Added link to better examples. Added more covers.

02_example.md

@@ -167,28 +167,14 @@ make it easier to identify and correct errors in the modeling process.
 
 ### More examples
 
-Here are some further examples, if you are not yet satisfied:
-
-- [N-queens](https://developers.google.com/optimization/cp/queens) (this one
-  also gives you a quick introduction to constraint programming, but it may be
-  misleading because CP-SAT is no classical
-  [FD(Finite Domain)-solver](http://www.gameaipro.com/GameAIPro2/GameAIPro2_Chapter26_Rolling_Your_Own_Finite-Domain_Constraint_Solver.pdf).
-  This example probably has been modified from the previous generation, which is
-  also explained at the end.)
-- [Employee Scheduling](https://developers.google.com/optimization/scheduling/employee_scheduling)
-- [Job Shop Problem](https://developers.google.com/optimization/scheduling/job_shop)
-- More examples can be found in
-  [the official repository](https://github.com/google/or-tools/tree/stable/ortools/sat/samples)
-  for multiple languages (yes, CP-SAT does support more than just Python). As
-  the Python-examples are named in
-  [snake-case](https://en.wikipedia.org/wiki/Snake_case), they are at the end of
-  the list.
-
-Ok. Now that you have seen a minimal model, let us look on what options we have
-to model a problem. Note that an experienced optimizer may be able to model most
-problems with just the elements shown above, but showing your intentions may
-help CP-SAT optimize your problem better. Contrary to Mixed Integer Programming,
-you also do not need to fine-tune any
+If you are not yet satisfied,
+[this folder contains many Jupyter Notebooks with examples from the developers](https://github.com/google/or-tools/tree/stable/examples/notebook/sat).
+
+Ok. Now that you have seen a minimal model, let us next look on what options we
+have to model a problem. Note that an experienced optimizer may be able to model
+most problems with just the elements shown above, but showing your intentions
+may help CP-SAT optimize your problem better. Contrary to Mixed Integer
+Programming, you also do not need to fine-tune any
 [Big-Ms](https://en.wikipedia.org/wiki/Big_M_method) (a reason to model
 higher-level constraints, such as conditional constraints that are only enforced
 if some variable is set to true, in MIPs yourself, because the computer is not

03_big_picture.md

@@ -4,6 +4,8 @@
 
 ## Alternatives: CP-SAT's Place in the World of Optimization
 
+![Cover Image World of Optimization](https://raw.githubusercontent.com/d-krupke/cpsat-primer/main/images/logo_8.webp)
+
 When you begin exploring optimization, you will encounter a plethora of tools,
 techniques, and communities. It can be overwhelming, as these groups, while
 sharing many similarities, also differ significantly. They might use the same

07_under_the_hood.md

@@ -4,6 +4,8 @@
 
 ## How does it work?
 
+![Cover Image How](https://raw.githubusercontent.com/d-krupke/cpsat-primer/main/images/logo_7.webp)
+
 CP-SAT is a versatile _portfolio_ solver, centered around a _Lazy Clause
 Generation (LCG)_ based Constraint Programming Solver, although it encompasses a
 broader spectrum of technologies.

08_benchmarking.md

@@ -385,6 +385,15 @@ provides the results for the `AddCircuit`-model.
 | lin318   |     318 |   92.43 |       41915 |     52025 |      19% |
 | pr439    |     439 |   94.22 |      105610 |    163452 |      35% |
 
+A last option is to split the y-axis into the part where the solving time is
+still within the time limit, and the part where it is not and the optimality gap
+becomes relevant. Such a plot has some benefits but can also be difficult to
+scale or aggregate.
+
+|                                                                                                                                                                 ![Split Plot](./images/split_plot.png)                                                                                                                                                                 |
+| :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
+| This plot splits the y-axis into a part where the instances can still be solved within the time limit, such that the time can be shown, and the part where the time limit is exceeded, such that the optimality gap can be shown. This example was generated by my student assistant Rouven Kniep, and he is working on creating an easy-to-use script for such plots. |
+
 This should highlight that often you need a combination of different benchmarks
 and plots to get a good understanding of the performance of your model.
 

