A DEV Community tutorial published September 16, 2026 builds a complete personal-data prototype. It pulls sleep and readiness data from the Oura Cloud API, reshapes it with Polars, fits a scikit-learn Random Forest regressor, and displays a predicted “Cognitive Load Score” in Grafana. The plumbing is worth copying. The idea that the output detects developer fatigue is not supported by anything the article reports: no participants, no dataset, no trained-model results, no accuracy figure, and no check that its outcome label measures anything real.
The sections below take the pipeline one stage at a time, then cover the choices that decide whether the output means anything, the access and consent requirements, and how to evaluate a model like this without flattering it.
As an Amazon Associate I earn from qualifying purchases.
What the pipeline does, stage by stage
The architecture has five parts. Each is simple to build, and each can quietly distort the result.
| Stage | Tool in the tutorial | Role in the prototype | Where it goes wrong |
|---|---|---|---|
| Data access | Oura Cloud API (API V2) | Supplies daily summary and heart-rate records after OAuth2 authorization | Missing account or membership, outdated API version, or broader scopes than the model needs |
| Transformation | Polars | Aligns records by date, computes sleep-stage proportions and a rolling readiness average | Records filed under the wrong day; rolling windows that reach into nights that had not yet ended |
| Model | scikit-learn RandomForestRegressor | Fits a regressor from features to a label; the code uses a train/test split and a score call | A shuffled split on time-ordered data can overstate performance |
| Label | Self-labels or work signals such as GitHub pull-request velocity | Defines what the model is trying to predict | The label may not measure fatigue at all |
| Display | Grafana | Plots the predicted “Cognitive Load Score” over time | A smooth line looks authoritative even when the model is weak |
Which Oura signals are candidate inputs
The tutorial proposes two kinds of feature. Neither is a direct measurement of cognitive load, and the article does not present them as one. Treat them as candidates that need to earn their place.
#1 Best Overall
- THE WORLD’S SMALLEST SMART RING - 40% smaller than the previous generation, Oura Ring 5 is thinner, lighter and smaller than ever. The updated all-titanium design is more scratch-resistant and comfortable than ever
- SIZING IS ESSENTIAL - Oura Ring 5 uses unique sizing different from standard jewelry rings and other Oura Ring generations; use the Oura Ring 5 Sizing Kit to find your perfect fit before purchasing
- EVERYDAY HEALTH SENSING - Oura tracks over 50 health metrics, including sleep, activity, stress, heart health, metabolic health and women’s health metrics. Oura seamlessly integrates with over 40 apps including Natural Cycles, Strava, & more
- OURA MEMBERSHIP - First month of membership is included with purchase, for new members only. Subscription is 5.99/mo afterwards. Membership is tied to your account via the Oura App, not your physical ring
- REBUILT FOR GREATER ACCURACY - Oura Ring 5 has been re-engineered and reconfigured, providing more precise personal health data. Larger, smarter sensors bring research-grade accuracy—all in Oura’s smallest form factor yet
Sleep-stage proportions
These are the shares of a night spent in each sleep stage, expressed as proportions of total sleep. The article poses the question “Is it lack of REM sleep or high resting heart rate?” That is the author’s framing for the experiment, not evidence that a REM deficit or a high resting heart rate causes developer fatigue. A model that finds a correlation with a chosen label has found a pattern in one dataset. It has not found a cause.
Rolling readiness average
Oura’s readiness score is a summary that combines several signals, so it is already one step removed from any single physiological measure. Averaging it over a window smooths day-to-day noise. The article does not state the window length, and a longer window leans more heavily on earlier days. Choose the window before you look at results, and write it down with the model.
Heart-rate signals
Oura’s documentation describes heart-rate data as a separate OAuth2 scope, so requesting it is a deliberate choice. Resting heart rate is half of the article’s question, but the article reports no relationship between it and any outcome. Add it only if you can state in advance what it should show and why.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Rank #2
- THE WORLD’S SMALLEST SMART RING - 40% smaller than the previous generation, Oura Ring 5 is thinner, lighter and smaller than ever. The updated all-titanium design is more scratch-resistant and comfortable than ever
- SIZING IS ESSENTIAL - Oura Ring 5 uses unique sizing different from standard jewelry rings and other Oura Ring generations; use the Oura Ring 5 Sizing Kit to find your perfect fit before purchasing
- EVERYDAY HEALTH SENSING - Oura tracks over 50 health metrics, including sleep, activity, stress, heart health, metabolic health and women’s health metrics. Oura seamlessly integrates with over 40 apps including Natural Cycles, Strava, & more
- OURA MEMBERSHIP - First month of membership is included with purchase, for new members only. Subscription is 5.99/mo afterwards. Membership is tied to your account via the Oura App, not your physical ring
- REBUILT FOR GREATER ACCURACY - Oura Ring 5 has been re-engineered and reconfigured, providing more precise personal health data. Larger, smarter sensors bring research-grade accuracy—all in Oura’s smallest form factor yet
The outcome label decides what the model learns
The article names two possible targets: a “Cognitive Load Score” that the model predicts, and a “Productivity Score” suggested as a label built from self-labels or work signals. A regressor learns whatever the label rewards. The label is therefore where a fatigue project succeeds or fails, and it deserves more scrutiny than the model.
