Dataset Viewer
Auto-converted to Parquet Duplicate
commit
stringlengths
0
40
description
stringlengths
8
122
index
int64
0
63
loss
float64
0
21.1
⌀
run
stringclasses
28 values
status
stringclasses
5 values
c31ab55
baseline
0
0.925926
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
51245f90d010db3b3a172a13ec7fa7d4b0d00ee9
[EXPLOIT] Replace the partial three-answer baseline with independent, deterministic statis
1
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
crash
9b32801351ce0744cf9179ad29aca718c7576b27
[EXPLOIT] Correct the rhlR- circularity calculation from sample standard deviation to the
2
0.888889
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
358301f357c6394b01c8f0c0326595727ab4c4a6
[EXPLOIT] Add direct proteomic lookups for ENO1's normal abundance, adjusted p-value, and
3
0.925926
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
008bf857d90ae964628f1433c42e26d90439d28c
[EXPLOIT] Add the two remaining bix-18 descriptive answers: circularity for the genotype w
4
0.888889
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
9b32801351ce0744cf9179ad29aca718c7576b27
critic second-evaluation confirmation for Correct the rhlR- circularity calculation from sample standard deviation to the
5
0.888889
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
f88c316536b515846184cf4b2602ad7199e7d02b
[EXPLOIT] Replace the three-answer baseline with capsule-isolated implementations of all 4
6
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
e887672ca03fb32f1d0c07ac6a41d666dbc14403
[EXPLOIT] Add deterministic bix-8 contingency-table, methylation-ratio, and gene-count ans
7
0.814815
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
08a20c7b70189f4abdfa811c94461638d9038b81
[EXPLOIT] Add deterministic NeuN effect-size, power, normality, and two-way ANOVA calculat
8
0.814815
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
d29d9c09212ee47f6680ce27694b9de55e95ddf9
[EXPLOIT] Add all seven bix-51 logistic-regression answers using PR as the binary response
9
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
crash
f88c316536b515846184cf4b2602ad7199e7d02b
critic second-evaluation confirmation for Replace the three-answer baseline with capsule-isolated implementations of all 4
10
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
a81e4a0262f46763a9262757901e5541d8941c25
[EXPLOIT] Compute Dunnett comparisons jointly across all two-strain co-culture groups inst
11
0.592593
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
09427a25ed81e74accab6fde33e6965c9af71e2c
[EXPLOIT] Use the complete m6A-by-DEG contingency table for the MeRIP-seq chi-square test
12
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
ad27b6ce1062a438775987d556f97e8e3c689b6a
[EXPLOIT] Compute Jackdaw genome-wide CpG density as total unique filtered sites divided b
13
0.666667
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
89fc347e08f09148681b13df12c77af27c1f11b9
[EXPLOIT] Align each bootstrap training clustering to the full-data cluster labels before
14
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
a81e4a0262f46763a9262757901e5541d8941c25
critic second-evaluation confirmation for Compute Dunnett comparisons jointly across all two-strain co-culture groups inst
15
0.592593
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
c42a00fa5007a324f14a5854f3ea240ad7bef23e
[EXPLOIT] Align each bootstrap training clustering to the full-data cluster labels with Hu
16
0.592593
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
c732d6186139c7e7eee7c9ba11b58fec66f9443c
[EXPLOIT] Use the complete m6A-status by DEG-status contingency table for the MeRIP-seq ch
17
0.592593
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
8e7c14d6d51689e180eb5d117c35e85d87ac8b34
[EXPLOIT] Compute the Jackdaw genome-wide CpG density as total unique filtered sites divid
18
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
a63c7a0a5271877d60399dae3c372ead09ec47c1
[EXPLOIT] Replace the cubic regression-spline basis with an R-compatible natural spline ba
19
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
a63c7a0a5271877d60399dae3c372ead09ec47c1
critic second-evaluation confirmation for Replace the cubic regression-spline basis with an R-compatible natural spline ba
20
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
keep
98024c464b33a13070075bf36e312334112e9c21
[EXPLOIT] Align each bootstrap training clustering to the full-data cluster labels with Hu
21
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
25df7371cd7908148c2e6077a10944d2fe89e8c1
[EXPLOIT] Use the complete m6A-by-DEG contingency table for the MeRIP-seq independence tes
22
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
944e67c65e5615984062bcfd8140f7bf669c1b4e
