Claim LD07-HQ-C13 (v1.0)
LLM judges can correlate with human evaluations while exhibiting systematic evaluator-specific biases, including position bias and self-preference; task-specific validation is therefore required.
Research opportunities this claim motivates
What this claim leaves open for research. Each is a hypothesis about a gap, not a finding.
- Do LLM judges inflate the quality of AI-written deliberation?LLM judges show position bias and self-preference, so construct-level validation is required.
Provenance and status
- Claim type
- Fact and Interpretation
- Publication status
- Eligible for publication
- Verification state
- Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
- Does not license
- LLM judge as sole unvalidated outcome for social constructs
- Limitations
- Mostly engineering/NLP constructs; social-science construct validation sparse
- Source (as cited)
- Shi et al. IJCNLP-AACL 2025; Panickssery, Bowman & Feng NeurIPS 2024; Calderon et al. ACL 2025
- Source type
- Peer-reviewed conference paper (recorded as “Peer-reviewed conference studies”)
- Institutional position
- Independent academic / mixed academic-industry author affiliations
- Maturity of the evidence
- Rapidly maturing evidence
- Recorded date
- 25 September 2026
- Currentness / correction
- This is the current version. No correction or concern is recorded on its sources.
Sources this version cites
- Shi et al., “Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge,” IJCNLP-AACL 2025 (Peer-reviewed conference paper)
Read the original - Panickssery, Bowman & Feng, “LLM Evaluators Recognize and Favor Their Own Generations,” NeurIPS 2024 (Peer-reviewed conference paper)
Read the original - Calderon, Reichart & Dror, “The Alternative Annotator Test for LLM-as-a-Judge,” ACL 2025 (Peer-reviewed conference paper)
Read the original
Version
This is the current version, v1.0. Earlier versions stay available, and their records never change.
Raw record: every recorded field, with its coded values
All 25 canonical Claim fields, verbatim, with the coded values derived from them. The plain-language labels above are read from these.
- Claim type (coded)
- FACT + INTERPRETATION (as printed: “FACT + INTERPRETATION”)
- Publication status (coded)
- ELIGIBLE (as printed: “ELIGIBLE”)
- Verification state (coded)
- CONTENT_VERIFIED_PRIMARY
- Source type (coded)
- PEER_REVIEWED_CONFERENCE_PAPER
- Maturity (coded)
- RAPIDLY_MATURING
| # | Canonical field | Canonical value (verbatim) | Normalized (Claim Value Vocabulary v1.0.2) |
|---|---|---|---|
| 1 | CLAIM ID | LD07-HQ-C13 | — |
| 2 | PUBLIC-FACING CLAIM | LLM judges can correlate with human evaluations while exhibiting systematic evaluator-specific biases, including position bias and self-preference; task-specific validation is therefore required. | — |
| 3 | CLAIM TYPE | FACT + INTERPRETATION | base_types: FACT, INTERPRETATION |
| 4 | RESEARCH STRAND | B | — |
| 5 | MATURITY | RAPIDLY MATURING | states: RAPIDLY_MATURING |
| 6 | SOURCE | Shi et al. IJCNLP-AACL 2025; Panickssery, Bowman & Feng NeurIPS 2024; Calderon et al. ACL 2025 | source_ids: SRC-LD07-SHI-JUDGING-THE-JUDGES-2025, SRC-LD07-PANICKSSERY-SELF-PREFERENCE-2024, SRC-LD07-CALDERON-ALTERNATIVE-ANNOTATOR-2025 |
| 7 | SOURCE TYPE | Peer-reviewed conference studies | classes: PEER_REVIEWED_CONFERENCE_PAPER |
| 8 | INSTITUTIONAL POSITION | Independent academic / mixed academic-industry author affiliations | — |
| 9 | SCOPE | LLM-as-judge NLP evaluation tasks | — |
