Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 84, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Observatory Lab - experiment evidence
Raw evidence for the Observatory Lab program: per-call transcripts (verbatim model responses), run logs, and aggregates for the executed experiments on multi-agent coordination, consensus, error propagation, instruction conflict, supervision cadence, and society-line cooperation.
- Model under study: gemini-3.1-flash-lite (Google AI Studio free tier); cross-provider cell also used llama-3.1-8b-fp8 (Cloudflare Workers AI free tier). Cost: EUR 0 (free tiers only).
- Evidence class: exploratory unless a cell says otherwise; no statistical or generalization claims beyond what each report states.
- Code, methodology and reports: https://github.com/olserra/observatory-lab
- Landing: https://observatory-lab.vercel.app/
- Evaluators are deterministic and included in the repo; every number in the reports traces to the JSONL files here.
Artifacts
1-observatory-lab-first-batch-data.tar.gz- first experiment batch (2026-10-27): five pilots, one per research track, plus the pre-registered cross-provider EXP-C01 cell and the instrument-fix validation. Evidence class: exploratory / inconclusive as labeled in reports/first-batch-report.md.2-observatory-lab-exp-c01-gemini-replication.tar.gz- pre-registered EXP-C01 replication cell on Gemini (2026-10-28): 60/60 unique trials (3 regimes x 10 fresh seeds x injection on/off), full transcripts, aggregation rule and report included. The pilot pattern (near-universal adoption once injected, near-zero detection) replicated in the same setting; n=10 per cell, one model, one task family - a replicated result in this setting, not a general effect.
Contents (inside each tarball)
- data/exp-/transcripts/.jsonl - one record per logical model call.
- data/exp-/summary.json(l) - per-trial and per-condition aggregates.
- reports and pre-registrations for the covered cells.
Known limitations (read before reusing)
- First pass of A01/C01/D01 persisted aggregates only (labeled preliminary); rerun 1 and later runs persist full transcripts.
- Prompts in pre-2026-10-27 records are stored hashed, not verbatim; responses are verbatim. Later runs log prompts verbatim.
- B01's mediator arm is incomplete (daily quota); completed-item accuracy is a descriptive subset.
- Infrastructure errors (quota, HTTP 503, timeout) are recorded and labeled, kept in denominators, never reported as behavioral results.
- All tasks are synthetic with fictional agents; no personal data.
- Downloads last month
- 23