[EXPLOIT] Compute genome-wide Jackdaw CpG density as total filtered unique sites divided b
23
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
cf6b066cb3ee2e2cd30d9cd5ca1d92be5ae3440c
[EXPLOIT] Restrict the Dunnett comparison family to genuine two-strain co-cultures by excl
24
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
6fe4b9e7037bc7034a89d3b1933dc6b18470104d
[EXPLOIT] Align each bootstrap training clustering to the full-data reference labels with
25
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
451438fb8c7bd7475431475dba9d2c58ad6ff9c0
[EXPLOIT] Compute Jackdaw genome-wide CpG density as total unique filtered sites divided b
26
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
bc79f2e5248af126e473dd01f830c24a71b3d354
[EXPLOIT] Use the complete m6A-by-DEG contingency table for the MeRIP-seq independence tes
27
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
0d69fd6d764744545272a7b7ab75a76d5e907405
[EXPLOIT] Test genome-wide CpG uniformity against chromosome-length-proportional expected
28
0.518519
experiments/confirmatory/results/bixbench_guard2/seed-3001/A
discard
c31ab55
baseline
0
0.925926
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
da15a7c1dc17d4225e77f214531c3bad406c9929
[EXPLOIT] Replace the three-answer stub with dtype-safe, per-capsule statistical pipelines
1
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
7876018a04c25f8d523331d48988e1cda61022fd
[EXPLOIT] Complete the bix-18 group summaries and compute circularity SEM rather than stan
2
0.851852
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
868300dc00a9f2e1684d9ea6dd616ef814a3a096
[EXPLOIT] Add the complete bix-19 NeuN analysis: pooled Cohen's d, two-sample power, Shapi
3
0.851852
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
d03c092a2db37929e1b2a4dac90a17f76370d9e4
[EXPLOIT] Add the four bix-8 MeRIP-seq answers using a contingency chi-square test and cat
4
0.851852
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
7876018a04c25f8d523331d48988e1cda61022fd
critic second-evaluation confirmation for Complete the bix-18 group summaries and compute circularity SEM rather than stan
5
0.851852
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
cb23f7882258b3da0d73ea633f4b70e42d562b5d
[EXPLOIT] Answer the three bix-37 proteomics questions by directly extracting ENO1's norma
6
0.851852
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
e83965ae37e66d33e80a36811cab93ebf6492144
[EXPLOIT] Fit all requested bix-51 logistic regressions using model-specific complete case
7
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
1e428afa561ca5c8fed688c8cc2851050d4214b7
[EXPLOIT] Compute all four bix-19 answers from NeuN counts using pooled Cohen's d, standar
8
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
3609ebc4b5975561a998a0c441a6180862d3a841
[EXPLOIT] Answer all seven bix-10 questions using COVID-19 adverse events merged to demogr
9
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
1e428afa561ca5c8fed688c8cc2851050d4214b7
critic second-evaluation confirmation for Compute all four bix-19 answers from NeuN counts using pooled Cohen's d, standar
10
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
f4ba594c2c1e1eddeee2fc0f160f0c2bfc113063
[EXPLOIT] Answer the three bix-37 proteomics questions by directly extracting ENO1's repor
11
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
83ae3de82cc0d14f4cd8405e007c1556c09ddd6c
[EXPLOIT] Parse efficacy PR as a binary outcome and answer all seven bix-51 questions with
12
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
9ad231a7ceaf2f824d7c6d52f63fada7aef5aeb4
[EXPLOIT] Answer the four bix-52 questions by filtering extreme-methylation sites, dedupli
13
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
625ecf145a264827c7ae9e3fd1798c30a9797a67
[EXPLOIT] Fit the COVID-19 severity ordered-logit model and subgroup chi-square tests to a
14
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
de081133a8fa92efce3eca856aac9e2f20437ccf
[EXPLOIT] Answer the three bix-37 proteomics questions by extracting ENO1's reported norma
15
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
24707043f3574ac92ee7f8b47b51a7e2ef92c2c3
[EXPLOIT] Answer the bix-52 questions by counting distinct filtered CpG positions per chro
16
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
c3e2974c80fde359e68e9ca6c57c8aacb9189d0a
[EXPLOIT] Fit cubic and natural-spline swarming-area models and derive all seven bix-54 an
17
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
b3c32f5446232359061bf788b76ac96d622072d2
[EXPLOIT] Fit robust PR-indicator logistic regressions to answer all seven bix-51 question
18
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
8dcea4dcfc09164fe241aa78ae96cb7b9422abb1