| 10 | DESIGN | Large multi-judge benchmark audits; self-recognition experiments; replacement test | — |
| 11 | SAMPLE / DATASET | Shi >150,000 evaluations, 15 judges, MTBench/DevBench; Panickssery controlled recognition/preference tests | — |
| 12 | MODEL / TOOL / VERSION | Multiple families incl GPT-4/Llama 2 and later judges | — |
| 13 | COMPARATOR | Human preferences; swapped presentation; other model outputs | — |
| 14 | MAIN RESULT | Position affects judgments; self-recognition contributes to self-preference; some judge replacement can pass task-specific tests | — |
| 15 | INFERENTIAL TARGET | Validity/reliability of model-based evaluation | — |
| 16 | DOES NOT LICENSE | LLM judge as sole unvalidated outcome for social constructs | — |
| 17 | VERIFICATION STATE | CLEARED | state: CONTENT_VERIFIED_PRIMARY · verified_by: UNKNOWN |
| 18 | LIMITATIONS | Mostly engineering/NLP constructs; social-science construct validation sparse | — |
| 19 | CONTRADICTORY / SUPPORTING SOURCES | 2026 judge studies may update magnitude | — |
| 20 | GEOGRAPHY / LANGUAGE | Mostly English | — |
| 21 | DISTRIBUTIONAL LEDGER | Language/group-specific judge bias remains under-studied | — |
| 22 | R01 / FIELD-MAP LINKS | D4b, D9, R-1 | — |
| 23 | WHAT WOULD CHANGE THIS CLAIM? | Construct-level validation showing biases immaterial under specified protocol | — |
| 24 | PUBLICATION STATUS | ELIGIBLE | state: ELIGIBLE |
| 25 | RECORDED DATE | 2026-09-25 | — |
Field 22 references resolved against Field Map v0.2: D4b (SUB_AREA); D9 (DOMAIN); R-1 (LENS)
The sources as recorded, with their record IDs and coded source types:
Pinned Sources (field 6, version-specific)
- “Shi et al. IJCNLP-AACL 2025” → Shi et al. (2025), “Judging the Judges: A Systematic…” (SRC-LD07-SHI-JUDGING-THE-JUDGES-2025--v1.0) — Shi et al., “Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge,” IJCNLP-AACL 2025; class PEER_REVIEWED_CONFERENCE_PAPER (explicit public provenance pin)
- “Panickssery, Bowman & Feng NeurIPS 2024” → Panickssery et al. (2024), “LLM Evaluators Recognize and…” (SRC-LD07-PANICKSSERY-SELF-PREFERENCE-2024--v1.0) — Panickssery, Bowman & Feng, “LLM Evaluators Recognize and Favor Their Own Generations,” NeurIPS 2024; class PEER_REVIEWED_CONFERENCE_PAPER (explicit public provenance pin)
- “Calderon et al. ACL 2025” → Calderon et al. (2025), “The Alternative Annotator Test…” (SRC-LD07-CALDERON-ALTERNATIVE-ANNOTATOR-2025--v1.0) — Calderon, Reichart & Dror, “The Alternative Annotator Test for LLM-as-a-Judge,” ACL 2025; class PEER_REVIEWED_CONFERENCE_PAPER (explicit public provenance pin)
Cited by Living Debate versions
Cite this page
SocioLens project. (2026, September 25). LLM judges can correlate with human evaluations while exhibiting systematic evaluator-specific biases, including position bias and self-preference; task-specific validation is therefore required [Claim LD07-HQ-C13, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld07-hq-c13/v1.0/
@misc{sociolens_ld07_hq_c13_v1_0,
author = {{SocioLens project}},
title = {LLM judges can correlate with human evaluations while exhibiting systematic evaluator-specific biases, including position bias and self-preference; task-specific validation is therefore required.},
year = {2026},
date = {2026-09-25},
howpublished = {\url{https://sociolens-web.pages.dev/claims/ld07-hq-c13/v1.0/}},
note = {Claim LD07-HQ-C13, version v1.0}
}The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.
Research-of-record provenance
Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 3bb4566dc3b52d50595581cae1788566ad1986f635b7301b48593b7af4e516f9.