README.md

@@ -374,28 +374,14 @@ make it easier to identify and correct errors in the modeling process.
 
 ### More examples
 
-Here are some further examples, if you are not yet satisfied:
-
-- [N-queens](https://developers.google.com/optimization/cp/queens) (this one
-  also gives you a quick introduction to constraint programming, but it may be
-  misleading because CP-SAT is no classical
-  [FD(Finite Domain)-solver](http://www.gameaipro.com/GameAIPro2/GameAIPro2_Chapter26_Rolling_Your_Own_Finite-Domain_Constraint_Solver.pdf).
-  This example probably has been modified from the previous generation, which is
-  also explained at the end.)
-- [Employee Scheduling](https://developers.google.com/optimization/scheduling/employee_scheduling)
-- [Job Shop Problem](https://developers.google.com/optimization/scheduling/job_shop)
-- More examples can be found in
-  [the official repository](https://github.com/google/or-tools/tree/stable/ortools/sat/samples)
-  for multiple languages (yes, CP-SAT does support more than just Python). As
-  the Python-examples are named in
-  [snake-case](https://en.wikipedia.org/wiki/Snake_case), they are at the end of
-  the list.
-
-Ok. Now that you have seen a minimal model, let us look on what options we have
-to model a problem. Note that an experienced optimizer may be able to model most
-problems with just the elements shown above, but showing your intentions may
-help CP-SAT optimize your problem better. Contrary to Mixed Integer Programming,
-you also do not need to fine-tune any
+If you are not yet satisfied,
+[this folder contains many Jupyter Notebooks with examples from the developers](https://github.com/google/or-tools/tree/stable/examples/notebook/sat).
+
+Ok. Now that you have seen a minimal model, let us next look on what options we
+have to model a problem. Note that an experienced optimizer may be able to model
+most problems with just the elements shown above, but showing your intentions
+may help CP-SAT optimize your problem better. Contrary to Mixed Integer
+Programming, you also do not need to fine-tune any
 [Big-Ms](https://en.wikipedia.org/wiki/Big_M_method) (a reason to model
 higher-level constraints, such as conditional constraints that are only enforced
 if some variable is set to true, in MIPs yourself, because the computer is not
@@ -412,6 +398,8 @@ very good at that).
 
 ## Alternatives: CP-SAT's Place in the World of Optimization
 
+![Cover Image World of Optimization](https://raw.githubusercontent.com/d-krupke/cpsat-primer/main/images/logo_8.webp)
+
 When you begin exploring optimization, you will encounter a plethora of tools,
 techniques, and communities. It can be overwhelming, as these groups, while
 sharing many similarities, also differ significantly. They might use the same
@@ -3275,6 +3263,8 @@ class KnapsackSolver:
 
 ## How does it work?
 
+![Cover Image How](https://raw.githubusercontent.com/d-krupke/cpsat-primer/main/images/logo_7.webp)
+
 CP-SAT is a versatile _portfolio_ solver, centered around a _Lazy Clause
 Generation (LCG)_ based Constraint Programming Solver, although it encompasses a
 broader spectrum of technologies.
@@ -3850,6 +3840,15 @@ provides the results for the `AddCircuit`-model.
 | lin318   |     318 |   92.43 |       41915 |     52025 |      19% |
 | pr439    |     439 |   94.22 |      105610 |    163452 |      35% |
 
+A last option is to split the y-axis into the part where the solving time is
+still within the time limit, and the part where it is not and the optimality gap
+becomes relevant. Such a plot has some benefits but can also be difficult to
+scale or aggregate.
+
+|                                                                                                                                                                 ![Split Plot](./images/split_plot.png)                                                                                                                                                                 |
+| :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
+| This plot splits the y-axis into a part where the instances can still be solved within the time limit, such that the time can be shown, and the part where the time limit is exceeded, such that the optimality gap can be shown. This example was generated by my student assistant Rouven Kniep, and he is working on creating an easy-to-use script for such plots. |
+
 This should highlight that often you need a combination of different benchmarks
 and plots to get a good understanding of the performance of your model.