pith:CRNS7LGM
Beyond Fixed Benchmarks and Worst-Case Attacks: Dynamic Boundary Evaluation for Language Models
Evaluating language models at each model's 0.5 success probability boundary reveals capability gaps that fixed benchmarks miss.
arxiv:2605.06213 v2 · 2026-05-07 · cs.AI
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{CRNS7LGMCSHJZ6IM7BF5WCESGL}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
We argue that the most informative evaluation signal lies at the boundary, where the per-prompt pass probability is near 0.5 under random-sampling decoding, and propose Dynamic Boundary Evaluation (DBE), which actively locates each model's boundary and places it on a globally comparable difficulty scale.
That the per-item difficulty labels validated across 9 reference LLMs will allow accurate placement of new models on the same scale, and that the boundary at 0.5 probability is indeed the most informative point for evaluation.
Dynamic Boundary Evaluation adaptively identifies each LLM's performance boundary on a shared difficulty scale using a calibrated item bank and a search algorithm.
Receipt and verification
| First computed | 2026-05-27T02:05:21.280792Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
145b2faccc148e9cf90cf84bdb089232dda0f6a930a35d3aceefc0edcc31e564
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/CRNS7LGMCSHJZ6IM7BF5WCESGL \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 145b2faccc148e9cf90cf84bdb089232dda0f6a930a35d3aceefc0edcc31e564
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "8fc4fa04fdb801a1bec4f8acff3dfb7c2bcd45a5d807a3ca47cfc98e1e544b09",
"cross_cats_sorted": [],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.AI",
"submitted_at": "2026-05-07T13:15:31Z",
"title_canon_sha256": "a710aa693ed0ba3acae331e88927f7353b96133d743204884b058c359ec47750"
},
"schema_version": "1.0",
"source": {
"id": "2605.06213",
"kind": "arxiv",
"version": 2
}
}