Frozen Drive-derived inputs
- R03 — LD-07 Canonical Claim Ledger v1.0 — LOCKED (v1.0) — Drive ID
1xVz_LrQAA8EA1tPhVWIrDgOUU3Zidlg9; revision: NOT_EXPOSED_BY_DRIVE_TEXT_EXPORT; Drive modified 2026-09-29T04:31:14.120Z; block “ENTRY 13 — LD07-HQ-C13” SHA-2563ff0bb8b809271926087a1c2088fd36c90d851d10d1291d42390397c0f3a9d6a; SocioLens — Programme Status & Authority Register v0.6 status “CURRENT.”; frozen fixturefixtures/frozen/r03-ld07-claim-ledger-v1.0-canonical.
Builder decisions (not Drive data)
- VERSION = v1.0 — basis: Claim version taken from the canonical ledger artifact version (R03 — LD-07 Canonical Claim Ledger v1.0); Implementation Architecture Selection §5
- CURRENT_POINTER = true — basis: SocioLens — Programme Status & Authority Register v0.6 status for "R03 — LD-07 Canonical Claim Ledger v1.0 — LOCKED": "CURRENT."
- FIELD_6_SOURCE_REFERENCE = SRC-LD07-SHI-JUDGING-THE-JUDGES-2025--v1.0 — basis: field 6 of ENTRY 13 — LD07-HQ-C13 cites "Shi et al. IJCNLP-AACL 2025"; the frozen field-6 segment of this ClaimVersion's ledger block; Drive-derived identity authority (Issue #39 Source inventory): Batch 04 §4; ledger C13; KEY EVIDENCE.
- FIELD_6_SOURCE_REFERENCE = SRC-LD07-PANICKSSERY-SELF-PREFERENCE-2024--v1.0 — basis: field 6 of ENTRY 13 — LD07-HQ-C13 cites "Panickssery, Bowman & Feng NeurIPS 2024"; the frozen field-6 segment of this ClaimVersion's ledger block; Drive-derived identity authority (Issue #39 Source inventory): Batch 04 §5; ledger C13; KEY EVIDENCE.
- FIELD_6_SOURCE_REFERENCE = SRC-LD07-CALDERON-ALTERNATIVE-ANNOTATOR-2025--v1.0 — basis: field 6 of ENTRY 13 — LD07-HQ-C13 cites "Calderon et al. ACL 2025"; the frozen field-6 segment of this ClaimVersion's ledger block; Drive-derived identity authority (Issue #39 Source inventory): Batch 04 §3; ledger C03/C13; KEY EVIDENCE.
- SOURCE_PROVENANCE_PIN = SRC-LD07-SHI-JUDGING-THE-JUDGES-2025--v1.0 — basis: Issue #5 proof rendering (Architecture Proof Protocol v0.1 AP-08 / Amendment v1.2 §6): explicit version-pinned public provenance of this ClaimVersion — the pinned field-6 Source record "Shi et al. IJCNLP-AACL 2025"; Drive-derived identity authority (Issue #39 Source inventory): Batch 04 §4; ledger C13; KEY EVIDENCE.
- SOURCE_PROVENANCE_PIN = SRC-LD07-PANICKSSERY-SELF-PREFERENCE-2024--v1.0 — basis: Issue #5 proof rendering (Architecture Proof Protocol v0.1 AP-08 / Amendment v1.2 §6): explicit version-pinned public provenance of this ClaimVersion — the pinned field-6 Source record "Panickssery, Bowman & Feng NeurIPS 2024"; Drive-derived identity authority (Issue #39 Source inventory): Batch 04 §5; ledger C13; KEY EVIDENCE.
- SOURCE_PROVENANCE_PIN = SRC-LD07-CALDERON-ALTERNATIVE-ANNOTATOR-2025--v1.0 — basis: Issue #5 proof rendering (Architecture Proof Protocol v0.1 AP-08 / Amendment v1.2 §6): explicit version-pinned public provenance of this ClaimVersion — the pinned field-6 Source record "Calderon et al. ACL 2025"; Drive-derived identity authority (Issue #39 Source inventory): Batch 04 §3; ledger C03/C13; KEY EVIDENCE.