In a conditionally accepted AJPS manuscript re-running 14 published political-science annotation pipelines, substituting LLM-generated labels changed the substantive statistical conclusion in approximately 35–42% of estimates across the tested prompt/in-context-learning settings.

SocioLens project,

Claim LD07-HQ-C16 (v1.0)

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication, with qualifiers: conditional acceptance, corpus
Verification state: Checked against the primary source; cleared with a source-specific qualifier (recorded as “CLEARED WITH QUALIFIER”). 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: Claim that 35–42% of political-science findings are wrong; claim that fresh experts would agree with original labels; generalization beyond the selected 14-study corpus
Limitations: Conditionally accepted, not yet version-of-record; selected corpus; exact model count internally inconsistent in current PDF
Source (as cited): Eddie Yang, Zoey Wang, Carl Zhou, Yaosheng Xu, “Data Annotation with Large Language Models: Lessons from a Large Empirical Evaluation”
Source type: Working paper (recorded as “Conditionally accepted journal manuscript / current pre-publication PDF”)
Institutional position: Independent academic
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

Yang, Wang, Zhou & Xu, “Data Annotation with Large Language Models: Lessons from a Large Empirical Evaluation,” conditionally accepted at AJPS; pre-publication manuscript (Working paper)
Read the original

Version

This is the current version, v1.0. Earlier versions stay available, and their records never change.

Versions: v1.0 · current

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): ESTIMATE (as printed: “ESTIMATE”)
Publication status (coded): ELIGIBLE (CONDITIONAL_ACCEPTANCE, CORPUS) (as printed: “ELIGIBLE WITH CONDITIONAL-ACCEPTANCE / CORPUS QUALIFIER”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY (SOURCE_SPECIFIC)
Source type (coded): WORKING_PAPER
Maturity (coded): RAPIDLY_MATURING

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD07-HQ-C16	—
2	PUBLIC-FACING CLAIM	In a conditionally accepted AJPS manuscript re-running 14 published political-science annotation pipelines, substituting LLM-generated labels changed the substantive statistical conclusion in approximately 35–42% of estimates across the tested prompt/in-context-learning settings.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	A — Annotation/coding/classification	—
5	MATURITY	RAPIDLY MATURING	states: RAPIDLY_MATURING
6	SOURCE	Eddie Yang, Zoey Wang, Carl Zhou, Yaosheng Xu, “Data Annotation with Large Language Models: Lessons from a Large Empirical Evaluation”	source_ids: SRC-LD07-YANG-DATA-ANNOTATION-EVALUATION
7	SOURCE TYPE	Conditionally accepted journal manuscript / current pre-publication PDF	classes: WORKING_PAPER
8	INSTITUTIONAL POSITION	Independent academic	—
9	SCOPE	14 political-science studies published 2018–2025 in leading journals, using discrete annotations in downstream statistical inference	—
10	DESIGN	Multi-study re-annotation and downstream re-estimation	—
11	SAMPLE / DATASET	More than three million original annotations across included studies; exact generated-annotation total version-sensitive	—
12	MODEL / TOOL / VERSION	Multiple contemporary LLMs; current manuscript contains an internal 15/16-model-count inconsistency, so no exact count is public-canonical	—
13	COMPARATOR	Original study labels and original published statistical conclusions	—
14	MAIN RESULT	LLM-derived estimates changed the substantive statistical conclusion in approximately 35–42% of cases across tested configurations; LLM-pair disagreement also produced sign/significance differences	—
15	INFERENTIAL TARGET	Preservation of original published downstream conclusions when labels are substituted	—
16	DOES NOT LICENSE	Claim that 35–42% of political-science findings are wrong; claim that fresh experts would agree with original labels; generalization beyond the selected 14-study corpus	—
17	VERIFICATION STATE	CLEARED WITH QUALIFIER	state: CONTENT_VERIFIED_PRIMARY · qualifiers: SOURCE_SPECIFIC · verified_by: UNKNOWN
18	LIMITATIONS	Conditionally accepted, not yet version-of-record; selected corpus; exact model count internally inconsistent in current PDF	—
19	CONTRADICTORY / SUPPORTING SOURCES	Baumann independently supports the downstream-failure class; Nakamura shows narrow fresh-expert equivalence under a different estimand	—
20	GEOGRAPHY / LANGUAGE	Political-science text tasks, primarily English	—
21	DISTRIBUTIONAL LEDGER	No broad subgroup-validity guarantee	—
22	R01 / FIELD-MAP LINKS	D9, R-1, R-2	—
23	WHAT WOULD CHANGE THIS CLAIM?	Frozen accepted/VOR manuscript materially revising the 35–42% result, or independent exact replication showing substantially greater preservation	—
24	PUBLICATION STATUS	ELIGIBLE WITH CONDITIONAL-ACCEPTANCE / CORPUS QUALIFIER	state: ELIGIBLE · qualifiers: CONDITIONAL_ACCEPTANCE, CORPUS
25	RECORDED DATE	2026-09-25	—

