Stub stuff out. Include markdown discussion

SOURCE_REGION_EXPERIMENTS.md

@@ -0,0 +1,16 @@
+# Source Region Parser Experiment Notes
+
+## Current Parser Experiment
+
+- The parser works directly with `CodePointIndex` and `CodePointSpan`; rich `Span` conversion is intentionally left to rendering or reporting code.
+- Structured errors are currently enough to identify the failing region and the syntactic fact that failed, without committing to message text.
+- Empty documents are valid because the document grammar is a sequence of expressions.
+
+## Source Region Observations
+
+- `SourceText.getSpan(CodePointSpan)` is the key boundary operation: parsers can stay low-level, while diagnostics can opt into line and column information later.
+- Zero-width EOF spans are important for parse errors such as an unclosed list. The existing `eofSpan` support fits this well.
+- Parser clients still need a few tiny local span helpers, such as current-code-point span and EOF cursor span.
+- `renderSpan` returning structured `LineView` data works well for UI rendering because the application can choose its own DOM and styling without reimplementing line slicing.
+
+## Potential Nice-To-Haves