Rewrite card in transparent uniform format

README-only documentation update: quick start, provenance, evaluation conditions, reproducibility, integrity identifiers, and limitations.

README.md

@@ -12,67 +12,129 @@ tags:
   - 4096-context
 ---
 
-# Qwen3.8-Kimi-Fable Distillation
+# Qwen3.8–Kimi–Fable Distillation
 
-A curated SFT dataset composed of three teacher datasets, mixed and filtered for training compact agentic models like Nanbeige4.2-3B.
+A 13,700-row curated SFT mixture of reasoning, coding, instruction, and tool-use traces from three teacher datasets.
 
-## Composition
+## Quick start
 
-Three teacher datasets, curated and mixed:
+```python
+from datasets import load_dataset
 
-- **[Qwen3.8-Max Distillation 50K](https://huggingface.co/datasets/r0b0tlab/qwen3.8-max-distillation-50k)** — ~10,000 rows distilled from Alibaba's Qwen3.8-Max Preview via Model Studio API. Covers math (33%), code (28%), reasoning (24%), and instruction-following (15%).
+dataset = load_dataset(
+    "json",
+    data_files="hf://datasets/j0no12/qwen38-kimi-fable-distillation/curated.jsonl",
+    split="train",
+)
 
-- **[Fable 5 Coding & Debugging Traces](https://huggingface.co/datasets/greghavens/fable-5-coding-and-debugging-traces)** — ~2,500 verified coding and debugging trajectories from Claude Fable 5, generated with moonshiner. Focus on tool-calling, multi-step planning, and build/test/fix loops. License: CC BY 4.0.
+print(dataset[0]["source"])
+print(dataset[0]["messages"])
+```
 
-- **[Kimi K3 Coding & Debugging Traces](https://huggingface.co/datasets/greghavens/kimi-k3-coding-and-debugging-traces)** — ~1,200 verified coding, tool-use, and instruction-following traces from Kimi K3. Focus on parallel tool calls, state tracking, and verification-driven completion. License: CC BY 4.0.
+## Dataset summary
 
-**Total: 13,700 training rows**
+| Property | Value |
+|---|---:|
+| Rows | 13,700 |
+| Split | Training collection only |
+| Format | JSON Lines |
+| Context policy | At most 4,096 tokens with the Nanbeige4.2-3B tokenizer |
+| Main domains | Math, code, reasoning, instruction following, tool use |
+| Intended use | SFT for compact agentic models |
 
-## Context Cap
+## Composition
 
-All examples are constrained to a **4096-token context window** (measured with the Nanbeige4.2-3B tokenizer). The raw content is capped at ~3800 tokens to leave headroom for chat-template tokens (`<im_start>`, `<im_end>`, `<eos>`, BOS).
+| `source` | Rows | Share |
+|---|---:|---:|
+| `qwen3.8-max` | 10,000 | 73.0% |
+| `fable-5-trace` | 2,500 | 18.2% |
+| `kimi-k3-trace` | 1,200 | 8.8% |
+| **Total** | **13,700** | **100%** |
+
+Recorded domain counts are 3,912 math, 2,916 reasoning, 2,650 code, 522 instruction, and 3,700 rows without a domain label.
 
 ## Schema
 
-Each row is a JSON object with:
+| Field | Type | Description |
+|---|---|---|
+| `messages` | list of objects | Chat messages with `role` and `content` |
+| `source` | string | One of the three source labels above |
+| `domain` | string or null | Domain tag when available |
+| `difficulty` | source-dependent / optional | Difficulty tag when supplied by the source |
 
-- `messages` — list of `{role, content}` dicts (system/user/assistant)
-- `source` — one of `"qwen3.8-max"`, `"fable-5-trace"`, `"kimi-k3-trace"`
-- `domain` — domain tag (when available)
-- `difficulty` — difficulty tag (when available)
+Example access pattern:
 
-## Quality Filters
+```python
+row = dataset[0]
+for message in row["messages"]:
+    print(f"{message['role']}: {message['content']}")
+```
 
-Applied during curation:
+## Source lineage
 
-- Quality score >= 7.5 on Qwen3.8 traces (median 9.0, mean 8.6)
-- No refusals, no empty responses
-- Near-deduplication via SHA-256 hashing
-- Context length <= 3800 raw tokens
-- Minimum response length: 30 tokens (non-trace), 15 tokens (trace)
+| Source dataset | Role | Selected rows | Terms recorded by this card |
+|---|---|---:|---|
+| [`r0b0tlab/qwen3.8-max-distillation-50k`](https://huggingface.co/datasets/r0b0tlab/qwen3.8-max-distillation-50k) | General reasoning, math, code, instruction | 10,000 | See the source dataset and Alibaba Cloud Model Studio terms |
+| [`greghavens/fable-5-coding-and-debugging-traces`](https://huggingface.co/datasets/greghavens/fable-5-coding-and-debugging-traces) | Coding/debugging trajectories | 2,500 | CC-BY-4.0 |
+| [`greghavens/kimi-k3-coding-and-debugging-traces`](https://huggingface.co/datasets/greghavens/kimi-k3-coding-and-debugging-traces) | Coding/tool-use trajectories | 1,200 | CC-BY-4.0 |
 
