diff --git a/experiments/convert_to_pdf.sh b/experiments/convert_to_pdf.sh
new file mode 100755
index 0000000..5c05bcc
--- /dev/null
+++ b/experiments/convert_to_pdf.sh
@@ -0,0 +1,15 @@
+#!/bin/bash
+
+# Define the input and output files
+input_file="parametric_expressivity.md"
+output_file="parametric_expressivity.pdf"
+
+# Check if the input file exists
+if [ -f "$input_file" ]; then
+    echo "Converting $input_file to PDF..."
+    pandoc "$input_file" -o "$output_file" --pdf-engine=pdflatex
+    echo "Conversion complete: $output_file created."
+else
+    echo "Error: $input_file not found."
+    exit 1
+fi
diff --git a/experiments/grid_xform_spec_10.py b/experiments/grid_xform_spec_10.py
new file mode 100644
index 0000000..de21a29
--- /dev/null
+++ b/experiments/grid_xform_spec_10.py
@@ -0,0 +1,23 @@
+# type: ignore
+
+"""
+Describe this transformation from grids to grids (ARC challenge).
+"""
+
+
+class GridTransformSpec:
+    colors = ["blue", "red", "green" "yellow"]
+
+    def Pre(self):
+        Objs = self.find_objects()
+        assert all(obj.getColor() == "grey" for obj in Objs1)
+        Objs1 = sorted(Objs, key=lambda obj: obj.height)
+        return Objs
+
+    def Post(self):
+        Objs = self.Pre()
+
+        def obj_func(obj, index):
+            color = colors[index]
+            return obj.copy().setColor(color)
+        return Grid.from_objects(Objs, obj_func=obj_func)
diff --git a/experiments/grid_xform_spec_5.py b/experiments/grid_xform_spec_5.py
new file mode 100644
index 0000000..6db2166
--- /dev/null
+++ b/experiments/grid_xform_spec_5.py
@@ -0,0 +1,39 @@
+# type: ignore
+
+"""
+Describe this transformation from grids to grids (ARC challenge).
+Notice that transformations are described as predicates `Pre --> Post` where `Pre` describes the grid before the transformation, and `Post` the grid after the transformation.
+To connect different predicates, `&&` is used for conjunction.
+For example ` Obj1.getColor() == Obj2.getColor && Obj1.getSize() = (3,3)`.
+
+Note that NestedGrid(origin, size) return a new nested grid, where there is a sub-grid of size `size` at each coordinate `(x,y)`. At coordinate `(0,0)`, the nested grid contains the grid of size `size` at position `origin` in the initial grid.
+All operations are immutable. In particular setXXX returns a copy of the receives with some property set as in the argument.
+"""
+
+
+class GridTransformSpec:
+    def Pre(self):
+        Objs = self.find_objects()
+        Obj = Objs.max()
+        Origin = Obj.getOrigin()
+        Size = Obj.getSize()
+        NestedGrid = self.grid_to_subgrids(subgrid_size=Size, origin=Origin)
+        return [NestedGrid, Obj]
+
+    def Post():
+        NG0, Obj0 = self.Pre()
+
+        def clamp(p):
+            return [max(min(p[0], 1), -1), max(min(p[1], 1), -1)]
+
+        def is_cardinal_or_diagonal(p):
+            return p[0] == 0 or p[1] == 0 or abs(p[0]) == abs(p[1])
+
+        def new_color(p):
+            Obj = NG0[clamp(p)]
+            return Obj.getColor() if is_cardinal_or_diagonal(p): else 0
+        NG1 = create_grid_of_objects(subgrid_size=Size,
+                                     subgrid_func=lambda p: Obj.copy().setColor(get_color(p)))
+        Grid1 = NG1.flatten()
+        Grid1 = Grid1.shift(Origin)
+        return Grid1
diff --git a/experiments/parametric_expressivity.md b/experiments/parametric_expressivity.md
new file mode 100644
index 0000000..b89a619
--- /dev/null
+++ b/experiments/parametric_expressivity.md
@@ -0,0 +1,78 @@
+
+
+### Introduction
+
+In the ARC challenge DSL, transformations (xforms) map one grid to another. To formalize these transformations, we use **shape types** to describe the sets of grids that a transformation can operate on. This allows us to distinguish between non-parametric and parametric transformations, depending on whether the transformations are specific to particular grid types or generalized using type variables.
+
+### Shape Types
+
+A **shape type** $S$ denotes a set of grids. This can capture structure such as dimensions, content patterns, or other properties. For example:
+
+- $S$ could represent all grids of a certain size, like all $3 \times 3$ grids.
+- $S$ could also represent grids with a specific pattern, such as grids where all cells along the diagonal are the same.
+
+Shape types include singleton types, denoting specific grids, and type variables `X` of kind `Grid` (for now).
+
+### Transformations
+
+A **transformation** $\text{xf}$ is a total function $Grid \rightarrow Grid$.
+
+### Specs
+A **spec** $[S,T]$ denotes the set of transformations $\text{xf}$ such that $\forall G \in Grid. G \in S \Rightarrow \text{xf}(G) \in T$
+
+In addition, $[S1,T1] \wedge [S2,T2] \wedge [S3,T3]$ is also a spec, denoting the intersection.
+
+### Matching
+A **spec** $[S,T]$ matches an example $(I,O)$ if
+
+1. $I \in S$
+2. $\forall \text{xf}. \text{xf} \in [S,T] \Rightarrow \text{xf}(I) = O$
+
+Intuitively, matching means that the spec admits the input $I$, and exactly specifies the behaviour on it.
+
+### Non-Parametric Matching
+
+Given examples $I_i \rightarrow O_i$ for $i = 1..3$, the following spec matches all of them:
+
+$$[I1,O1] \wedge [I2,O2] \wedge [I3,O3]$$
+
+where we write `I` as the singleton shape type denoting exactly `I`.
+
+Here's an example
+
+```python
+def flip_2x2(grid):
+    # Assumes grid is 2x2
+    return [[grid[0][1], grid[0][0]],
+            [grid[1][1], grid[1][0]]]
+```
+
+This transformation, `flip_2x2`, is specifically designed to work only for $2 \times 2$ grids. It does not generalize to other grid sizes.
+
+
+### Parametric Spec
+
+Consider this example
+
+```python
+def flip_grid(grid):
+    # Flip any grid horizontally
+    return [row[::-1] for row in grid]
+```
+
+This transformation, `flip_grid`, works for grids of any size by flipping each row horizontally.
+
+In addition to the non-parametric spec above, this transformation satisfies spec $[X, Flip(X)]$ where:
+
+- $X$ represents any grid.
+- $\text{Flip}(X)$ denotes the grid $X$ flipped horizontally.
+
+### Genericity of Specs
+
+A spec $\text{spec1}$ is more specific than $\text{spec2}$ if it is a strict subset.
+In the examples above, the non-parametric spec is more specific than the parametric one.
+
+### Genericity of Transformation
+
+A transformation $\text{xf1}$ is more generic than $\text{xf2}$ if it satisfies more generic specs.
+In the examples above, `flip_grid` is more generic than `flip_2x2`.
diff --git a/experiments/parametric_expressivity.pdf b/experiments/parametric_expressivity.pdf
new file mode 100644
index 0000000..72af4f1
Binary files /dev/null and b/experiments/parametric_expressivity.pdf differ