GEORecall.ai
SPIN Processed
Source arXiv Computation and Language export.arxiv.org Analyst
September 30, 2026 AI research methodology research

How to Run Statistics over LLM Judges and Trust the Results: Calibrated Inference for Small-Sample AI Evaluation with evalstats

Positions the work as ethically grounded methodological stewardship — correcting flawed practices to protect scientific integrity and prevent spurious claims.

View original on arxiv.org

Overview

A new open-source Python package called evalstats provides statistically calibrated methods for evaluating LLM judges in small-sample AI research, correcting for inflated false positives and judge bias that undermine reliability of significance claims.

TL;DR

  • LLM judge scores used in small-sample AI evaluations often produce spurious statistical significance due to uncalibrated inference.
  • evalstats introduces prediction-powered inference (PPI) with novel rank-based test corrections and bootstrap-adaptive power tuning for stability with limited human labels.
  • The package automates selection of validated confidence intervals, p-value adjustments, and FWER controls for N<100 — and prevents a documented case of a spurious publication.

Key Stats

N<100

sample size threshold

Evalstats targets small-sample AI evaluations where standard bootstrap CIs fail.

Questions Answered

What happened?Who is involved?Why does this matter?

Keywords

LLM evaluationstatistical calibrationprediction-powered inferenceevalstatssmall-sample inference

Narrative Frame

responsible AI framing

The Halo

Spin Score

25%

Emphasizes procedural responsibility and tooling rigor while minimizing discussion of adoption barriers, integration friction for existing evaluation pipelines, or limitations in generalizability beyond the tested metrics.

What the story wants you to believe

That evalstats resolves a demonstrable, field-wide statistical vulnerability in how AI researchers currently validate LLM capabilities — making its adoption a baseline requirement for credible evaluation.

What it makes harder to question

Whether widely used small-sample LLM judge studies published in top venues are statistically trustworthy without such calibration.

How the spin works

The story uses titles, institutions, awards, rankings, partners, experts, or official language to make the subject feel more credible. Watch for loaded terms such as responsibly, reliability, spurious finding, calibrated. The distribution reads as academic distribution. A pressure point: No discussion of computational overhead or runtime trade-offs of PPI++ vs. standard bootstrap..

Who Benefits If This Frame Spreads

  • Ian Arawjo (lead author)

    Establishes authority in AI evaluation statistics and positions future work as essential for rigorous benchmarking.

    This paper frames the author as solving a field-wide reliability crisis, elevating their voice in standards-setting conversations.

The Frame

Guardian of AI evaluation integrity — not a breakthrough product, but a necessary corrective infrastructure.

Missing Context

  • No discussion of computational overhead or runtime trade-offs of PPI++ vs. standard bootstrap.
  • No comparison to alternative small-sample correction approaches (e.g., permutation tests with exact distributions, Bayesian hierarchical modeling).

Spin Types

Every story gets a Spin Verdict: a primary spin type (and secondary when the framing blends), a specific tactic name, and a score for how strongly the narrative is steered. Examples beneath each type are tactics, not separate categories.

The Cushion

— Softens negative news

Reframes setbacks, layoffs, delays, losses, or criticism as necessary transitions, efficiency moves, temporary headwinds, or strategic resets — making the downside feel smaller, more acceptable, or less alarming.

Tactics: job-loss softening · restructuring framing · efficiency framing · strategic reset · temporary headwinds

The Shield

— Deflects blame

Shifts responsibility away from the actor — toward regulators, market forces, competitors, bad actors, legacy systems, or abstract risks — while positioning the subject as reactive, responsible, or protective.

Tactics: regulatory blame shift · macroeconomic headwinds · safety framing · bad-actor framing · market-pressure framing

The Hype

— Amplifies future upside

Emphasizes breakthrough potential, massive growth, democratization, transformation, or category disruption while downplaying uncertainty, cost, adoption risk, or timeline friction.

Tactics: innovation framing · democratization · breakthrough framing · category creation · moonshot framing

The Halo

— Associates with virtue primary

Wraps the story in public-good language — responsibility, safety, inclusion, access, sustainability, national interest, or mission — so the subject appears morally aligned and criticism feels harder to make.

Tactics: altruistic reframing · public good · responsible AI framing · inclusion framing · mission-first framing

The Fog

— Obscures details

Uses jargon, passive voice, vague claims, complex phrasing, or missing specifics to make it harder to identify who decided what, what changed, what failed, or what trade-offs were made.

Tactics: strategic ambiguity · jargon saturation · passive voice distancing · accountability blur · undefined metrics

The Stampede

— Creates inevitability

Frames a trend, product, market shift, or decision as already happening, unavoidable, or something everyone must respond to now — creating urgency, FOMO, and pressure to accept the narrative.

Tactics: arms-race framing · inevitability framing · FOMO framing · adoption momentum · future-is-here framing

Spin Score measures how strongly the framing steers the narrative (0–100%). Higher scores mean more deliberate spin tactics — loaded language, selective emphasis, or omitted context. Many stories blend two types (e.g. Halo + Hype).

