TimeGPT: A Scientific Management System for Time, Energy, and Tasks
TimeGPT is a scientifically sound and comprehensive management system for time, energy, and tasks, designed to help everyone better achieve their goals. The entire system is goal-oriented, decomposing objectives into smaller projects and executable to-do items. It provides a complete UCEVI scoring system for tasks, aimed at maximizing the realization of high-value goals, and ultimately intelligently schedules tasks based on task types and energy zones. Time, energy, and tasks eventually achieve a perfect positive cycle to facilitate better goal attainment.
Task Management
UCEVI Scoring System
UCEVI is an acronym for Urgent, Cost, Effort, Value, and Impact. This system is developed based on the Eisenhower Matrix, commonly known as the Urgent-Important Matrix. Simply put, the Eisenhower Matrix divides tasks into four quadrants based on their urgency and importance to determine priority. However, in my practice, I discovered two problems with this evaluation standard:
There is no differentiation in priority for tasks within the same quadrant.
Urgency and Importance focus on the starting and ending nodes of a task but overlook the process.
Addressing these two issues, I expanded the Eisenhower Matrix. First, I required the final result to be a scoring system, ensuring that priority differences still exist even for tasks within the same quadrant.
Secondly, I added consideration for the task process, specifically the resources required. I added a new dimension: Cost, representing the time expenditure of each task.
| Start | Progress | Result |
|---|---|---|
| Urgent | Cost | Importance |
However, this may still present issues. The current three dimensions only focus on the task itself, without involving the upper-level projects or the overall grand goals. In other words, if a task itself has a high cost and high importance, but the goal behind it is not significant, its final priority might end up higher than a task with a high cost but which belongs to a vital goal. Therefore, to incorporate consideration of the upper hierarchy, I added two more dimensions: Effort and Impact. Effort refers to the exertion required for the entire goal behind the task. Meanwhile, the previous "Importance" is decomposed into the task's own Value and the Impact brought by the task's underlying goal.
Ultimately, five different dimensions are obtained:
| Start | Progress | Result |
|---|---|---|
| Urgent | Cost | Value |
| Effort | Impact |
The beginning of a task depends on its degree of urgency; the process depends on the resources required to be spent; and the conclusion depends on the value brought after completion. Below are the specific definitions of these five dimensions:
Urgent: The degree of urgency of the task.
Cost: The resources (time) spent on the task itself.
Effort: The effort required to complete the Goal to which the task belongs.
Value: The value the task itself can bring.
Impact: The influence of the goal behind the task.
Since this is a scoring system, we need to set specific values for each dimension. Simply grading based on subjective psychological evaluation (1-10) is clearly not an objective scoring system. Therefore, for these five dimensions, we need to find corresponding actual values for conversion—values we need to actually record. There are different records for different levels.
| Goal | Project | Todo | |
|---|---|---|---|
| Start Date | Start Date | Start Date | Start Date |
| End Date | Deadline | Deadline | Deadline |
| Time Spent | Time Spent | Time Spent | Time Spent |
| Annual Income Increase | Increase in Annual Income | Parent Goal | Parent Project |
| One-time Income | One-time Income | ||
| Probability of Success | Success Probability | ||
| Income Conversion Ratio | Income Conversion Ratio |
Urgent: Days from the current date to the deadline.
Cost: The time expenditure of the task itself.
Effort: Time spent on the goal the task belongs to + Time spent on the project the task belongs to.
Value: The time spent on the task as a proportion of the total time spent on the goal, multiplied by Impact.
Impact: (Expected income if the goal succeeds (Income $\times$ Probability)) - (Opportunity cost if the goal fails (Probability of failure $\times$ Hourly wage)) + Mental pleasure brought by the goal.
At this point, we have obtained relatively objective data for UCEVI. Of course, there are still coefficients within this that have not been explained, which we will address later. With the UCEVI data, we face a new problem: the influence of the dimensions within UCEVI varies for every individual. Some might feel they have plenty of time and only care about the final value; such a person would believe that the influence of Cost and Effort should be lower, and vice versa. Assigning a coefficient to each variable purely by oneself is not an objective or rational approach. To solve this, we need to introduce the Analytic Hierarchy Process.
