The dataset viewer is not available for this subset.
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/harbor/harbor.py", line 171, in _split_generators
raise DataFilesNotFoundError("No task.toml or instruction.md files found")
datasets.exceptions.DataFilesNotFoundError: No task.toml or instruction.md files found
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.
ConMgmt-Bench
An open evaluation set for AI agents in construction project management — from tender to claim. Version 1.0.
Twenty scenarios from one synthetic project (Polaris DC-1, a data-campus design-and-build in Finland), covering the work a general contractor's planning, commercial and claims teams actually do: baseline programme, productivity rates, contract analysis across three revisions, bid levelling and award, subcontract gap analysis, programme compliance, drawings, site logistics, a board pack, three month-ends of progress reporting, two claims argued in opposite directions, a variation valuation and a determination response.
Each scenario is a prompt a construction professional would actually send, the documents they would have at that moment, outcome expectations, an answer key where one was written before the run, and our own outputs as baseline 1.0 — what an acceptable answer looks like, not a ceiling.
project/— the tender-stage document set (contract, drawings, specifications, bill, offers, permits, surveys). Many are deliberate scans.scenarios/— one folder per scenario:prompt.md,expectations.md,key/,inputs/(documents that arrive for that scenario),baseline-1.0/(model, outputs, grading).chain.jsongives the order and dependencies;_BASELINE_NOTE.mdthe caveats.scoring/— how to score, judge rubrics where they exist, the results template, the leaderboard.tools/— programme metrics for PRE-01 and a results-sheet builder.
Running it
Chained is the default: follow scenarios/chain.json, add each scenario's inputs/ to the project before starting it, keep earlier deliverables in the project. One-shot variants (prompt-oneshot.md) exist for the tasks the baseline ran as conversations. One run per scenario is enough; say which model, which date, chained or one-shot.
What you need that is not here: the FIDIC Yellow Book 2017 General Conditions (copyright; see project/01 Contract and Commercial/_NOT_SHIPPED_FIDIC_Yellow_Book_2017.md).
Scoring
Per scenario: each expectation 0 / 0.5 / 1, and each key item found / partial / missed. Report both; do not collapse them into one number. scoring/HOW_TO_SCORE.md has the detail and scoring/results-template.csv the sheet. Send results as a pull request to scoring/leaderboard.md on GitHub (https://github.com/AshrafMMahdy/conmgmt-bench), which is the primary home of this set; this dataset is a mirror.
Sponsored runs
If you want a model or an agent run against the full set and graded by the authors, see sponsored-runs.md.
Licence
Project documents, prompts, keys and baseline outputs: CC BY 4.0 (LICENSE-DATA.md). Scripts: MIT (LICENSE-CODE.md). The project, companies, people and figures are fictional.
- Downloads last month
- 143