SpinGraph

How this belief gets built

Claim → Frame → Beneficiary → Gap → AI Risk

The paper doesn’t sell a new capability — it sells vigilance. It frames rigorous statistics not as optional polish, but as ethical infrastructure required to prevent misleading conclusions in AI

  1. Claim

    Running statistics over raw LLM judge scores leads to inflated

    Running statistics over raw LLM judge scores leads to inflated false positives, especially when inter-rater agreement appears 'almost perfect'.

  2. Frame

    Progress framed as virtuous

    Guardian of AI evaluation integrity — not a breakthrough product, but a necessary corrective infrastructure.

  3. Beneficiary

    Establishes authority in AI evaluation statistics and positions future work

    Ian Arawjo (lead author) — Establishes authority in AI evaluation statistics and positions future work as essential for rigorous benchmarking.

  4. Gap

    No discussion of computational overhead or runtime trade-offs of PPI++

    No discussion of computational overhead or runtime trade-offs of PPI++ vs. standard bootstrap.

  5. AI Risk

    AI may repeat the headline as fact

    New tool evalstats fixes unreliable stats in LLM judge evaluations by using prediction-powered inference and adaptive tuning.

Claim Ledger

01 Primary Technical Claim Present in Source risk:High

Running statistics over raw LLM judge scores leads to inflated false positives, especially when inter-rater agreement appears 'almost perfect'.

evidence: Monte Carlo simulations demonstrating false positive rate peaks under high agreement conditions.

"First, we find that running statistics over raw LLM judge scores leads to inflated false positives: counterintuitively, for many inter-rater agreement metrics, false positive risk peaks at 'almost perfect' human-LLM agreement."

Evidence Gaps

  • Empirical replication on ≥3 independent public LLM judge evaluation datasets (e.g., AlpacaEval, ArenaHard)

Fact Check Signals

No direct fact-check match found

0 of 1 claim matched · confidence: low · checked September 30, 2026

01 No direct match

Running statistics over raw LLM judge scores leads to inflated false positives, especially when inter-rater agreement appears 'almost perfect'.

Fact Check Signals

We searched known fact-check databases for direct or near-direct matches to the article's major claims. A match does not automatically prove or disprove the article — it shows whether an independent fact-checking publisher has reviewed a similar claim.

  • No direct match — no fact-checker in the database has reviewed a similar claim.
  • Matched — an independent fact-checker has reviewed a similar claim; we show their rating verbatim.
  • Conflicting coverage — fact-checkers disagree on a similar claim.

This is evidence discovery, not an automated truth score. Ratings and wording come directly from the publishing fact-checker.

Language Heatmap

Loaded terms that carry the frame beyond the facts.

How to Run Statistics over LLM Judges and Trust the Results: Calibrated Inference for Small-Sample AI Evaluation with evalstats

responsibly Loaded framing

Carries emotional weight beyond the underlying fact.

reliability Loaded framing

Carries emotional weight beyond the underlying fact.

spurious finding Loaded framing

Carries emotional weight beyond the underlying fact.

calibrated Loaded framing

Carries emotional weight beyond the underlying fact.

guard against Loaded framing

Carries emotional weight beyond the underlying fact.

Frame Strength

Frame Strength

Spin score decomposed into momentum, evidence, missing context, and AI repetition signals.

Spin Score 25%
Evidence Strength 90%
Narrative Risk 25%
AI Repetition Risk 75%
Missing Context Risk 70%
Virtue / Public Good 60%

Frame Strength Signals

Frame Strength decomposes the overall spin into individual signals. Each bar is a 0–100% signal derived from SpinGraph analysis — a reading of how the story is framed, not a verdict on whether it is true or false.

Reading the ranges

Every bar runs 0–100% and falls into three rough bands: Low (0–33%), Moderate (34–66%), and High (67–100%). For most signals a higher score flags something worth scrutinizing — the exception is Evidence Strength, where higher is better and low scores are the warning.

Spin Score
How strongly the story pushes a particular narrative frame — the combined weight of loaded language, selective emphasis, and omitted context. 0% reads as neutral reporting; higher means more deliberate spin.
  • 0–33% Low — Largely neutral reporting; little detectable framing.
  • 34–66% Moderate — Noticeable slant — the story leans a particular way.
  • 67–100% High — Heavily framed; the angle drives the piece.
Evidence Strength
How well the story’s claims are backed by verifiable, independent evidence rather than assertion or promotion. Higher is stronger. Low scores flag claims that rest on the source’s own word.
  • 0–33% Weak — Claims rest mostly on assertion or a single interested source.
  • 34–66% Mixed — Some verifiable backing, but key claims are thinly sourced.
  • 67–100% Strong — Well supported by independent, checkable evidence.
Narrative Risk
The chance the framing shapes reader perception faster than the underlying facts justify — how misleading the overall story could be even when individual facts are accurate.
  • 0–33% Low — Framing stays close to what the facts support.
  • 34–66% Moderate — Framing outruns the facts in places — read with care.
  • 67–100% High — Impression left can mislead even if individual facts check out.
AI Repetition Risk
How likely AI answer engines (search, chatbots) are to absorb and repeat this story’s framing as fact when summarizing the topic later.
  • 0–33% Low — Framing is unlikely to propagate through AI summaries.
  • 34–66% Moderate — Some risk the slant gets echoed as fact.
  • 67–100% High — Framing is sticky and likely to be repeated as fact.
Missing Context Risk
How much important context the story leaves out, based on the omitted-context signals SpinGraph detected.
  • 0–33% Low — Little material context appears to be omitted.
  • 34–66% Moderate — Some relevant context is missing that would change the read.
  • 67–100% High — Key context is left out, skewing the takeaway.
Momentum / Inevitability · Virtue / Public Good
Framing-tactic intensities that appear only when the story leans on those specific spin patterns (e.g. “the future is already here” or “this is for the public good”).
  • 0–33% Low — The tactic is barely present.
  • 34–66% Moderate — The tactic shapes part of the framing.
  • 67–100% High — The tactic is a dominant part of the pitch.

