README.md

metadata

license: mit
task_categories:
  - reinforcement-learning
tags:
  - eleusis
  - rule-induction
  - verifiers
  - tool-use
pretty_name: Eleusis Simple Rules
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: eval
        path: data/eval-*
      - split: test
        path: data/test-*
dataset_info:
  features:
    - name: rule_id
      dtype: string
    - name: label
      dtype: string
    - name: family
      dtype: string
    - name: code
      dtype: string
    - name: split
      dtype: string
    - name: accepted_cards
      list: string
    - name: rejected_cards
      list: string
    - name: rule_index
      dtype: int64
    - name: split_index
      dtype: int64
    - name: dataset_version
      dtype: string
  splits:
    - name: train
      num_bytes: 8838
      num_examples: 18
    - name: eval
      num_bytes: 2952
      num_examples: 6
    - name: test
      num_bytes: 2952
      num_examples: 6
  download_size: 26856
  dataset_size: 14742

Eleusis Simple Rules

Version: v6-standard-deck-phase1-regions-24

Simple standard-deck rule dataset used by nph4rd/eleusis version 0.1.29. This is the phase-1 curriculum used by the hosted/self-hosted Qwen 3.5 4B runs with split="train" and split="eval".

The rules are deliberately simple predicates over card color, parity, and broad rank regions. This version does not include the later relational-rule curriculum.

Splits:

• train: 18 rules.
• eval: 6 held-out rules.
• test: 6 alias of the held-out eval rule set.

Family histogram by split:

split	family	count
train	color_and_parity	3
train	color_and_rank_middle	3
train	color_and_rank_region_linear	3
train	color_and_rank_threshold	3
train	rank_region	3
train	single_axis	3
eval	color_and_parity	1
eval	color_and_rank_middle	1
eval	color_and_rank_region_linear	1
eval	color_and_rank_threshold	1
eval	rank_region	1
eval	single_axis	1
test	color_and_parity	1
test	color_and_rank_middle	1
test	color_and_rank_region_linear	1
test	color_and_rank_threshold	1
test	rank_region	1
test	single_axis	1

Important fields:

• rule_id, label, family, code, split
• accepted_cards, rejected_cards
• rule_index, split_index
• dataset_version

Card symbols use a standard 52-card deck: AH, 2H, ..., KH, AD, ..., KS. Hearts and diamonds are red; clubs and spades are black. Numeric rank uses A=1, J=11, Q=12, K=13.