Add/update README.md for Quaestro Bombardment-200 challenge
#3
by ingeekim - opened
README.md
CHANGED
|
@@ -1,3 +1,83 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
license: cc-by-nc-4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
license: cc-by-nc-4.0
|
| 5 |
+
pretty_name: Quaestro Bombardment-200
|
| 6 |
+
size_categories:
|
| 7 |
+
- n<1K
|
| 8 |
+
task_categories:
|
| 9 |
+
- text-generation
|
| 10 |
+
- other
|
| 11 |
+
tags:
|
| 12 |
+
- challenge
|
| 13 |
+
- hallucination
|
| 14 |
+
- fail-closed
|
| 15 |
+
- neuro-symbolic
|
| 16 |
+
- constrained-generation
|
| 17 |
+
- sophus
|
| 18 |
+
- quaestro
|
| 19 |
+
- lie-superalgebra
|
| 20 |
+
dataset_info:
|
| 21 |
+
features:
|
| 22 |
+
- name: id
|
| 23 |
+
dtype: string
|
| 24 |
+
- name: category
|
| 25 |
+
dtype: string
|
| 26 |
+
- name: question
|
| 27 |
+
dtype: string
|
| 28 |
+
configs:
|
| 29 |
+
- config_name: default
|
| 30 |
+
data_files:
|
| 31 |
+
- split: traps
|
| 32 |
+
path: quaestro_200.jsonl
|
| 33 |
---
|
| 34 |
+
|
| 35 |
+
# Challenge: Replicate the SOPHUS Defense
|
| 36 |
+
|
| 37 |
+
**If scaling laws are sufficient to police algebraic hallucination, this corpus is where you prove it.**
|
| 38 |
+
|
| 39 |
+
This release packages the **Quaestro Bombardment-200** challenge: two hundred adversarially structured prompts spanning valid Lie-superalgebra intents and Fail-Closed trap classes, together with the **Key–Value answer targets** extracted under the **SOPHUS** dual-control architecture (typed constitution + residual excision) on a **9B**-class model.
|
| 40 |
+
|
| 41 |
+
SOPHUS closed this split under the author's protocol at **Intent Accuracy 200/200**, with **Fail-Closed** exits on every trap row and **zero** `budget_exhausted` outcomes. The extraction machinery is **not** shipped here. The theoretical and systems methodology is published as Bona Sapiens Technical Report **BS-7**:
|
| 42 |
+
|
| 43 |
+
> [Sophus: Constrained Generation for Verified Algebra](https://doi.org/10.5281/zenodo.21712705)
|
| 44 |
+
> DOI: `10.5281/zenodo.21712705`
|
| 45 |
+
|
| 46 |
+
## The challenge
|
| 47 |
+
|
| 48 |
+
Reproduce, by any means you prefer—frontier parameter count, long context, tool use, RLHF, constitutional AI, or prompt alchemy—**both** of the following on all 200 items:
|
| 49 |
+
|
| 50 |
+
1. **Defense:** constitutionally admissible intents lock; every trap Fail-Closes (no fluent improvisation past the algebraic boundary; no wall-clock budget collapse substituted for typed refusal).
|
| 51 |
+
2. **Extraction:** recover the SOPHUS Key–Value targets for the admissible sector (see `sophus_kv_target.jsonl`).
|
| 52 |
+
|
| 53 |
+
We do not provide the C++ orchestrator, residual forge, or intercept lattice. We provide the **problems**, the **targets**, and a public whitepaper. If scale alone solves verifiable algebra, the leaderboard writes itself.
|
| 54 |
+
|
| 55 |
+
## Files
|
| 56 |
+
|
| 57 |
+
| File | Role |
|
| 58 |
+
|------|------|
|
| 59 |
+
| `quaestro_200.jsonl` | Challenge inputs: `id`, `category`, `question`, `scoring_rubric` |
|
| 60 |
+
| `sophus_kv_target.jsonl` | SOPHUS reference targets: lock/Fail-Closed labels, `quaestro_last_result`, and extracted `kv_segments` |
|
| 61 |
+
| `README.md` | This card |
|
| 62 |
+
|
| 63 |
+
## Rules of engagement
|
| 64 |
+
|
| 65 |
+
- Cite the Zenodo report when you publish attempts, successes, or partial closures.
|
| 66 |
+
- Report protocol clearly: model, decoding, tools, wall-clock budget, and how you score “Fail-Closed” versus soft refusal.
|
| 67 |
+
- Partial credit is welcome; **100/100 defense plus target extraction** is the bar SOPHUS already cleared on this split under the stated harness.
|
| 68 |
+
- This is an author-controlled corpus and scoring surface, not a claim of universal fixed points or external SOTA rankings.
|
| 69 |
+
|
| 70 |
+
## Citation
|
| 71 |
+
|
| 72 |
+
```bibtex
|
| 73 |
+
@misc{kim2026sophus,
|
| 74 |
+
author = {Kim, In-Gee},
|
| 75 |
+
title = {Sophus: Constrained Generation for Verified Algebra},
|
| 76 |
+
year = {2026},
|
| 77 |
+
doi = {10.5281/zenodo.21712705},
|
| 78 |
+
url = {https://doi.org/10.5281/zenodo.21712705},
|
| 79 |
+
note = {Bona Sapiens, Inc. Technical Report (BS-7)},
|
| 80 |
+
}
|
| 81 |
+
```
|
| 82 |
+
|
| 83 |
+
Dataset: `bonasapiensinc/Quaestro-Bombardment-200`.
|