Self-labels
A daily rating of energy or focus is the most direct proxy for how a person feels. Its weaknesses are rating drift, day-to-day mood, and the fact that people often rate their workload as well as their state. Use a fixed scale, ask the same question every day, and write a short definition of what each value means before collecting anything.
Pull-request velocity
PR counts and merge rates are easy to pull from GitHub, but they describe output, not fatigue or code quality. Task size, review delays, team practices, work hours and project context all move them. A developer working through a hard problem may open fewer, larger PRs, and a rested developer may split the same work into many small ones. The score can change for reasons that have nothing to do with how the person feels.
Rank #3
- THE WORLD’S SMALLEST SMART RING - 40% smaller than the previous generation, Oura Ring 5 is thinner, lighter and smaller than ever. The updated all-titanium design is more scratch-resistant and comfortable than ever
- SIZING IS ESSENTIAL - Oura Ring 5 uses unique sizing different from standard jewelry rings and other Oura Ring generations; use the Oura Ring 5 Sizing Kit to find your perfect fit before purchasing
- EVERYDAY HEALTH SENSING - Oura tracks over 50 health metrics, including sleep, activity, stress, heart health, metabolic health and women’s health metrics. Oura seamlessly integrates with over 40 apps including Natural Cycles, Strava, & more
- OURA MEMBERSHIP - First month of membership is included with purchase, for new members only. Subscription is 5.99/mo afterwards. Membership is tied to your account via the Oura App, not your physical ring
- REBUILT FOR GREATER ACCURACY - Oura Ring 5 has been re-engineered and reconfigured, providing more precise personal health data. Larger, smarter sensors bring research-grade accuracy—all in Oura’s smallest form factor yet
| Criterion | Self-rated energy or focus | PR velocity or merge counts |
|---|---|---|
| Construct validity (does it measure what its name claims?) | Measures self-perceived state, not cognitive load | Measures merged output, not fatigue or code quality |
| Repeatability | Depends on the rating scale and the person’s day | Easy to repeat from the platform, but shaped by team conventions |
| Work-context confounding | High: workload shapes ratings | High: task size, review delays and project context |
| Time granularity | Can be recorded daily, matching nightly sleep data | Must be cut into intervals that line up with each night’s sleep |
Align dates before you model anything
Date alignment is a quiet source of false signal. Oura records each night against a time, while work activity happens in local-time intervals, and the two can be mismatched by up to a full day depending on how each record is filed.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Take a night that ends at 06:30 on Tuesday and a pull request merged at 23:00 on Monday. The merge happened before that sleep, so the sleep cannot have influenced the merge. A calendar-day join can still pair them, depending on which date the sleep record is filed under. The model then learns a relationship that cannot be used for prediction. Settle these rules before writing the join:
- Assign each sleep record to the work interval that follows it, using one fixed time zone.
- Build rolling features only from nights that ended before the prediction time.
- Apply the same time zone to Oura timestamps and to work logs.
Evaluate the model without flattering it
The article’s code includes a train/test split and a score call. The split proportion and the model’s hyperparameters are code settings, not results. The article reports no model score, no error figure and no validation of its label, so nothing in it shows that the model beats a simple guess.
Rank #4
- SIZE BEFORE YOU BUY - It’s important to size yourself using our Oura Ring 5 Sizing Kit before purchasing an Oura Ring 5. Oura Ring 5 sizes are different from standard ring sizes and Oura Ring 4 sizes
- AVAILABLE SIZES - Oura Ring 5 is available in sizes 6-13. There are no half sizes
- CHOOSE A FINGER - Oura Ring 5 works on all fingers and can be worn on either hand. For optimal accuracy, we recommend wearing your Oura Ring 5 on your index, middle, or ring finger. Size the exact finger you plan to wear your Oura Ring on
- CHOOSE A FINGER - Oura Ring 5 works on all fingers and can be worn on either hand. For optimal accuracy, we recommend wearing your Oura Ring 5 on your index, middle, or ring finger. Size the exact finger you plan to wear your Oura Ring on
- COMFORT AND ACCURACY - A properly fitting Oura Ring 5 will result in the highest data accuracy, comfort, and overall experience
The main hazard is the random split. Adjacent days of sleep and work are similar, so a shuffled split lets the model train on Thursday and test on Wednesday. The score then measures recognition of a neighbouring day, not prediction. Use a time-ordered split: train on earlier weeks and test on later ones. Then check the following:
- Leakage in features. Every rolling average must use only data that existed before the prediction time.
