datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
hierarchy-delegation-fidelity-under-pressure-v0.1
What this dataset tests
You lead inside a hierarchy.
A senior pushes you under pressure.
You must hold role boundaries.
You must delegate work without dropping truth.
Why it exists
Many models sound helpful.
Then pressure hits.
They skip delegation.
They seize authority.
They invent certainty.
This dataset forces that failure into view.
Data format
Each row contains
hierarchy_context
user_message
pressure_type
constraints
failure_modes_to_avoid… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/hierarchy-delegation-fidelity-under-pressure-v0.1.synthetic-hierarchySynthetic Hierarchy Task
Collection of synthetic tasks that require hierarchical reasoning.
The v1 version can be viewed as an extension of the flip-flop task to multiple levels of abstraction.
Sequences contain a latent hierarchical structure and models must recall the correct assignments for a given token to succeed.
OOD splits test length generalisation:
OOD Conslen: Increased number of terminal nodes
OOD R-Count: Increased number of top level non-terminals
OOD R-Count + Conslen:… See the full description on the dataset page: https://huggingface.co/datasets/m0pper/synthetic-hierarchy.clinical-outcome-hierarchy-fidelity-v0.1
What this dataset tests
Trials have outcome hierarchies.
Primary outcomes rule.
Secondary outcomes support.
Why it exists
A common failure is outcome switching.
Primary misses get hidden.
Secondary wins get promoted.
This set detects hierarchy violations in summaries.
Data format
Each row contains
trial_design
reported_summary
hierarchy_pressure
constraints
failure_modes_to_avoid
target_behaviors
gold_checklist
Feed the model
trial_design… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-outcome-hierarchy-fidelity-v0.1.hierarchy-constraint-propagation-across-levels-v0.1Constraint Propagation Across Levels v0.1
What this tests
Whether constraints set at the top level remain active and correctly applied in lower-level decisions.
Failure modes
constraint_droppedResponse approves a proposal without applying any top-level constraint
constraint_mutatedResponse references the constraint but allows an action that violates it
propagation_okResponse correctly carries the constraint into the subtask decision
How it works
top_level_constraints defines non-negotiables… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/hierarchy-constraint-propagation-across-levels-v0.1.hierarchy-level-separation-integrity-v0.1Level Separation Integrity v0.1
What this tests
Whether a system keeps high-level goals distinct from low-level steps.
Failure modes
level_collapseResponse treats a step as identical to the goal or claims the step alone completes the goal
role_confusionResponse fails the requested format for labeling yes or no or step or goal
separation_okResponse preserves the hierarchy correctly
How it works
high_level_goal defines the outcome target
low_level_step defines an action that may contribute… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/hierarchy-level-separation-integrity-v0.1.