diff --git a/README.md b/README.md
index 65aa9b8..607a7f5 100644
--- a/README.md
+++ b/README.md
@@ -2,8 +2,8 @@ SixtyPical
 ==========
 
 _Version 0.21_
-| _See also:_ [Bubble Escape 2K](https://github.com/catseye/Bubble-Escape#readme)
-∘ [SITU-SOL](https://github.com/catseye/SITU-SOL#readme)
+| _See also:_ [Bubble Escape 2K](https://codeberg.org/catseye/Bubble-Escape#bubble-escape)
+∘ [SITU-SOL](https://git.catseye.tc/SITU-SOL/)
 
 - - - -
 
@@ -138,7 +138,7 @@ Documentation
 *   [Output formats supported by `sixtypical`](doc/Output%20Formats.md)
 *   [TODO](TODO.md)
 
-[Future directions for SixtyPical]: https://gist.github.com/cpressey/f35e104b3e3cf555824aa2b4d15ea858
+[Future directions for SixtyPical]: doc/Future%20directions%20for%20SixtyPical.md
 [6502]: https://en.wikipedia.org/wiki/MOS_Technology_6502
 [MOS Technology 6502]: https://en.wikipedia.org/wiki/MOS_Technology_6502
 [effect system]: https://en.wikipedia.org/wiki/Effect_system
diff --git a/doc/Future directions for SixtyPical.md b/doc/Future directions for SixtyPical.md
new file mode 100644
index 0000000..4327bf6
--- /dev/null
+++ b/doc/Future directions for SixtyPical.md	
@@ -0,0 +1,60 @@
+Future directions for SixtyPical
+================================
+
+[SixtyPical](https://codeberg.org/catseye/SixtyPical) has reached a mature
+stage of development.  There are small features that could be added, but
+they are minor compared to the main features
+(abstract-interpretation/symbolic-execution/flow-typing-based
+static analysis and optimization of low-level 6502 programs).
+So the question arises -- what would be next for SixtyPical?
+
+It would be nice to port SixtyPical to work on a ISA other than the 6502 --
+the Z80, for example.  Or, a more practical choice would be the X86
+architecture.
+
+It would also be nice if it could somehow be made to work on regular
+assembly language programs.  It would still be acceptable if, in this
+case, it only worked on programs written in a particular style, to be
+compatible with SixtyPical (structured "for" loops, no arbitrary jumps,
+and so forth).
+
+It would also be nice to simply generalize the idea: a generic
+low-level language with a generic set of registers
+(global variables), the rules for which we can specify as part of
+the program.  The static analyzer then checks the program according to the rules.
+
+It would also be nice to design a more formal theory behind all
+this.  While SixtyPical works well for what it does, much of it was
+"unit-tested into existence" and the theories behind these parts are not made explicit.
+
+In fact, I think we can combine all these ideas.  (And, if these are
+really all nice to have in a next version of SixtyPical -- we should.)
+
+We can do this by splitting it up into a few different phases:
+
+The first phase is a "frontend" that takes a 6502 assembly language program in a common
+6502 assembly language format (perhaps with annotations, or perhaps accompanied by a
+configuration file) and translated it into a program description in the generic language,
+including the specification of the rules particular to the 6502.
+
+There is then a generic analyzer which checks the program in the
+generic language.
+
+If all the checks pass, then our guarantees have been met and the
+original assembly language program can simply be assembled, using any
+assembler that can understand its format.  (This approach is similar
+to using a model checker.)
+
+Unlike SixtyPical currently,
+which is an optimizing translator, in this method there would be no
+optimizations applied.  But I think this is an acceptable trade-off.
+It is especially acceptable if the assembly language input can be hand-optimized
+and still be checked by the analyzer, but this would be trickier to accomplish.
+
+Then, to support other architectures, one could define a similar "frontend"
+which works on Z80 assembly code, or X86 assembly, or what have you.
+
+In practice, it would probably be easiest to start with a
+"frontend" which converts 6502 assembly to the existing SixtyPical
+language.  Then design and implement the generic language.
+Then re-target the frontend to the generic language.