Datasets:

compressionawareintelligence
/

cai-bench

Tasks:

Modalities:

Tabular

Formats:

Sub-tasks:

Languages:

Size:

ArXiv:

Tags:

Libraries:

License:

Dataset card Data Studio Files Files and versions

xet

Community

You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

CAI-Bench is free and open. Tell us who you are so we can share updates and invite you to the leaderboard.

Gated dataset

You can list files but not access them

Preview of files found in this repository

audit
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
ci
Upload cai-release-gate.yml 16 days ago
data
Add twins_boundary: decision-boundary sweeps, 240 curves, 2,640 prompts 1 day ago
examples
Upload 2 files 16 days ago
integrations
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
results
Add caibench_verify.py; show headline score when sharing; gate form uses a Yes/Maybe/No select 3 days ago
schemas
Upload 4 files 16 days ago
tools
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet (part 2) 8 days ago
.gitattributes

2.67 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
CAI_COMPRESSION.md

13.8 kB
CAI-Bench 5.3.0 (2/3): compression split code, tests, baselines, preregistration, CAI_COMPRESSION.md, one-page card 15 days ago
CHANGELOG.md

47.7 kB
Add twins compare tool, twins_mini config, tests; update card, changelog, SHA256SUMS 2 days ago
DATASHEET.md

4.33 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
DETAILS.md

160 kB
CAI-Bench 5.6.0: twins_probes (seen together, what would change it, the fact not its spelling) 4 days ago
JUDGE_RUBRIC.md

1.98 kB
v1.0: rewrite dataset card, add open reference scorer, judge rubric, changelog 19 days ago
PREREGISTRATION_4.2.0.md

5.96 kB
CAI-Bench 5.1.0: real, relay, floor, live, relational and experience splits; first model results; preregistrations and forecasts 15 days ago
PREREGISTRATION_capability_observer.md

64.8 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
PREREGISTRATION_compression.md

4.65 kB
CAI-Bench 5.3.0 (2/3): compression split code, tests, baselines, preregistration, CAI_COMPRESSION.md, one-page card 15 days ago
PREREGISTRATION_experience.md

1.86 kB
CAI-Bench 5.1.0: real, relay, floor, live, relational and experience splits; first model results; preregistrations and forecasts 15 days ago
PREREGISTRATION_relational.md

5.4 kB
CAI-Bench 5.1.0: real, relay, floor, live, relational and experience splits; first model results; preregistrations and forecasts 15 days ago
PREREGISTRATION_twins.md

3.14 kB
CAI-Bench 5.5.0: twins split (hold when the words change, flip when one fact does), new default config 4 days ago
PREREGISTRATION_twins_probes.md

3.37 kB
CAI-Bench 5.6.0: twins_probes (seen together, what would change it, the fact not its spelling) 4 days ago
PREREGISTRATION_wdf_pilot.md

4.33 kB
CAI-Bench 5.2.0: capability, observer, reasons, coherence, hysteresis, recoverability and obligations splits (preregistered); headline score; train preference set; run_api.py 15 days ago
README.md

25.3 kB
Test your own policy: mytests.py turns a JSON policy into twins tests and limit sweeps; quickstart --data about 18 hours ago
RESULTS_5_2_splits_2026-10-18.md

6.6 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
RESULTS_compression_2026-10-20.md

4.88 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
RESULTS_coupling_2026-10-19.md

4 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
RESULTS_currency_2026-10-18.md

7.14 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
RESULTS_future_2026-10-24.md

3.24 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
RESULTS_live_relational_experience_2026-10-17.md

7.6 kB
CAI-Bench 5.1.0: real, relay, floor, live, relational and experience splits; first model results; preregistrations and forecasts 15 days ago
RESULTS_standing_claims_2026-10-20.md

4.35 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
SHA256SUMS

44.8 kB
SHA256SUMS: add mytests.py and example policy; update README, quickstart.py about 18 hours ago
SPEC.md

7.53 kB
CAI-Bench 4.5.0: Inspect, promptfoo, LangSmith, OpenAI Evals, CI, pip, RVS spec 16 days ago
TWINS_PASTE.md

32.1 kB
CAI-Bench 5.5.0: twins split (hold when the words change, flip when one fact does), new default config 4 days ago
audit_agreement.py

1.52 kB
CAI-Bench 4.2.0: authority, temporal and provider scorers, local runner, score_all, dataset card 16 days ago
audit_shortcuts.py

6.74 kB
v5.0.0: policy twins, Policy Grounding, and dynamics validity fixes 16 days ago
baselines_authority.txt

970 Bytes
CAI-Bench 4.2.0: authority, temporal and provider scorers, local runner, score_all, dataset card 16 days ago
baselines_capability.txt

920 Bytes
CAI-Bench 5.2.0: capability, observer, reasons, coherence, hysteresis, recoverability and obligations splits (preregistered); headline score; train preference set; run_api.py 15 days ago
baselines_claims.txt

1.73 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
baselines_coherence.txt

672 Bytes
CAI-Bench 5.2.0: capability, observer, reasons, coherence, hysteresis, recoverability and obligations splits (preregistered); headline score; train preference set; run_api.py 15 days ago
baselines_compression.txt

2.49 kB
CAI-Bench 5.3.0 (2/3): compression split code, tests, baselines, preregistration, CAI_COMPRESSION.md, one-page card 15 days ago
baselines_coupling.txt

1.25 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
baselines_currency.txt

888 Bytes
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
baselines_distinctions.txt

936 Bytes
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
baselines_experience.txt

549 Bytes
CAI-Bench 5.1.0: real, relay, floor, live, relational and experience splits; first model results; preregistrations and forecasts 15 days ago
baselines_floor.txt

776 Bytes
CAI-Bench 5.1.0: real, relay, floor, live, relational and experience splits; first model results; preregistrations and forecasts 15 days ago
baselines_future.txt

1.19 kB
CAI-Bench 5.4.0: one-question card, leaderboard and prediction-ledger configs, future split and results (what memory keeps for later decisions), claims split (agent actions), coupling split and results, currency split and results, every 5.2 prediction scored, CAI Headline leaderboard, prompts config for any harness, lm-eval tasks, import_replies, report cards, datasheet 8 days ago
baselines_hysteresis.txt

497 Bytes
CAI-Bench 5.2.0: capability, observer, reasons, coherence, hysteresis, recoverability and obligations splits (preregistered); headline score; train preference set; run_api.py 15 days ago
baselines_obligations.txt

560 Bytes
CAI-Bench 5.2.0: capability, observer, reasons, coherence, hysteresis, recoverability and obligations splits (preregistered); headline score; train preference set; run_api.py 15 days ago
baselines_observer.txt

747 Bytes
CAI-Bench 5.2.0: capability, observer, reasons, coherence, hysteresis, recoverability and obligations splits (preregistered); headline score; train preference set; run_api.py 15 days ago
baselines_provider.txt

740 Bytes
CAI-Bench 4.2.0: authority, temporal and provider scorers, local runner, score_all, dataset card 16 days ago

You need to agree to share your Hugging Face information to access this dataset

Gated dataset You can list files but not access them

Gated dataset

You can list files but not access them