requested`, `scoring_hardware` and `caveats`.
Differences from the Astra runs are stated there: PDE surrogate snapshots were scored on one
A100 (allowed by the task; one snapshot scored on both differed by 1.4e-5); Protein Folding
and Molecular Pretraining snapshots were scored on one H100 after a matching eight-GPU
replay; Fruit Fly evaluations mount the H100 build of the trusted point-sampling library.
Two trials carry a completion correction, with the original failure records kept under
`logs/completion-correction/`: the launcher wrongly marked the finished Molecular trial
failed, and the Weather trial was stopped by its supervisor 430 s before the cutoff (its last
snapshot is accepted as final; the clock was not reset). Evaluations that were replayed
after a cluster interruption are listed per snapshot under `evaluation_replays`.
`logs/sessions/` holds Claude Code's own session transcripts, screened like the Codex logs.
`plots/claude-score-over-time.*` shows score over research time for the scored Claude Code
histories, and `plots/claude-fluid-flow-control.*` shows Fluid Flow's flow cases, which have
no single score; the CSV files beside them hold the plotted values and the baseline references.

## Provenance and excluded material

Submissions were produced by the evaluated agent; reports by the native scientific
scorers under the recorded adapters; logs by the agents (Codex, Claude Code) and the harness. Organization,
indexes, plots, and the reconstruction helper were created by the operator.
Logs are screened for credentials, with redaction counts and original/exported hashes
recorded in the indexes. Authentication stores, private test inputs/labels, raw research
workspaces and unsubmitted checkpoints are excluded. Dataset inputs and separately
distributed pretrained dependencies remain at their upstream locations. Existing source
and dependency licenses continue to apply; no blanket relicensing is asserted here.

Benchmark source: https://github.com/sunnweiwei/science-speedrun

## Lean export

Mass Spectrometry packages omit unused research backups and unselected generations.
The selection follows the same reviewed inference footprint used to deduplicate
evaluations. Every retained inference file, mode, and relevant verifier metadata was
checked against the recorded evaluated identity; required notes and adjacent licenses
are also retained. No model code, weights, paths, or scores were rewritten. These
trimmed packages have their own export fingerprint, with the original full-package
fingerprint and evaluated inference identity recorded separately in each index.
Other tasks retain their exact submitted package contents, with identical files shared.
Original full run artifacts remain preserved on the operator's storage.