FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteNo-Evidence version preferred136 / 162 · 6724cf6bba101e4a

Generalizable machine learning for stress monitoring from wearable devices: A systematic literature review

Psychology and cognitive neuroscience · 2209.15137v3

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

Generalizable machine learning for stress monitoring from wearable devices: A systematic literature review

4 Discussion > 4.4 Lack of Generalization

…tatistical power given a two-tailed power test with a correlation coefficient of 0.5 [ 73 ] , and an assumed significance level of α \alpha =0.05 for each of the studies reviewed, based on the number of unique test subjects contained within each dataset when used for training and validation. Of these, datasets utilized by Greco et al. [ 32 ] and Ehrhart et al. [ 48 ] achieved at least 80% power by using non-public datasets while the public WESAD [ 45 , 27 , 28 , 30 , 6 , 14 , 47 ] and SWELL [ 49 , 14 ] datasets achieve 45% and 70% power respectively, based on number of test subjects included.

…ngs of previously observed subjects during training. This cannot ensure generalizability of the developed model to other subjects or datasets. Recently some studies are evaluating the potential of person-specific models and their promise in improving generic stress detection models [ 14 ] . Of the studies reviews in this paper, and scored using the IJMEDI checklist [ 2 ] , three studies [ 48 , 47 , 32 ] were found to likely achieve generalization (Table 8 ), based on model validation and the use of sufficiently large training datasets based on the number of individual study subjects included.

↓ 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

The reliance on small, non-public datasets limits the statistical power of many reviewed studies and constrains their ability to demonstrate generalization across subjects and datasets.

Passages supplied to the Evidence version

Automated Word Stress Detection in Russian ↗

In this study we address the problem of automated word stress detection in Russian using character level models and no part-speech-taggers. We use a simple bidirectional RNN with LSTM nodes and achieve the accuracy of 90% or higher. We experiment with two training datasets and show that using the da…

Read full passage excerpt

In this study we address the problem of automated word stress detection in Russian using character level models and no part-speech-taggers. We use a simple bidirectional RNN with LSTM nodes and achieve the accuracy of 90% or higher. We experiment with two training datasets and show that using the data from an annotated corpus is much more efficient than using a dictionary, since it allows us to take into account word frequencies and the morphological context of the word.

No Evidence

Same model and position · no retrieved passages · first output

Preferred by blind review

These three studies relied on non-public datasets with larger numbers of unique test subjects, whereas the public WESAD and SWELL datasets provided fewer test subjects and correspondingly lower statistical power. This pattern suggests that dataset size and subject diversity, rather than model architecture alone, may underlie the observed differences in generalization potential across the reviewed studies.

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 = Evidence; Output B = No Evidence.

Both outputs are coherent continuations grounded in the draft, but B is more useful and specific. Output A is a generic summary statement that largely restates points already made in the preceding paragraph (small datasets limit power and generalization). It adds little new information and reads as a redundant conclusion rather than advancing the narrative. Output B specifically ties the three generalization-achieving studies [48, 47, 32] to their non-public dataset status and larger subject counts, explicitly contrasting this with the public WESAD and SWELL datasets' lower power—a contrast implied but not fully drawn in the draft. B also introduces a substantive interpretive claim (dataset size and subject diversity 'rather than model architecture alone' may underlie generalization differences) that is a natural, source-grounded inference from the draft's discussion of power calculations and generalization assessment. B provides more analytical value with less editing needed to integrate into the manuscript's flow.

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.