- Leakage in labels. The label must not be computed from the features, or from data that only appears after the prediction time.
- A naive baseline. Compare the model with predicting the average label, and with predicting yesterday’s label. A model that cannot beat yesterday’s label has learned little that is useful.
- Scope. One person’s data describes that person. Claims about developers in general require results across several people, tested separately for each.
Getting Oura data: access requirements
Personal data requires the account, application, membership and authorization steps below. The sample-data route in the article does not require them.
An Oura account and an API application
Oura’s support documentation says API use requires an Oura account and an API application. Set up both before writing the authorization code, because the OAuth2 flow depends on the application you register.
Best Value
- 【Size Before You Buy】Before purchasing the Milavan Smart Ring, we recommend referring to the size chart and carefully measuring your finger circumference. This helps ensure a comfortable fit and avoid unnecessary returns or exchanges
- 【Wellness Tracking】The slim ring features advanced 4.0 sensors that automatically track your heart rate every 30 minutes while worn. Daily data is available in the app for reference only and is not intended for medical use
- 【Intelligent Sleep Tracker】The fitness ring automatically tracks your sleep quality, including deep and light sleep, REM (eye movement), and awake periods. You can view all sleep data anytime in the app for daily reference
- 【8 Sports Modes Tracking】Supports 8 sports modes, including Running, Cycling, Jump Rope, Walking, Hiking, Yoga, Dancing, and Golf. View workout duration, steps, heart rate, speed, calories burned, distance, and other activity data in the app
- 【Stylish Fitness Companion】Featuring a slim 2mm profile and a specially polished inner surface, this smart ring is designed for comfortable all-day wear and most sports activities. View detailed workout data in the app after selecting a sport mode
An active membership for Gen3 rings
Oura’s current support guidance says Gen3 users without an active Oura Membership cannot access their data through the API. Check the support documentation for other ring generations rather than assuming the same rule applies or does not apply.
Use API V2
API V1 was removed on January 22, 2024, so any code that calls V1 endpoints will fail against current access. Build on V2. Oura’s support documentation also says a recent Oura app version may be needed for newer V2 data types. If a field is missing, update the app and check the data type before debugging your code.
OAuth2 and minimal scopes
Oura documents OAuth2 authentication with named scopes for categories such as daily summaries and heart-rate data, and recommends requesting only the scopes an integration needs. The article’s bearer-token snippet is a compact illustration. Pasting a token into a script can support a local experiment, but it is not a production authentication pattern. Follow Oura’s current OAuth2 setup instructions, keep tokens out of source code, and revoke and replace credentials if they are exposed.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesConsent, privacy and workplace use
Oura’s API agreement describes user data and user consent, and restricts certain uses or combinations of personal data made without consent. Joining health-related signals to work activity is exactly the combination that needs the most care. A build that takes this seriously meets the following requirements:
- Every person whose data is used gives informed permission for this specific purpose, and can withdraw it.
- Only the fields the model needs are collected. Raw sleep and heart-rate streams are not copied into work-tracking systems.
- Tokens and personal data are stored behind access controls, and only the people who need the scores can see them.
- The label source, such as Git activity, is collected under the same consent and for the same stated purpose.
A workplace deployment raises employment and privacy questions that depend on jurisdiction. Neither Oura’s agreement nor the article settles those obligations. Complete privacy and employment review before any colleague’s data enters the pipeline.
Choosing a starting point
There are two real ways into the tutorial: use an Oura Ring with your own authorized data, or start from sample data. They serve different goals.
Quick Recap
| Factor | Sample data | Your own Oura Ring data |
|---|---|---|
| Hardware cost | None. The article says a ring is not needed merely to explore the code | Requires an Oura Ring. The article gives no price, so check Oura’s current product listing |
| Representativeness | Not your physiology or work pattern. Useful for checking code paths | Reflects your own pattern, but is a sample of one person |
| API and account setup | Not described in the article | Oura account, API application, OAuth2 authorization, and an active membership if you use a Gen3 ring |
| Privacy exposure | Low, because no personal health record is involved | Personal health data on your machine and in any dashboard you build |
| Best use | Learning the pipeline and debugging transformations | Testing whether your own signals track a label you trust |
Choose based on what you are trying to learn:
- To understand the pipeline, start with sample data and stay there until the transformations and split behave as intended.
- To run a personal experiment, use your own data only after the label is defined, the date alignment is fixed, and the evaluation is time-ordered.
- To monitor other people, complete the consent and review steps above before collecting anything.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Recommended Free Tools
