FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteNo-Evidence version preferred68 / 162 · c449ec78bad0f12b

Profile Likelihoods in Cosmology: When, Why and How illustrated with $Λ$CDM, Massive Neutrinos and Dark Energy

Astrophysics · 2408.07700v2

FLOWING EVIDENCE BENCHMARK

How do we tell whether Evidence helps?

At the same writing position, with the same model and task, how does supplying retrieved paper passages change the first output? We compare matched versions and retain ties, unusable outputs and incomplete reviews.

SAME DRAFT · EVIDENCE ON OR OFF

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED INPUTS

SHARED BY BOTH

Manuscript context, model, task and prompt

A · With Evidence

Retrieved paper passages supplied

B · Without Evidence

No retrieved passages supplied

LLM

Same model and version

A → first output

B → first output

Blind judge agent

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Order check: X / Y → Y / X

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The judge sees the same draft and both first outputs without knowing which received Evidence.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are autocomplete positions stratified?

Before seeing generation outcomes, we check whether a retrieved passage contains a specific proposition that directly supports the next writing move. Those positions appear in the left opportunity group; the rest are ordinary positions on the right. We select a balanced sample from admitted papers in each field. The 50/50 split is experimental, not a measure of how often either type occurs in writing.

AUTOCOMPLETE · 110

Biology, statistics and astrophysics

55 positions on each side; statistics uses the ten-paper rerun.

AUTOCOMPLETE · 52

Psychology and climate science

26 positions on each side; psychology includes nine papers and climate science four.

How are the table percentages calculated?

Across five fields there are 81 source-opportunity positions. The original blind review preferred Evidence in 51; a task check moved one empty Evidence continuation to both unusable, leaving 50 in public counts. In the ordinary group, another pair of empty outputs moved from no winner to both unusable. Original verdicts remain visible on case pages.

50Evidence version preferred
÷
81All positions in this group
=
62%Evidence preference in this group

Source contribution is a separate review: 16 of 21 Evidence wins entered into source review directly used retrieved papers; another 29 wins await review.

These are development-stage model judgments pending independent human review. They are not formal Benchmark conclusions and do not, on their own, establish causality.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

Profile Likelihoods in Cosmology: When, Why and How illustrated with $Λ$CDM, Massive Neutrinos and Dark Energy

I Introduction

…ry can lead to an overestimation of the uncertainty. . Proximity to a boundary can also lead to a dependence on the choice of lower limit in the prior as well as to the shape of the prior, e.g. linear vs. logarithmic prior. A strong sensitivity of the parameter constraints on the prior was reported in the cosmology literature in different contexts, e.g. neutrino mass bounds [ 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 ] ; early dark energy [ 22 , 23 ] , full-shape galaxy clustering with the effective field theory of large scale structure [ 24 , 25 , 26 , 27 ] , and stage-IV forecasts [ 28 ] .

… experiments. and the likelihood is parametrization invariant, which makes it insensitive to the problems above. This lack of dependence on priors makes frequentist constraints especially useful in situations when the inferred parameter interval has a strong dependence on the prior, when prior volume effects are dominating the constraints, or when the cosmological limits are close to a physical boundary. In those cases, the frequentist intervals can provide important complementary information about the analysis, in particular when used along the Bayesian analysis to explore the effects above.

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionJudgment: No-Evidence version preferred

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Evidence

Same model and position · retrieved passages supplied · first output

Not preferred in this comparison

A frequentist profile-likelihood analysis of New Early Dark Energy, for instance, was motivated by the impact of prior volume effects on the Bayesian constraints around recombination.

Passages supplied to the Evidence version

Profiling Cold New Early Dark Energy ↗

Recent interest in New Early Dark Energy (NEDE), a cosmological model with a vacuum energy component decaying in a triggered phase transition around recombination, has been sparked by its impact on the Hubble tension. Previous constraints on the model parameters were derived in a Bayesian framework…

Read full passage excerpt

Recent interest in New Early Dark Energy (NEDE), a cosmological model with a vacuum energy component decaying in a triggered phase transition around recombination, has been sparked by its impact on the Hubble tension. Previous constraints on the model parameters were derived in a Bayesian framework with Markov-chain Monte Carlo (MCMC) methods. In this work, we instead perform a frequentist analysis using the profile likelihood in order to assess the impact of prior volume effects on the constraints.

No Evidence

Same model and position · no retrieved passages · first output

Preferred by blind review

Profile likelihoods offer a frequentist route to parameter inference that bypasses the prior-volume and boundary effects discussed above, and they have consequently been applied across a range of cosmological problems.

Judgment and task check

The blinded model review preferred the no-Evidence version. Its original rationale appears below.

Read the original model judgment

In this rationale, Output A = No Evidence; Output B = Evidence.

Output A provides a coherent, general continuation that bridges the draft's discussion of prior sensitivity to the next paragraph's discussion of frequentist methods. It accurately states that profile likelihoods bypass prior-volume and boundary effects (grounded in the draft's preceding sentences) and notes their application 'across a range of cosmological problems'—a mild, defensible generalization that matches the draft's own list of cosmological contexts. Output B, while factually grounded in the NEDE source, introduces a specific example (NEDE) that is jarringly narrow and somewhat repetitive: the next paragraph already begins with 'experiments. and the likelihood is parametrization invariant...' which speaks generally about frequentist methods, not about NEDE specifically. The NEDE source is about a specific model's frequentist analysis, but inserting it here would disrupt the flow from general prior problems → general frequentist solution → specific applications. Output A's broader framing better prepares for the subsequent paragraph's general claims about frequentist intervals. Both are usable and without unsupported claims, but A is more coherent as a transitional paragraph.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.