FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteEvidence version preferred45 / 162 · 69bb5ba7aebc8ab0

A review of machine learning applications in wildfire science and management

Climate and environmental science · 2003.00646v2

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

A review of machine learning applications in wildfire science and management

5 Discussion > 5.5 Opportunities

…mbine multiple trees in an optimized way, ensemble tree classifiers are less interpretable than single tree classifiers. On the other hand, BNs are one example of an ML technique where good explanations for results can be inferred due to their graphical representation; however, full Bayesian learning on large-scale data is very computationally expensive which may have limited early applications; however, as computational power has increased we have seen an increase in the popularity of BNs in wildfire science and management applications (e.g., Penman et al. [2015] , Papakosta et al. [2017] ).

…n). The ML community, however, recognizes the problem of interpretability and work is underway to develop methods that allow for greater interpretability of ML methods, including methods for DL (see for example, McGovern et al. [2019] ) or model-agnostic approaches [ Ribeiro et al., 2016 ] . Runge et al. [2019] further argue that casual inference methods should be used in conjunction with predictive models to improve our understanding of physical systems. Finally, it is worth noting that assessing variable importance (see Sec. 4.3.4 ) for a given model can play a role in model interpretation.

↓ 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: 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

Preferred by blind review

Model-agnostic variable importance measures offer one route to this end, as they can be applied to any trained supervised learning algorithm for which predictions on new data can be obtained, including ensembles that combine multiple base learners.

Passages supplied to the Evidence version

A Simple and Effective Model-Based Variable Importance Measure ↗

Secondly, our method is suitable for use with any trained supervised learning algorithm, provided predictions on new data can be obtained. For example, it is often beneficial (from an accuracy standpoint) to train and tune multiple state-of-the art predictive models (e.g., multiple RFs, GBMs, and de…

Read full passage excerpt

Secondly, our method is suitable for use with any trained supervised learning algorithm, provided predictions on new data can be obtained. For example, it is often beneficial (from an accuracy standpoint) to train and tune multiple state-of-the art predictive models (e.g., multiple RFs, GBMs, and deep learning NNs (DNNs)) and then combine them into an ensemble called a super learner through a process called model stacking . Even if the base learners can provide there own measures of variable importance, there is no logical way to combine them to form an overall score for the super learner. However, since new predictions can be obtained from the super learner, our proposed variable importance measure is still applicable (examples are given in Sections 5 – 6 ).

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

The product produced no insertable continuation text.

The model explicitly returned a no-completion marker.

Judgment and task check

The blinded model review preferred the Evidence version. Its original rationale appears below.

Read the original model judgment

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

Output A provides a coherent, grounded continuation that directly connects to the preceding discussion about model-agnostic approaches and variable importance. The content is supported by the supplied source, which explicitly states that the method is 'suitable for use with any trained supervised learning algorithm, provided predictions on new data can be obtained' and specifically mentions combining models into ensembles/super learners where 'our proposed variable importance measure is still applicable.' Output A accurately captures this without adding unsupported claims, maintaining the manuscript's focus on interpretability solutions. Output B is empty, which fails the task requirement of providing 'one coherent academic paragraph' to continue the manuscript. The blank output does not advance the discussion or connect the interpretability problem to the variable importance theme established in the draft.

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.