THREE STRUCTURAL ASSUMPTIONS TESTED THIS TURN -- ALL WRONG.  Recorded so the
next person does not repeat them.

1) "0x7F638C2 is the resource allocator"  -> WRONG.
   file 0x1238C2 is a printf-style logger: it loads format descriptors
   (MOVI 0x3fb1cea0 / 0x3fb1ce60 / 0x3fb1cf20 / 0x3fb1cf60), fetches args from
   A11's fields and calls 0x7FC22DE / 0x7FC1BD2 / 0x7FC2F04.  The real
   allocation is the NEXT call, 0x7EACA82 -> file 0x6CA82.

2) "0x7F174A4 is the XIE resource-table lookup" -> WRONG.
   file 0xD74A4 indexes a table at bit 0x21FFE0 by
     N = (id & 0xFFF00000) >> 12
   and bounds-checks N against the count at 0x225870, which reads 16.
   For our client ids (0x40C000, 0x40D000) N = 1024, i.e. far out of range,
   and the bucket slot (0x227FE0) reads 0.
   (id & 0xFFF00000)>>12 is the CORE-X client index, so this is the core-X
   resource lookup by XID, not anything to do with XIE resources.

3) "@0x1AD980 / @0x1AD990 / @0x1AD9A0 are the XIE resource list heads" -> WRONG.
   Reading 32 bits at those bit addresses yields 0x02000002, 0x04000200 and
   0x00000400 -- small 16-bit-field values, and following them as a pointer
   chain reads zeros.  They are not list heads; the arguments built around
   them at 0x679B0E / 0x67B28 are something else (they are read with `MOVE
   @addr,A11,0x0`, an 8-bit field, which is suspicious in itself).

WHY THIS MATTERS: the BadMatch(8) / BadAccess(10) / BadIDChoice(14) codes we
have been comparing all session were being attributed to structures I had
mis-identified.  The measurements themselves are sound (S19LOOK a8 = 0x08 /
0x0A / 0x0E at 0x6A7B4 is a direct observation), but every explanation built on
top of these three functions is now void.

CONCLUSION: further static inference about the XIE resource model is producing
wrong answers faster than it produces progress.  The remaining work needs the
protocol definition, not more guessing.
