FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteEvidence version preferred96 / 162 · 2cf381c55ec9e312

Human Stress Assessment: A Comprehensive Review of Methods Using Wearable Sensors and Non-wearable Techniques

Psychology and cognitive neuroscience · 2202.03033v2

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

Human Stress Assessment: A Comprehensive Review of Methods Using Wearable Sensors and Non-wearable Techniques

6 Future Directions

…the lab environment or real-life scenario, we don’t have the ground truth and the majority of the studies in the literature have to rely on subjective scoring which can vary from person to person. It has been observed that physiological data from two participants may have the same pattern but one of the participant’s labeled as ”stressed” whereas the other is labeled as ”non-stressed” in subjective labeling ( Liapis et al., 2015 ) . This type of labeling ambiguity degrades the performance of the system and thus needs to be rectified by developing a more robust mechanism for labeling the data.

…all the devices whether they are EEG headsets, skin conductance, and heart rate measurement modules or smartwatches all have limited battery life and can last approximately around 4-5 hours. But to record the activities of a user for a complete day, the power consumption of the devices needs to be kept at a minimum level so that the battery can work for more time. Existing literature has not examined this power consumption factor and needs to be addressed to be able to manufacture efficient and long-lasting data acquisition devices for monitoring stress, especially for real-life environments.

↓ 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

Battery life remains a critical constraint for continuous stress monitoring in real-life settings, as the wearable devices used for physiological data acquisition typically operate for only a few hours on a single charge. Addressing power consumption is therefore essential for enabling all-day monitoring outside controlled laboratory conditions.

Passages supplied to the Evidence version

Optimised Convolutional Neural Networks for Heart Rate Estimation and Human Activity Recognition in Wrist Worn Sensing Applications ↗

The battery life of smartwatches and fitness trackers vary greatly depending on the features and functionality available on-board the wearable. The Apple Watch Series 5, which is more of a lifestyle and fitness tracker, can run for a period of up to 18 hours whereas the Fitbit Charge 3 fitness track…

Read full passage excerpt

The battery life of smartwatches and fitness trackers vary greatly depending on the features and functionality available on-board the wearable. The Apple Watch Series 5, which is more of a lifestyle and fitness tracker, can run for a period of up to 18 hours whereas the Fitbit Charge 3 fitness tracker can go for up to 7 days on a single charge. Continuous activity and heart rate monitoring speed up the depletion of the battery of most wearables. Gathering and processing of simultaneous sensor data can further increase the power consumption of the devices.

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

Privacy and ethical considerations also require attention before wearable stress monitoring can be deployed at scale in real-life settings. Continuous collection of physiological signals such as EEG, skin conductance, and heart rate raises concerns about data ownership, consent, and the potential misuse of sensitive health information, and these concerns are compounded when data are streamed or stored on cloud platforms.

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 = No Evidence; Output B = Evidence.

Output B is grounded in the draft and source. The draft explicitly discusses limited battery life (4-5 hours), the need for all-day recording, and that existing literature has not examined power consumption. The source supports that continuous activity and heart rate monitoring speeds up battery depletion and simultaneous sensor data increases power consumption. Output B directly continues this thread with no unsupported claims. Output A introduces an entirely new topic—privacy and ethical considerations about data ownership, consent, cloud platforms—that has zero support in the draft or supplied source. This is a material factual claim about concerns, risks, and deployment barriers that are not mentioned anywhere in the provided materials. The draft discusses labeling ambiguity and power consumption; the source discusses battery life and power consumption. Privacy/ethical issues are completely absent, making Output A an unsupported claim that changes the manuscript's scope and introduces unsourced material.

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.