-## License
+## Construction story
+
+The curation report records 61,150 input rows: 44,796 Qwen, 12,448 Fable, and 3,906 Kimi. Quality and length filtering retained 40,689 candidate rows before source caps were applied. The final deterministic mixture contains the 13,700 rows shown above.
+
+Recorded filtering outcomes:
+
+| Source | Main recorded exclusions |
+|---|---|
+| Qwen | 7,413 low-quality scores, 711 over-length contexts, 132 refusals, 6 short assistant responses |
+| Fable | 8,196 short assistant responses, 1,380 over-length contexts, 54 refusals |
+| Kimi | 2,386 short assistant responses, 174 near-duplicates, 9 refusals |
 
-The component datasets carry different licenses:
+The context check used the Nanbeige4.2-3B tokenizer. User messages range from 5 to 3,027 tokens (median 75), and assistant messages range from 15 to 3,625 tokens (median 259). The raw-content target was approximately 3,800 tokens to leave space for chat-template tokens inside a 4,096-token window.
 
-- Qwen3.8 traces: see original source terms (Alibaba Cloud Model Studio)
-- Fable 5 traces: CC BY 4.0
-- Kimi K3 traces: CC BY 4.0
+## Quality checks
 
-This curated collection is made available for research and personal use under the same terms as its components.
+- Qwen rows required a quality score of at least 7.5; the prior card reports median 9.0 and mean 8.6.
+- Refusals and empty/very short completions were filtered.
+- Context length was checked with the target tokenizer.
+- Kimi processing records a near-duplicate filter; the previous card also described SHA-256 hashing, which should be understood as exact content hashing unless a separate similarity method is supplied.
+- Source caps fix the final mixture at 10,000 / 2,500 / 1,200 rows.
 
-## Intended Use
+## Contamination screening
 
-Supervised fine-tuning of compact agentic models, particularly those with novel architectures (Looped Transformer, etc.) that benefit from diverse, high-quality distilled traces.
+No benchmark-contamination report is included. Hash-based deduplication and source-specific near-duplicate filtering do not prove independence from HumanEval, MBPP, GSM8K, MATH, or other public evaluations. Run task-specific decontamination before reporting benchmark results from models trained on this collection.
+
+## Known limitations
+
+- All responses are teacher-generated and may contain incorrect reasoning, insecure code, fabricated tool results, or stylistic artifacts.
+- There is no held-out validation or test split.
+- The 4,096-token guarantee is tokenizer-specific and may not hold under another tokenizer or chat template.
+- Domain labels are missing for 3,700 rows, and optional metadata is source-dependent.
+- The mixture intentionally oversamples Qwen relative to the two trajectory sources.
+- Component licensing is mixed; no single permissive license is asserted for every row.
+
+## Files and integrity
+
+| File | Size | Integrity identifier |
+|---|---:|---|
+| `curated.jsonl` | 64,322,634 bytes | SHA-256 `78d2d52229b49e057fad41cf3ec66a60e5be7914997b55938cb3ba863365bb8a` |
+| `report.json` | 1,303 bytes | Git blob `095b1324ac986777402e3cb58a1fbfa792a67fd5` |
+| Repository revision documented by this card update | — | `95156b7351b5a7689b5aa6eee037337d3ee5293f` |
 
 ## Citation
 
 ```bibtex
-@misc{j0no2026qwen38kumifable-distillation,
-  title        = {Qwen3.8-Kimi-Fable Distillation},
-  author       = {j0no},
+@misc{j0no12_qwen38_kimi_fable_2026,
+  author       = {j0no12},
+  title        = {Qwen3.8--Kimi--Fable Distillation},
   year         = {2026},
   publisher    = {Hugging Face},
-  howpublished = {\url{https://huggingface.co/datasets/j0no12/qwen38-kimi-fable-distillation}},
+  howpublished = {\url{https://huggingface.co/datasets/j0no12/qwen38-kimi-fable-distillation}}
 }
 ```
+
+Please also cite the three source datasets.
+
+## License
+
+Mixed / other. Fable and Kimi components are recorded as CC-BY-4.0; Qwen-derived traces remain subject to their source dataset and Alibaba Cloud Model Studio terms. This card does not grant rights beyond those upstream terms.