Pith Number

pith:YWL5L3BM

pith:2026:YWL5L3BMRWK7EETKAU55XA7SQG

not attested not anchored not stored refs resolved

Confidence Geometry Reveals Trace-Level Correctness in Large Language Model Reasoning

Ding Liu, Shi-Ju Ran, Shuo Liu

Token-level confidence trajectories in LLMs form low-dimensional geometries that separate correct from incorrect reasoning traces without using question or text content.

arxiv:2605.16824 v1 · 2026-05-16 · cs.LG · cs.CL

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{YWL5L3BMRWK7EETKAU55XA7SQG}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

Using only token-level confidence values, without access to the input question, reasoning text, hidden states, or external verifiers, low-dimensional representations of confidence trajectories separate correct from incorrect reasoning traces.

C2weakest assumption

That the observed low-dimensional separation arises specifically from trace-level correctness rather than from other correlated properties of the generation process such as length, token distribution, or model-specific artifacts, and that this separation generalizes beyond the three evaluated benchmarks without content information.

C3one line summary

Token-level confidence trajectories in LLMs encode a content-agnostic geometry that separates correct and incorrect reasoning traces and supports a lightweight correctness estimator called NeuralConf.

References

32 extracted · 32 resolved · 7 Pith anchors

[1] Chain-of- thought prompting elicits reasoning in large language models, 2022

[2] Large language models are zero-shot reasoners, 2022

[3] Least-to-most prompting enables complex reasoning in large language models, 2023

[4] Self-consistency improves chain of thought reasoning in language models, 2023

[5] Tree of thoughts: Deliberate problem solving with large language models, 2023

Receipt and verification

First computed	2026-05-20T00:03:24.518186Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

c597d5ec2c8d95f2126a053bdb83f281ab06f4ba52ee9c216bdeec9c98193bd1

Aliases

arxiv: 2605.16824 · arxiv_version: 2605.16824v1 · doi: 10.48550/arxiv.2605.16824 · pith_short_12: YWL5L3BMRWK7 · pith_short_16: YWL5L3BMRWK7EETK · pith_short_8: YWL5L3BM

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/YWL5L3BMRWK7EETKAU55XA7SQG \
  | 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: c597d5ec2c8d95f2126a053bdb83f281ab06f4ba52ee9c216bdeec9c98193bd1

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "076a6624ab838da8b08dc2461a7151832c8763030ca32bb3b8b238ddc853a2cc",
    "cross_cats_sorted": [
      "cs.CL"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-16T05:57:00Z",
    "title_canon_sha256": "2f60d239e717102168fbc9d854150d6736dca42974dc85134ff1492f763d4d03"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.16824",
    "kind": "arxiv",
    "version": 1
  }
}