The Analytic Hierarchy Process (AHP) is a decision-making analysis method. It is primarily used for the analysis and decision-making of complex problems, especially those difficult to quantify entirely. For example, if I want to buy a computer, I might consider three aspects: CPU, GPU, and Motherboard. However, due to a limited budget, I cannot buy the best of everything, so I must make trade-offs. I perform pairwise comparisons of these three variables. For instance, if I value the GPU more than the CPU, I assign the GPU a 2, while the CPU vs. GPU comparison can only get a 1/2. Ultimately, I obtain the following table:
| CPU | GPU | Motherboard | |
|---|---|---|---|
| CPU | 1 | 1/2 | 1 |
| GPU | 2 | 1 | 3 |
| Motherboard | 1 | 1/3 | 1 |
Afterward, I calculate the percentage for each column and then sum the percentages to obtain the weight for each variable.
| CPU | GPU | Motherboard | Weight | |
|---|---|---|---|---|
| CPU | 0.25 | 0.27 | 0.2 | 0.72 |
| GPU | 0.5 | 0.55 | 0.6 | 1.65 |
| Motherboard | 0.25 | 0.18 | 0.2 | 0.63 |
This weight is still derived from our subjectivity (as it indeed needs to be), but it is more objective than purely grading variables based on intuition. After calculation, we can obtain our own UCEVI weights. Multiplying the weights by the variables gives us the final score (this also involves many operations like normalization and inverse value conversion).
How to Define Value
Value differs for different goals. Some goals directly bring money, while others bring a sense of achievement or pleasant time. We first discuss goals that directly yield money, as these are usually more intuitive and are the goals most of us pursue. For monetary income, there are two types: one-time income and an increase in annual income. To compare one-time income versus an increase in annual income, the common practice is to use the Net Present Value (NPV) method, usually using the bank interest rate as the discount rate.
However, if we only make comparisons this way, there are still issues. Under this algorithm, extreme cases like buying a lottery ticket might yield extremely high income compared to everything else. Therefore, when calculating the true value of a goal, what needs to be calculated is the Expected Utility. Expected utility, simply put, is the sum of various possible values brought by the final goal multiplied by the probability of those possibilities occurring.
If a simple increase in annual income is used as the impact of an event, the problem is that buying a lottery ticket would result in a very high increase in annual income without considering the low probability. Thus, the final decision should be the Result (Outcome) of an Action taken under a certain State of the world. Each outcome has its own Utility function, and a certain Probability exists for reaching that outcome. The final result is as follows: the expected utility of a certain action is the sum of the results of that action in different states of the world multiplied by their probabilities.
$$E[u(a, s)] = \sum_{s \in S} P(s) u(a, s)$$
The above method might be difficult to understand. Let’s use a simple example. The decision of whether or not to carry an umbrella is related to the state of the world—i.e., whether it rains or not. Different actions corresponding to different world states lead to different outcomes.
| Raining | Not Raining | |
|---|---|---|
| Carry Umbrella | Comfortable | Physically comfortable but carrying extra weight |
| No Umbrella | Body wet | Comfortable |
Different outcomes have different utility evaluations for everyone. Based on the probability of it raining or not raining, the expected utility of an action is obtained. Therefore, in extreme cases, one of my goals might be "finding a job that earns 10 million a year," which would give its corresponding projects and tasks very high priority. But since the possibility of achieving this is very low, we need to consider the probability of success in the current state of the world before deciding.
The results obtained through this method are much more reasonable than just using the increase in annual income. Some may ask how to accurately determine the probability of each event. It is impossible to predict the probability of future events with complete accuracy, but these probabilities can change as you progress. Whether using Bayesian theory to update probabilities with prior probability and new evidence, or Jeffreys' update after obtaining uncertain evidence, both can effectively help us move further in the right direction. These probabilities are related to many factors, including your internal beliefs and the state of the world; different people might give completely opposite probabilities for the same event. However, essentially, it is about working toward the direction of your internal beliefs.
How to Compare Mental Gains and Monetary Income
When setting goals, if the system only considers monetary income, goals with no temporary monetary income would never be considered. Some goals are just for mental pleasure, such as developing hobbies, fitness, etc. Therefore, we need to consider both mental gains and monetary income. The method used in TimeGPT is to introduce a Lottery Mechanism. Suppose at a certain time you face a choice: you can choose to go to work or stay home and play games. Usually, when people cannot decide, they flip a coin—heads for work, tails for gaming. A lottery is a similar concept, except you can set the probabilities yourself.
