---
dataset_info:
  features:
  - name: prompt
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
  - name: chosen
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
  - name: rejected
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
  - name: preference_id
    dtype: string
  splits:
  - name: train
    num_bytes: 56332497
    num_examples: 9000
  - name: validation
    num_bytes: 2758935
    num_examples: 435
  - name: test
    num_bytes: 2737900
    num_examples: 435
  download_size: 37334938
  dataset_size: 61829332
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

## Split swap, 2026-08-20

`validation` and `test` were exchanged in this revision. `train` is unchanged.

**Why.** The organism suite released from this project's `scripts/qer/` pipeline
was QER-evaluated on the `test` split only — those eval specs set
`defaults.trigger.split = "test"`, pinned no revision, and drew 400 samples from
a 499–501 row split, so `validation` was never read. Those readings informed the
published targets and per-variant learning rates, which made the old `test` a
**selection** split wearing a held-out split's name.

**What this revision contains**, measured against the previous one:

| split | in common with the prompts that suite read |
|---|---|
| `validation` | **435 of 435** — the prompts selection already consumed |
| `test` | **0 of 435** — never read by that suite |

**Correction.** An earlier version of this note gave those two figures the other
way round. The swap performed was always the one described here; only the note
was wrong.

**Scope — read this before citing `test` as held out.** "Never read" is specific
to the `scripts/qer/` suite. A separate campaign (the published `automo-*`
organisms) evaluated this family with a reader that topped a short split up from
its sibling, drawing 1000 prompts as 501 from `test` plus 499 from `validation`
— the entire pool. Every prompt in this revision's `test` sat in the selection
pool behind those models. `test` is held out with respect to the `scripts/qer/`
suite, not unconditionally.

**Reproducibility.** Every earlier revision remains reachable by sha, and
consumers that pin a revision are unaffected until they re-pin.