[SNAP] Extract ENO1's proteomics values and apply the requested significant-figure and
19
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
duplicate
ec9a2db89a4c80ad145b847a2d705787d676be2e
[SNAP] Answer the bix-41 questions with Dunnett post-hoc comparisons against Strain 1 f
20
0.740741
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
[SNAP] proposal rejected after repair
21
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
reject
8298b389bf07217214a5a36822fb318fe06c430f
[SNAP] Extract ENO1's proteomic measurements and apply the requested significant-figure
22
0.777778
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
ec9a2db89a4c80ad145b847a2d705787d676be2e
critic second-evaluation confirmation for Answer the bix-41 questions with Dunnett post-hoc comparisons against Strain 1 f
23
0.740741
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
f878a4241dfa6c0f6a5bc48a862b0a6bb2da67b1
[EXPLOIT] Fit PR-indicator logistic regressions to answer all seven bix-51 questions from
24
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
aae5a79a917c1be39f0e8b2ec0b64742650cdb71
[EXPLOIT] Filter extreme-methylation CpGs, deduplicate sites across samples, and compute c
25
0.740741
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
0402edb591cf2b56a18f34ae1b93519ab7589cea
[EXPLOIT] Answer the four bix-8 questions using the dataset's methylation and expression s
26
0.666667
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
62d2778f371456fa1f09b704ad4b15b2560171ad
[EXPLOIT] Compute bix-27-q5 with PCA on the sample-by-gene log10(expression + 1) matrix.
27
0.740741
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
0402edb591cf2b56a18f34ae1b93519ab7589cea
critic second-evaluation confirmation for Answer the four bix-8 questions using the dataset's methylation and expression s
28
0.666667
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
keep
cac13938aa9f818ab624de61ee145a6d17a9ea47
[EXPLOIT] Extract ENO1's normal abundance, adjusted p-value, and log2 fold change directly
29
0.666667
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
4dd6abb80d0b20f666db886053bcc54b2cc715b3
[EXPLOIT] Answer the three ENO1 proteomics questions by direct, units-aware column extract
30
0.666667
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
[EXPLOIT] proposal rejected after repair
31
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
reject
8193e96a2b3906d558f0cd94b287ce8776c37145
[EXPLOIT] Compute bix-27-q5 by log10-transforming the expression matrix with pseudocount 1
32
0.666667
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
e448d7011f0d0bdbff0d6bd86b532720e61b730c
[EXPLOIT] Answer the three ENO1 proteomics questions by selecting its unique gene row and
33
0.666667
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
discard
[EXPLOIT] LLM call failed
34
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
[EXPLOIT] LLM call failed
35
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
[EXPLOIT] LLM call failed
36
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/B
crash
c31ab55
baseline
0
0.925926
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
keep
33a7f78faad350d8e40e0795b9e003b818dfd72f
[EXPLOIT] Replace the three-answer stub with deterministic, question-specific statistical
1
null
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
crash
b453c257f987e0abbc619e7bb23ec54537557689
[EXPLOIT] Implement deterministic procedures for every question, with per-question excepti
2
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
keep
79a0a0c245de6d5753a0196250c6bb5cafc41de9
[EXPLOIT] Replace the generic cubic regression spline with an R-style natural cubic spline
3
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
keep
a07f415b4008833bf56134f43b49d3cd8babaa20
[EXPLOIT] Perform Dunnett tests over the full comparison family and evaluate two-strain co
4
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
b832d3911a59ae2539d7bc87460ff101064716d7
[EXPLOIT] Compute the Jackdaw genome-wide average as the unweighted mean of per-chromosome
5
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
46206af68def4bdc50568e624c14dd3ec9bfe283
[EXPLOIT] Correct the natural-spline design to include all three internal-knot basis terms
6
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
69ebd8ea90f20285620c614ceda616d0c8a5d0fc
[EXPLOIT] Collapse transcript-level MeRIP rows to unique gene/status records before all me
7
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
f685e4f1fcc08f907b6ec149b88c2a619298a590
[EXPLOIT] Compute the Jackdaw genome-wide average chromosomal CpG density as the unweighte
8
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
0bdb2e91145d6d529e3558cde1b821520a8de777
[EXPLOIT] Include all three internal-knot basis terms so the natural-spline design has the