For example, I want to compare working versus playing volleyball. I create a lottery where there is a 1/10 chance of working and a 9/10 chance of playing volleyball, making me feel that regardless of the outcome, I am satisfied. In my mind, I want to play volleyball more than work. Thus, the equivalent value $V$ is equal to a constant $k$ divided by the probability $p$.
$$V_{eq} = \frac{k}{p}$$
Assuming that when the probabilities are each 1/2, we consider work and mental pleasure to be equal. Thus, we get $k = 0.5W$ (where $W$ is the hourly wage).
$$V_{eq} = \frac{0.5W}{p}$$
Looking at the model curve, this model basically meets our needs; only in extreme cases is there a significant surge, while at other times it stays within a normal range around $W$.
Therefore, for goals without direct income, you can use the time required to reach the goal multiplied by the equivalent hourly income calculated by the above formula. This is equivalent to saying that you obtained these hours of pleasure, which is the same as earning that much money.
Recommendation System
By now, we have a corresponding score for every task, but this still doesn’t satisfy our recommendation system. If we simply sort by score, several problems may arise:
A task might not have reached its start date yet, but it gets a high score due to high final value.
Similar tasks might get the same score because they belong to the same goal or result.
It cannot recommend specifically based on the daily situation.
To address these, we need to add extra constraints, but first, it is most important to classify tasks. Through long-term practice, I have divided tasks into three types: Tasks, Reminders, and Events. Their required methods are completely different:
Task: Tasks without a specific time, such as completing research.
Event: Events with a clear time point, such as seeing a doctor or holding a meeting.
Reminder: Small tasks that need to be completed quickly, such as paying rent.
| Start Time | Time Spent | Belongs to Goal |
|---|---|---|
| Task | Not fixed | Not fixed |
| Event | Fixed | Fixed |
| Reminder | Not fixed | Fixed and short |
Among these, Reminders do not need to be scored in our task system because they are usually very brief, small tasks. Therefore, the following discussion only involves the remaining two types. Tasks have no fixed start time and no fixed duration. Conversely, Events have very fixed start times and durations. Clarifying these task types is crucial; without classification, mixing them all together results in chaos.
After classifying tasks, we can provide reasonable recommendations for different types. For Tasks, since they have no fixed start time, they can be chosen almost anytime. For Events, if the day has not arrived, the event is meaningless because it requires a fixed time; it only happens when the time comes. For daily routines, we might have many simultaneously, but doing one a day is enough. Thus, the recommendation system needs to select if there are Events for the day, and then use the remaining time to choose a reasonable combination of daily routines and Tasks.
In plain terms, the recommendation system aims to arrange suitable tasks for each day within limited time. This sentence contains the system's constraints: limited time and suitable tasks. Limited time refers to the free time available for disposal that day. Suitable tasks involve obtaining necessary Event tasks first, then striving to use the remaining time for the highest-scoring combination of routines and Tasks. To implement this, we use Linear Programming. Linear programming is a mathematical method used to find the maximum or minimum value of a linear function under a series of linear inequality or equality constraints. It includes:
Variables: Elements you adjust in the objective function to reach a maximum or minimum. In business, these might be the quantities of different products produced.
Objective Function: The linear function you wish to maximize or minimize. For example, maximizing profit or minimizing cost.
Constraints: Restrictions in the form of linear inequalities or equalities that limit the feasible range of variables. For example, raw material or budget limits.
For our system, the variable is whether a task is executed. A task is either executed (1) or not (0).
The objective function is the sum of the scores of all executable tasks; we wish to maximize this sum.
$$\text{Maximize } Z = \sum_{i=1}^{n} Score_i \times x_i$$
Constraints are that the total time spent on executable tasks must be less than the active time of the day, Event tasks must be executed on the day, and only one of similar routine tasks is executed per day.
$$\sum_{i=1}^{n} Cost_i \times x_i \leq Time_{active}$$
With these settings, we can obtain the optimal task plan for the day.
Time Management
At the very beginning of the article, I mentioned that this is a system designed to help everyone. The common resource for everyone is Time. Time acts as a fair dimension to help measure all projects and reveals the "effort" one truly puts into a certain goal.
Most people have very low sensitivity to time. People usually evaluate a skill using time, saying "I've studied guitar for five years." But "five years" is a very vague expression, only indicating a time span. Whether one practiced once a week or every day during those five years makes a world of difference for the skill. Conversely, if one cannot calculate the actual time spent, one might be deceived by this "five years," feeling that they have studied for so long, so why is there no significant progress?
