The aim is simple: the same task earns the same credit no matter which computer runs it, and the credit reflects how much work the task actually took. Here is how we do that.

What we don't use
Many BOINC projects use BOINC's built-in credit system "CreditNew", which estimates the work from your computer's benchmark scores and how long the task ran. That can come out uneven, for example when a computer's benchmarks haven't been run or are out of date, and it can differ between CPUs, GPUs and operating systems. Kessler@home doesn't use it. Your computer's benchmarks, its speed and the time a task took play no part in your credit.

Counting the work
Each task is a simulation of the debris environment in low Earth orbit, and tasks differ a lot: some scenarios run quietly for two centuries, others spiral into a collision cascade within a few decades and end up tracking two million objects. So instead of guessing, the application counts the work it does while it runs, and reports those counts with its result. The counts cover the parts of the simulation that take the time:
- the atmospheric drag calculations that slowly bring objects down
- the Sun and Moon gravity calculations for objects in high orbits
- the satellites flying in the dense constellation layers
- every fragment made when objects collide or explode
- a fixed amount for loading and setting up each task

The simulation is exactly reproducible, so every computer that runs a task reports exactly the same counts for it, and the server checks them along with the rest of the result.

Turning the counts into credit
To learn what each kind of work costs, we timed a set of very different runs on one fixed reference computer (a laptop with an Intel Core i7-1165G7) and measured how much time each kind of work takes. With those costs, the counts predict how long a task takes on that computer to within about 4% on average. A task's credit is its predicted amount of work, scaled so that our standard reference task, the baseline scenario, is worth 100 credits. On the reference computer that comes to roughly 740 credits per hour for one CPU core.

We measure the costs again on the same reference computer for every new version of the application, so credit stays consistent from one version to the next.

What this means for you
- A task earns the same base credit on any computer: Windows, Linux or Mac, CPU or GPU.
- A faster computer or GPU earns more per hour only by finishing more tasks.
- Every task is run on two computers, and credit is granted when both results agree exactly. Until the second copy comes back, your task shows as pending.
- Both copies earn the same base credit. If the two disagree, a third computer runs the task, and only the copies that agree earn credit. Results with errors or wrong answers earn nothing.

The fast-return bonus
Returning tasks quickly helps the project move through the scenarios faster, so we reward it:
- returned within 24 hours of being sent to your computer: +50% credit
- returned within 48 hours: +25% credit
- returned later, but before the 5-day deadline: the base credit

The clock starts when the server sends the task to your computer and stops when your BOINC client reports the finished task. Your bonus depends only on your own copy: if the other computer is slow, you still keep your bonus. The bonus is only paid on results that validate.

To get the full bonus, keep your BOINC work buffer small, so tasks don't wait in your queue for hours before they start. BOINC usually reports finished tasks the next time it asks for new work; with a small buffer that happens often.