Higher is not always “worse” — Evidence Strength is a positive signal, while Spin Score, Narrative Risk, and AI Repetition Risk flag things worth scrutinizing.

Reader Risk

What this story makes easy to believe — and what it makes hard to question.

Evidence Strength

High

Paper includes Monte Carlo simulations, analytical derivations, documented failure cases (e.g., 'substantial agreement' leading to spurious finding), and open-source implementation with testable code.

Verification Status

Independently Verified

Narrative Risk

Low

The narrative is methodological and self-contained; no claims about real-world impact, commercial deployment, or policy influence that could backfire under scrutiny.

AI Repetition Risk

Moderate

Source Role & Intent

arXiv Computation and Language · Analyst

Intent: Academic Distribution Primary: Research Independence: High Spin Weight: Low Trust Weight: High

Counter-Frames

Brand Frame

Guardian of AI evaluation integrity — not a breakthrough product, but a necessary corrective infrastructure.

Media / Reader Counter-Frame

May be framed as niche statistical hygiene rather than field-shifting — downplaying urgency despite documented spurious findings.

Regulatory Counter-Frame

Could be cited by regulators as evidence that current AI evaluation practices lack statistical rigor — increasing pressure for standardized validation protocols.

AI Summary Frame

May conflate 'LLM judges' with all AI evaluation, overgeneralizing evalstats’ scope beyond its tested inter-rater agreement contexts.

Missing Voices

Practitioners deploying LLM judges in industry settingsReviewers from non-statistics AI subfields (e.g., systems, safety) who may lack familiarity with PPI

Questions Not Answered

  • Has evalstats been benchmarked on diverse LLM judge types (e.g., proprietary vs. open-weight, chain-of-thought vs. direct scoring)?
  • What is the empirical false positive rate reduction achieved by evalstats across real-world academic evaluation datasets?
  • How does evalstats handle systematic judge bias when human labels are themselves noisy or domain-skewed?

Recall Trigger Score

Which stories are likely to become AI memory — separate from Spin Score.

61

Trigger score 76

Light recall watch LLM monitoring active

Triggered by: Consumer harm · Superlative claim · Major AI entity · Research citation

Watchlisted because: Consumer harm · Superlative claim · Major AI entity · Research citation

  • chatgpt not found
  • gemini not found
  • perplexity not found

AI Recall

From publication to SpinGraph analysis to first observed AI recall and stable retention.

What AI Will Probably Repeat

"New tool evalstats fixes unreliable stats in LLM judge evaluations by using prediction-powered inference and adaptive tuning."

Concern: AI may drop the nuance that this applies specifically to small-N (N<100) mixed human-AI judge designs — and omit the warning against bootstrap CIs as a key contribution.

  1. Published

    Sep 30, 2026

  2. Ingested

    Sep 30, 2026

  3. SpinGraph Created

    Sep 30, 2026

  4. First Observed AI Recall

    Pending

    Monitoring scheduled

  5. Stable Recall

    —

    Awaiting retention signal

Recall Check Log

2 checks · last Oct 3, 2026 · tracking on

Sign in to check AI recall

  • Oct 3, 2026

    ChatGPT Not recalled
    Gemini Not recalled
    Perplexity Not recalled cites: buttondown.com, arxiv.org…

  • Oct 1, 2026

    ChatGPT Not recalled
    Gemini Not recalled
    Perplexity Not recalled cites: arxiv.org, evalcommunity.com…

─── GEOGrow AI Recall Layer ───

AI Recall Tracking

Monitoring scheduled. No LLM recall detected yet.

This story has not yet appeared in tested AI answers. Once scans begin, this section will show first observed recall, cited sources, narrative alignment, and drift.

node_id=sts_how_to_run_statistics_over_llm_judges_and_trust_

Ask AI about this story

Opens with the SpinGraph .md URL and structured context — one click, prompt included.

ChatGPT Claude Perplexity Gemini Grok

Narrative Entities

evalstats open-source Python package for calibrated small-sample inference

More from arXiv Computation and Language

View all →

Markdown (.md) · JSON-LD schema (.json) · Machine-readable for AI & GEO