Field 22 references resolved against Field Map v0.2: D9 (DOMAIN); R-1 (LENS); R-2 (LENS)

The sources as recorded, with their record IDs and coded source types:

Pinned Sources (field 6, version-specific)

“Eddie Yang, Zoey Wang, Carl Zhou, Yaosheng Xu, “Data Annotation with Large Language Models: Lessons from a Large Empirical Evaluation”” → Yang et al., “Data Annotation with Large Language Models…” (SRC-LD07-YANG-DATA-ANNOTATION-EVALUATION--v1.0) — Yang, Wang, Zhou & Xu, “Data Annotation with Large Language Models: Lessons from a Large Empirical Evaluation,” conditionally accepted at AJPS; pre-publication manuscript; class WORKING_PAPER (explicit public provenance pin)

Cited by Living Debate versions

AI in the research apparatus (v0.1) LD-07
AI in the research apparatus (v0.2) LD-07

Cite this page

SocioLens project. (2026, September 25). In a conditionally accepted AJPS manuscript re-running 14 published political-science annotation pipelines, substituting LLM-generated labels changed the substantive statistical conclusion in approximately 35–42% of estimates across the tested prompt/in-context-learning settings [Claim LD07-HQ-C16, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld07-hq-c16/v1.0/

@misc{sociolens_ld07_hq_c16_v1_0,
  author = {{SocioLens project}},
  title = {In a conditionally accepted AJPS manuscript re-running 14 published political-science annotation pipelines, substituting LLM-generated labels changed the substantive statistical conclusion in approximately 35–42\% of estimates across the tested prompt/in-context-learning settings.},
  year = {2026},
  date = {2026-09-25},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld07-hq-c16/v1.0/}},
  note = {Claim LD07-HQ-C16, 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 869c683f6541617c3647cd82026b337aa7c3c2ff63679feb49e500adbd949a02.

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 16 — LD07-HQ-C16” SHA-256 70c22940247ca3b84bece80284654946b4e89ca05011a100f63ce85e0bfc2fd7; SocioLens — Programme Status & Authority Register v0.6 status “CURRENT.”; frozen fixture fixtures/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_17_CONTEXT_QUALIFIERS = SOURCE_SPECIFIC — basis: R03 — LD-07 Canonical Claim Ledger v1.0 — LOCKED (v1.0, Drive 1xVz_LrQAA8EA1tPhVWIrDgOUU3Zidlg9)
FIELD_6_SOURCE_REFERENCE = SRC-LD07-YANG-DATA-ANNOTATION-EVALUATION--v1.0 — basis: field 6 of ENTRY 16 — LD07-HQ-C16 cites "Eddie Yang, Zoey Wang, Carl Zhou, Yaosheng Xu, “Data Annotation with Large Language Models: Lessons from a Large Empirical Evaluation”"; the frozen field-6 segment of this ClaimVersion's ledger block; Drive-derived identity authority (Issue #39 Source inventory): Batch 01 §3; ledger C16 fields 6/7/8; KEY EVIDENCE.
SOURCE_PROVENANCE_PIN = SRC-LD07-YANG-DATA-ANNOTATION-EVALUATION--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 "Eddie Yang, Zoey Wang, Carl Zhou, Yaosheng Xu, “Data Annotation with Large Language Models: Lessons from a Large Empirical Evaluation”"; Drive-derived identity authority (Issue #39 Source inventory): Batch 01 §3; ledger C16 fields 6/7/8; KEY EVIDENCE.