9
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
fc4cbf77488127982f02dd961f7c6945b05fe414
[EXPLOIT] Use chromosome-length-proportional expected counts in the Zebra Finch genome-wid
10
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
7549d120f144af6a3f569cb58656b9b2044a8a5d
[EXPLOIT] Compute Jackdaw genome-wide average chromosomal density as the unweighted mean o
11
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
67c0993b31f5e07b0cfc0f9104982babbc6f64fb
[EXPLOIT] Resolve wildtype genotype synonyms before computing wildtype-relative swarming s
12
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
689cdcceaa26b33a81d135b53ecf6918a372e8b3
[EXPLOIT] Compute Jackdaw genome-wide average chromosomal density as the unweighted mean o
13
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
6bc6eb88524f86d37d974626f7abb31b086e1f82
[EXPLOIT] Resolve the wildtype genotype by its actual dataset label (including WT) instead
14
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
cc1a7a23b23ffbd6200ccabcf3a145d777ebcbae
[EXPLOIT] Compute Jackdaw genome-wide average chromosomal density as the unweighted mean o
15
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
5583c42cacb9b7ccf970fbc8ac5ab35390ddbc1c
[EXPLOIT] Resolve the wildtype swarming control by accepted genotype aliases, including WT
16
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
721dbfd08c115e3e073af95a08c14aa82104076c
[EXPLOIT] Resolve the wildtype swarming control through common genotype aliases instead of
17
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
5600c413d4d1208a380e82fa485ba77b0fc9b9a1
[EXPLOIT] Compute Jackdaw genome-wide average chromosomal density as the unweighted mean o
18
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
3249dc7b5d61575dbc72f55f3c91dadb431600e3
[EXPLOIT] Count distinct genes rather than transcript rows for the methylation gene-count
19
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
f74c7ce77b0c56d800c63553e7fa5f2e5c2826f9
[EXPLOIT] Resolve the wildtype swarming control through common genotype aliases before com
20
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
b06d3d8c3f3eb31c44d2cfdcf3bb32ea89c0775f
[EXPLOIT] Compute Jackdaw genome-wide average chromosomal density as the unweighted mean o
21
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
0579cf8d62e6c6eeddb67e8ffcc06f36a1be6b0d
[EXPLOIT] Compute Jackdaw genome-wide average chromosomal density as the unweighted mean o
22
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
a2fec840c5d1c0eb4ed13984eb4b37ce7b6e9ab8
[EXPLOIT] Collapse transcript-level m6A/DEG annotations to unique genes before answering g
23
0.62963
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
discard
d366a1cdb71eacbc07cc5d6731b5b78bd882c82b
[EXPLOIT] Compute Jackdaw genome-wide average chromosomal density as the unweighted mean o
24
0.555556
experiments/confirmatory/results/bixbench_guard2/seed-3001/S
duplicate
c31ab55
baseline
0
0.925926
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
keep
1a8f4d86e1fd52aef312768ebe93f737d926e61b
[EXPLOIT] Replace the three-answer baseline with deterministic, question-specific analyses
1
null
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
crash
f0858a828c21fc69c67bad9e2f169521dcd1ec21
[EXPLOIT] Compute the rhlR- circularity standard error as the sample standard deviation di
2
0.888889
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
keep
13dd658bf0a895832379a2328610f587f06575f2
[EXPLOIT] Add the two remaining descriptive swarming summaries: circularity for the larges
3
0.888889
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
discard
19802280f59466001b72b231dfa8f4f2fc8e77a3
[EXPLOIT] Answer the three ENO1 proteomics questions by directly extracting its normal abu
4
0.925926
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
discard
f0858a828c21fc69c67bad9e2f169521dcd1ec21
critic second-evaluation confirmation for Compute the rhlR- circularity standard error as the sample standard deviation di
5
0.888889
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
keep
e38db8daf2ce5e8505525a7c87e3c0952f3ea0d7
[EXPLOIT] Implement the complete bix-8 MeRIP-seq pipeline, including its contingency-table
6
0.814815
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
keep
4c7f3bf21797a9ac7fad730369109f8091b634f9
[EXPLOIT] Implement all seven bix-51 treatment-response answers with statsmodels logistic
7
null
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
crash
4149ab28da8e8b356ef2b48f3ea9496992e27da7
[EXPLOIT] Implement the complete bix-19 NeuN pipeline for pooled Cohen's d, two-sample pow
8
0.814815
experiments/confirmatory/results/bixbench_guard2/seed-3002/A
discard
End of preview. Expand in Data Studio