Take my own example. I started playing volleyball in November 2022. Before I had time awareness, I only knew I had played for a couple of years. But at the end of 2024, I reviewed the time I spent on volleyball over the past two years: it was only about 350 hours. According to the "10,000-Hour Rule" proposed by K. Anders Ericsson, reaching a world-class level requires 10,000 hours of investment. Comparing these examples, I initially thought I had invested a lot of time in volleyball over two years, but actually, 350 hours is a drop in the bucket compared to 10,000. Although my goal is not to reach a world-class level, from the 10,000-hour standard, this time is clearly insufficient. According to the "80/20 Rule," we might only need to invest 20% of the time to master 80% of basic skills. However, to refine the remaining 20% and approach world-class levels, one usually needs an extra 80% of the time. In other words, even without pursuing a world-class level, an average person needs to invest at least 2,000 hours to reach a high level of ability in a skill. So, at the end of 2024, I clearly realized my level was only 350 hours. To reach 2,000 hours as quickly as possible, I set a goal for 2025 to achieve 300 hours in a single year.
This is the significance of time management. If I didn't know the exact time I spent, I would only know I had played for two years without much progress, and in 2025, I might continue to invest only about 150 hours.
How to Record
Currently, there are many time-recording software options on the market (Timing, Toggl, Tyme), but ultimately none of them could be sustained. Market options generally fall into two types: automatic background recording on computers (Timing) or manually starting and ending a task (Toggl, Tyme). Both methods have issues with sustainability. Automatic recording cannot record time spent outside of computer use, and the recorded data is often too cluttered and overly realistic, leading to a loss of interest. Manual recording requires a start and an end action for every single task, leading to frequent forgetfulness. Ultimately, my choice is to use Interstitial Journaling. Interstitial journaling is very simple: just record a timestamp and a brief text. It not only helps record time usage but also helps organize thoughts when switching to the next task. After recording each day, one only needs to calculate the time difference between two timestamps and classify them.
For time classification, my current categories are: Core Work, Daily Life, Self-Improvement, Health, Relationships, and Rest. These six categories cover almost most life scenarios.
The most common problem in classification is: if a time period belongs to two different categories, how should it be distinguished? Through my practice, I have the following solutions:
Classify according to the primary purpose of the time period. For example, playing games with friends—should it count as gaming or maintaining relationships? That depends on the primary purpose of choosing that activity. If it's for relaxation, it's Rest; if it's for a gathering with friends, it's Relationships.
Record separately according to original time. For example, how to record watching TV while eating? Eating is a daily necessity; the normal eating time is recorded as such, and the portion exceeding that time is recorded under your extra purpose. For instance, if the whole process takes 2 hours, one hour is Daily Life for normal eating, and the remaining hour is Rest for watching TV.
Situations where simultaneous recording is possible. For example, how to record doing personal tasks while slacking off at work? The original work time already brings value, and the extra time used counts as its corresponding category. That is, we might have a situation where a day exceeds 24 hours.
Energy Management
Previous content only obtained the optimal daily task combination. The next goal is how to do the most suitable task at the most suitable time. Through a period of time recording, we can know which time periods of the day were spent on tasks we consider important work; these periods are the high-energy periods of each day. Dividing a day into small 5-minute segments results in 288 segments, which is a vector $V$ of length 288. For each day, we can obtain such a vector, where high-energy segments get a 1 and low-energy segments get a 0.
$$V = (v_1, v_2, \dots, v_{288}), v_i \in \{0, 1\}$$
Collecting the vectors for all days allows us to obtain an average energy vector $\bar{V}$.
$$\bar{V} = \frac{1}{n} \sum_{i=1}^{n} V_i$$
With this average energy vector, the system knows when the high-energy and low-energy periods are. After the recommendation system suggests reasonable tasks, they can be assigned to the highest energy periods.
$$\text{Maximize } \sum \text{Score}_i \times \text{Energy}(t_i)$$
The Trinity
So far, the management of time, tasks, and energy within this system has been fully explained; they exist as a trinity. After new tasks are generated, they require time to complete; time reflects the levels of energy, helping to better arrange tasks next time. This system hopes to progress in this virtuous cycle, completing tasks until goals are achieved.
FluxTime
The entire system described above looks very complex with many calculations required; it is almost impossible for an individual to implement it alone. To allow more people to use this system, I developed a software called FluxTime, allowing you to focus only on time recording and task creation. Other functions like task recommendation, energy curve calculation, and result feedback are handled by the software. Currently, FluxTime is available on the App Store and is completely free. Readers are welcome to download and test it.