Redesign dataset card for pretraining, SFT, and SDFT

README.md

@@ -7,26 +7,85 @@ language:
 tags:
 - synthetic
 - state-space-models
+- pretraining
 - sft
 - sdft
+size_categories:
+- 10M<n<100M
 configs:
 - config_name: default
   data_files:
   - split: train
-    path: data/train/*
+    path: "data/train/*"
   - split: validation
-    path: data/validation/*
+    path: "data/validation/*"
 ---
 
-# StateBench v1 training corpus
+<div align="center">
 
-This is a synthetic, deterministic state-reasoning corpus containing 10000000 canonical rows.
-Every row can be viewed as pretraining text, SFT prompt/completion, or TRL on-policy SDFT
-prompt/privileged_context without regenerating semantics.
+# StateBench v1
 
-For SFT select prompt and completion. For TRL SDFT select prompt and privileged_context;
-the privileged context is teacher-only and must never be concatenated into the student prompt.
+### 10 million executable state programs for memory, recurrence, and state control.
 
-Generator version: StateBench-v1.0. See dataset_manifest.json for checksums, distributions,
-generation configuration and resumable progress. Public training data must not be used as the
-official private StateBench evaluation split.
+One deterministic corpus. Three training views: **pretraining · SFT · on-policy SDFT**.
+
+</div>
+
+| Scale | Coverage | Storage | Splits |
+|:--|:--|:--|:--|
+| **10,000,000 rows** | **36 families · M1–M11** | **8.09 GiB · 800 Parquet shards** | **9,949,600 train · 50,400 validation** |
+
+StateBench teaches models to execute finite state programs rather than recall facts. A typed program is executed by a deterministic reference evaluator; only then is it rendered into varied textual surfaces.
+
+## Load it
+
+    from datasets import load_dataset
+
+    ds = load_dataset("aabbdev/StateBench-v1", split="train", streaming=True)
+
+    pretrain = ds.select_columns(["text"])
+    sft = ds.select_columns(["prompt", "completion"])
+    sdft = ds.select_columns(["prompt", "privileged_context"])
+
+## One corpus, three views
+
+| Regime | Student input | Training signal |
+|:--|:--|:--|
+| Pretraining | text | Complete causal document |
+| SFT | prompt | completion |
+| SDFT | prompt | privileged_context, visible only to the teacher |
+
+> **SDFT contract:** privileged_context contains the verified rendered target completion. Never concatenate it into the student prompt. It is designed for TRL SDFTTrainer teacher conditioning.
+
+## What is inside
+
+| Capability | Families |
+|:--|:--|
+| Retain and address | passive retention, capacity, MQAR, multi-token keys, collision resistance |
+| Edit and select | overwrite, erase/reuse, masked edits, selective write/copy, scoped reset |
+| Track dynamics | DFA, Mealy, permutations, monoids, modular counters, finite groups |
+| Adapt online | one-shot binding, remapping, rule switching |
+| Survive long execution | active recurrence, cycles, checkpoints, read/generation pollution |
+
+Difficulty is balanced across levels 0, 2, 4, 5, and 6. Family allocation differs by at most one row across the full corpus.
+
+## Canonical columns
+
+| Column | Meaning |
+|:--|:--|
+| prompt | Conversational student-facing input |
+| completion | Verified assistant target for SFT |
+| privileged_context | Teacher-only verified solution for on-policy SDFT |
+| text | Prompt and completion serialized for causal pretraining |
+| semantic_fingerprint | Deduplication and semantic identity |
+| family, macro, level | Curriculum controls |
+| surface_json, transformation_ids | Rendering provenance |
+
+## Trust and scope
+
+- Every oracle is produced by the reference evaluator, not written by a teacher model.
+- Programs are deduplicated by semantic fingerprint.
+- dataset_manifest.json records generation parameters, distributions, shard sizes, and SHA-256 checksums.
+- The public corpus is training data, not the private StateBench evaluation split.
+
+Generated with **StateBench-v1.0**. The canonical dataset contains both SFT and SDFT fields so no semantic example is duplicated between training formats.