Put the SPC on the 68000, and find P4 was blocked on a file nobody needed

ROADMAP P4, first half. Every byte the player has ever consumed was placed in
emulated RAM by a host: decode.lua preloaded a container, stream.lua answered a
mailbox at a modelled rate. src/player/scsi.i selects a SCSI target on a real
MB89352 and issues READ(10) itself -- 4,096 B from LBA 0 and 2,048 B from LBA
1000, both byte-for-byte against the host's copy of the same volume, with no
IOCS and no host in the transfer path. The non-zero LBA is the half that
matters: a driver that emits a malformed LBA field still passes block 0,
because zero is what a malformed field usually is.

P4 was recorded as blocked in this tree and was not. Session 21's handoff said
MAME's x68000 has no MB89352 path; -exp1 cz6bs1 instantiates one next to the
HD63450, and FINDINGS 32.4 had read that card's DMA glue in session 9. The
session-21 note is a regression in the record, not a discovery. What is
genuinely absent is the 8 KB scsiexrom.bin MAME requires to INSTANTIATE the
card and the player never executes -- driving the SPC registers directly has
been the plan since BENCHMARK item 4 in session 2 -- so scsi_run.sh supplies a
zero-filled placeholder on its own rompath, leaves the user's romset alone, and
lets MAME print WRONG CHECKSUMS as it should. B3 is untouched: it wants that
ROM's bytes disassembled and a blank one has none.

The register map is measured, not inferred, and it corrects MAME's own
documentation. The probe walks $EA0000..$EA003F one address at a time with a
bus-error handler that records the fault and steps the index, because a
sequential dump reports the first hole as the answer -- the earlier version
took a bus error at $EA0006 and knew nothing about the other 57. 60 of 64
answer; the two holes are exactly the TMOD and EXBF the MB89352 omits and the
MB87030 has. MAME leaves HOLES and does not shift the later indices down, which
its own device summary claims it does, and that is what keeps DREG at $EA0015.

The data register is DMA-only here and a PIO write vanishes. x68k_scsiext.cpp
glues $EA0015 and nothing else, and with exown() asserted and DRQ low the byte
is discarded: no error bit, no status change, no interrupt. Quieting all four
DMAC channels does not change it. Measured rather than reasoned about -- write
$5A, read back $00 with the FIFO still empty -- because ten command bytes
vanishing without trace looks exactly like a target refusing a command, which
is how it first presented. So every transfer runs the SPC in DMA mode and the
CPU moves the bytes through the DMAC's own door.

That costs the argument something, and it is easy to overclaim here: with exown
asserted at idle MAME cannot distinguish a CPU-driven byte at $EA0015 from a
DMAC-driven one. This shows the DATA PATH and cannot by itself show that the
HD63450 is driving it, which is precisely what ROADMAP calls P4's first job.
Whether a real CZ-6BS1 also refuses PIO there is not settled; it is a property
of MAME's model and it wants a board.

W did not move by one clock, and could not have. MAME's device models are
functional rather than transfer-timing accurate and 42.5 reads its DMAC
configured in wall-clock attotimes, so this is BENCHMARK Tier 1 -- does the
read path work -- and never Tier 2. W is still the largest open number here.

Five bugs, four of them silent, recorded in 57.5 because the pattern is the
finding: a chain of rol.l #8 that loaded a transfer counter of ZERO from a
count of 10; a byte handed to a FIFO mistaken for a byte on the bus; a fixed
phase sequence where the bus decides the order; the discarded PIO write; and an
initiator that must drop ACK and only then release the bus. The last appeared
only once there were TWO reads -- one passed byte-exact and every conclusion
from it was sound, and the second could not select. A player issues one command
per record, so that failure would have been universal in the ring and invisible
in a one-read demonstration.

No decoder code changed; decode.bin is still 1,296 B at the same MD5. check.sh
gains a SCSI stage that builds the volume out of the same stream_disk.bin the
ring rig reads, gates the register window at 60 of 64 and both reads
byte-exact, and skips when chdman is absent. ALL GREEN before and after.

Claude-Session: https://claude.ai/code/session_01194oWYW8DQXK1SZ2DnChW6

docs/ROADMAP.md

@@ -5,6 +5,7 @@ Amended end of session 21: P1 done, P2 half done (FINDINGS 53).
 Amended end of session 22: P3 done (FINDINGS 54).
 Amended end of session 23: P5 done (FINDINGS 55).
 Amended end of session 24: G1 done (FINDINGS 56).
+Amended end of session 25: P4 HALF done (FINDINGS 57).
 
 **THE COMPLETION TARGET IS M3, THE VERTICAL SLICE** (USER DECISION): one scene
 tree — a decision point, two outcomes, a death clip — with audio, streaming from
@@ -159,7 +160,39 @@ period from `htotal - 8`), so the tree's "1/55.46 s granularity" was 1/56.69 s
 throughout. No 68000 cycle figure moves — the CPU clock is unrelated to the
 screen — but anything paced by the raster does. 54.5.
 
-**P4. Real transport.** Drive the MB89352 instead of a host file. **Session 23
+**P4. Real transport. HALF DONE, session 25 — FINDINGS 57.**
+~~Drive the MB89352 instead of a host file.~~ `src/player/scsi.i` selects a SCSI
+target and issues READ(10) on the 68000, with no IOCS and no host in the
+transfer path: **4,096 B from LBA 0 and 2,048 B from LBA 1000, both byte-exact**
+against the host's copy of the same volume.
+
+**This item was listed as blocked and was not.** Session 21 recorded "MAME's
+`x68000` has no MB89352 path"; `-exp1 cz6bs1` instantiates one, and FINDINGS
+32.4 had read that card's DMA glue back in session 9. The real gap was the 8 KB
+`scsiexrom.bin` MAME needs to instantiate the card and **the player never
+executes**; a blank placeholder on a separate rompath settles it. **B3 still
+wants the real ROM's bytes** and is untouched by this.
+
+**What is left is the half that decides the project**, and it is now two pieces:
+
+**P4a. A DMAC configuration that HOLDS THE BUS.** Unchanged as the first job,
+but 57.3 changes how it can be shown: `x68k_scsiext.cpp` glues `$EA0015` so that
+a write is discarded when `exown()` is asserted and DRQ is low, and on this
+machine `exown()` is asserted at idle. Every transfer therefore runs the SPC in
+DMA mode with the CPU moving the bytes through the DMAC's own door — and **MAME
+cannot then distinguish a CPU-driven byte from a DMAC-driven one at that
+address**. "The DMAC held the bus" needs evidence that does not come from
+watching `$EA0015`.
+
+**P4b. `scsi.i` behind `ring.i`'s `XF_*` mailbox**, in place of `stream.lua`'s
+modelled transport, gated on the same pixel-exact 120 frames. The seam already
+exists: `XF_ACK` stops being a word a host synthesises from emulated time and
+becomes one the 68000 bumps when a transfer lands.
+
+`W` **did not move by one clock** and is still the largest open number here.
+
+*(original item, still the standing description of the `W` question:)*
+Drive the MB89352 instead of a host file. **Session 23
 added a second axis to it:** `W` is the clocks stolen per delivered byte, and
 55.3 measured that the player's own request loop gives away 3-7% of the pipe
 before `W` is even asked about. A transport design has to answer both. The `W`