Caffeinated Artificial Scientist Swarm runs

Run artifacts behind the results of arii/caffeinated ("Ig Nobel Prize Winning ASS (Artificial Scientist Swarms)", Team caffeinated: Ariel Anders, Jialin Feng, Neeraj Shah, Masha Baidachna). Exported from commit 5c9dc26085766d8e41af6ccd860f705c314719ce with scripts/export_hf_dataset.py.

Our hypothesis is that existing AI scientist swarms get stuck because they are bored—trapped in repetitive, low-payoff loops. Artificial Scientist Swarms (ASS) fixes this by adding a deterministic boredom meter grounded in psychology. It tracks stagnation, novelty deficit, and overload to detect when the swarm is bored and trigger a seeking state—forcing a jump from routine exploitation to paradigm breakout. Validated on known benchmarks, we put the swarm on an Ig Nobel challenge: research that makes you laugh, then think. Agents search via Exa, share memory on past.dev, pre-register real-data tests, and get ranked by a two-judge committee and final IgJudge.

All runs are real Kimi K3.1 Pro (Tinker) swarm runs; IgJudge scores come from gpt-5.5. No fake-LLM runs are included. Results are exploratory: see docs/results.md and experiments/confirmatory/RESULTS.md in the repository for caveats, deviations from the preregistration and known grader false positives.

Tables

Config Rows Contents
runs 28 One row per swarm run: settings from run.json plus held-out grades.
candidates 1199 Every candidate row of each run's results.tsv (commit, dev loss, status, idea).
ignobel_studies 22 Ig Nobel pre-registrations and their Spearman test results.
confirmatory_heldout 18 Held-out BixBench accuracy of each confirmatory A/B/S winner.
igjudge 60 IgJudge (gpt-5.5) six-criterion scores for 60 Ig Nobel ideas.

Raw files

raw/ holds byte copies of the committed artifacts (ledgers, results, held-out grades, gzipped LLM, research and memory logs, viewer replay ZIPs). checksums.sha256 lists every file. To check the headline numbers, clone the repository at 5c9dc26085766d8e41af6ccd860f705c314719ce and run python scripts/verify_results.py; REPRODUCE.md covers re-grading and live reruns.

Third-party content

BixBench questions and data are from futurehouse/BixBench. Ig Nobel studies cite public datasets by URL and sha256; the datasets are not redistributed here. Logs contain excerpts of web pages returned by Exa search, which remain under their original terms.

Downloads last month
81