README.md

---
license: other
license_name: mixed-upstream
license_link: https://github.com/OpenBallast/ballast
language:
  - en
pretty_name: Ballast eval sets
size_categories:
  - 10K<n<100K
tags:
  - question-answering
  - factuality
  - hallucination
  - rag
configs:
  - config_name: matrix_probes
    data_files: matrix_probes.parquet
  - config_name: halluc_probes
    data_files: halluc_probes.parquet
---

# Ballast eval sets

The probe sets behind the [Ballast](https://github.com/OpenBallast/ballast)
measurements. Two files:

- **`matrix_probes.parquet` — 50,147 factual questions** (recall), each with a
  gold answer, alias set, 7 type-matched distractors, and — where resolvable —
  a Wikidata subject Q-id linking it to the
  [Ballast T0 corpus](https://huggingface.co/datasets/OpenBallast/ballast-t0)
  (90.5% linked at full corpus). Measured on two model families: Gemma-4
  (E2B/E4B/12B) and Qwen3.5 (0.8B/2B/4B/9B), plus a bf16/fp8/nf4 quantization
  sweep.
- **`halluc_probes.parquet` — 43,137 hallucination probes** (beyond recall),
  98.3% entity-linked, run over seven (model, quant) cells.

## Recall probes (`matrix_probes`)

| source | origin | notes |
|---|---|---|
| `popqa` | PopQA | native subject Q-ids + popularity; the primary instrument |
| `wikidata` | self-generated from T0 | rank-stratified, uncontaminated by construction, regenerable |
| `simpleqa` | SimpleQA | typed distractors from answer_type × topic |
| `nq_open` | Natural Questions (open) | entity-linked via normalized n-gram match |
| `triviaqa` | TriviaQA (rc.nocontext) | linked via wiki-title hints + spans |

## Hallucination probes (`halluc_probes`)

| family | n | what it tests |
|---|---|---|
| `hop2` | 24,320 | self-generated 2-hop compositions from T0 ("where was the director of this film born?") — the answer exists only through a join across two evidence lines, so copy-extraction cannot work. Includes a 3,200-probe adversarial subset (`adversarial=true`) where the competing join path's answer is also present in evidence. |
| `unanswerable` | 12,000 | 9,600 `absent_prop` (the property was never recorded) + 2,400 `false_premise`. Every candidate answer is false; any confident pick is a fabrication. |
| `wikimulti` | 6,000 | 2WikiMultiHopQA compositional + inference questions, both hops entity-linked. |
| `truthfulqa` | 817 | TruthfulQA mc1 as a control family: grounding is expected to leave it unmoved. |

What they showed (full analysis in the
[thesis](https://github.com/OpenBallast/ballast/blob/main/THESIS.md)):
grounding collapses multi-hop hallucination in all 21 family×cell readings
(3–20×, monotone in corpus level), but on unanswerable probes fabrication
*rises* with corpus coverage in all 7 cells — evidence that mentions the
question's entities without containing an answer suppresses exactly the
abstentions that were correct. The TruthfulQA control is imperfect: 3 of 7
cells hold within the 2% margin.

## Schema

Both files: `question_id`, `source`, `question`, `prop`, `subj`, `subj_qid`,
`gold`, `gold_aliases[]`, `distractors[]`, `s_pop`, `pop_decile`.
`halluc_probes` adds `family`, `subfamily`, `adversarial`, `bridge_qid`
(the join entity for 2-hop probes).

## Intended use

Logprob choice scoring: score gold + distractors under the same prompt,
prediction = argmax length-normalized answer logprob, confidence = softmax mass on
argmax, abstain below 0.5. Grounded condition prepends the subject's rendered
evidence block from T0 at a chosen level (2-hop probes prepend both hops'
blocks). Two passes (no corpus / full corpus) compose exactly to every corpus
level via the subject's rank bucket — see the
[methodology](https://github.com/OpenBallast/ballast/blob/main/THESIS.md).

## Contamination stance

Models have seen the upstream benchmarks' sources in pretraining. The `wikidata`
recall probes and the `hop2` compositions are uncontaminated by construction
(generated from the corpus, not the web). The headline quantity
(grounded − ungrounded delta) is contamination-conservative: pretraining
exposure inflates the ungrounded floor, shrinking the measured gain.

## Licenses

Upstream sets keep their upstream licenses (PopQA/TriviaQA/NQ/2WikiMultiHopQA/
TruthfulQA: research-standard open licenses; SimpleQA: MIT). The self-generated
`wikidata` and `hop2` probes are CC0. Check upstream terms before commercial
use of those subsets.