diff --git a/.gitignore b/.gitignore
index 654e81f..e2a056b 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1,9 @@
 # Developer hard-drive dump — not part of the release tree
 /sda4/
+
+# Emulator working files (downloaded binaries, game image, captures)
+/emulator/dosbox-x/
+/emulator/image/
+/emulator/*.png
+/emulator/dbx_*.txt
+/emulator/screen*.png
diff --git a/emulator/.gitignore b/emulator/.gitignore
new file mode 100644
index 0000000..bc0527f
--- /dev/null
+++ b/emulator/.gitignore
@@ -0,0 +1,6 @@
+# Local working artifacts
+dosbox-x/
+image/
+*.png
+dbx_out.txt
+dbx_err.txt
diff --git a/emulator/PHASE0-RESULTS.md b/emulator/PHASE0-RESULTS.md
new file mode 100644
index 0000000..0381a40
--- /dev/null
+++ b/emulator/PHASE0-RESULTS.md
@@ -0,0 +1,121 @@
+# Phase 0 — Baseline Boot Test: Results
+
+**Status: complete.** The DOS game boots on Windows 10 (via DOSBox-X) all the
+way to the VPX link protocol and fails exactly where the plan predicted,
+confirming the emulation architecture end-to-end.
+
+## Environment
+
+- **Emulator:** DOSBox-X `v2026.06.02`, MinGW 64-bit build (the standard build,
+  *not* the `-osfree` variant — that one disables the built-in DOS shell and
+  cannot run an autoexec). Extracted to `emulator/dosbox-x/`.
+- **Host:** Windows 10, cockpit-class PC.
+- **Toolchain present** for later phases: MSYS2 mingw64 (`gcc`/`g++` at
+  `C:\msys64\mingw64\bin`).
+
+## Game image selection — a version-matching gotcha
+
+The first image (from `sda4/BTLIVE`, the "release" install) **fails a version
+check before reaching the renderer**:
+
+```
+BattleTech v1.0.1.4
+Error - Resource file btl4.res v1.0.1.2 is obsolete?
+```
+
+The `BTL4OPT.EXE` in BTLIVE is newer than its `BTL4.RES`. The resource version
+is the **first 4 bytes** of `BTL4.RES`:
+
+| Install | BTL4.RES version | Notes |
+|---|---|---|
+| `sda4/BTLIVE`   | `1.0.1.2` | **mismatched** — EXE expects 1.0.1.4 |
+| `sda4/BTDAVE`   | `1.1.0.4` | self-consistent |
+| `sda4/BTRAVINE` | `1.1.0.6` | self-consistent; **matches repo source** |
+
+`BTRAVINE`'s `1.1.0.6` matches this repository's
+[`CODE/BT/BT_L4/BTL4VER.HPP`](../CODE/BT/BT_L4/BTL4VER.HPP)
+(`MAJOR_DATA_VERSION 1`, `MINOR_DATA_VERSION 6`), so **BTRAVINE is the
+development image**: its EXE, RES, content, and this repo's source are all the
+same build. (For production deployment we will revisit whether to ship the
+BTLIVE content set with a rebuilt/renumbered RES; not a blocker for emulator
+work.)
+
+The image is assembled at `emulator/image/` from `sda4/BTRAVINE`, plus the
+Borland 32-bit runtime (`32RTM.EXE`, `DPMI32VM.OVL`) which every launch needs.
+
+## Baseline result (stock DOSBox-X, no VPX device yet)
+
+Config: [`baseline.conf`](baseline.conf) — replicates the pod environment
+(`L4CONTROLS=RIO,KEYBOARD`, `VIDEOFORMAT=svga`, the production `DPLARG` with
+`/device 0x150`), serial ports disabled, then runs `btl4opt.exe -egg test.egg`.
+
+Console output:
+
+```
+32rtm.exe -x
+32rtm version 1.5 Copyright (c) 1992-94 Borland International
+32rtm resident
+btl4opt.exe -egg test.egg
+BattleTech v4.2
+BTL4Application::BTL4Application
+RIO never came back from check request?
+Protocol error : length 65535 too big, length_word 0xffffffff
++++ERROR : rcv_protocol fail during iserver handling, xmits=0
+```
+
+### What each line proves
+
+1. **`32rtm resident`** — the Borland DPMI/32-bit protected-mode runtime loads
+   and runs under DOSBox-X. (Resolves the "DPMI/32RTM incompatibility" risk in
+   the plan: **low → none**.)
+2. **`BTL4Application::BTL4Application`** — the game passed the RES version
+   check and constructed its top-level application. We are *past* startup
+   validation and into hardware bring-up.
+3. **`RIO never came back from check request?`** — the game polled the **RIO on
+   COM1** (`L4CONTROLS=RIO`, driver `CODE/RP/MUNGA_L4/L4RIO.CPP`). Serial was
+   disabled in this probe, so no reply — expected. Confirms the COM1 control
+   path is live and exercised at boot. (Phase 5 will pass COM1 through to the
+   real RIO.)
+4. **`Protocol error : length 65535 too big, length_word 0xffffffff`** — the
+   game read the 32-bit length/route word from the **link adapter at I/O port
+   0x150** and got `0xffffffff`. DOSBox-X returns `0xFF` for reads of unmapped
+   I/O ports, so the C012 status register (`0x152`) always reads "ready" and
+   the data register (`0x150`) always yields `0xFF`; four such reads assemble
+   `0xffffffff`. `VR_COMMS.C` computes `nb = length_word & 0xffff = 65535` and
+   rejects it — *the exact code path and constant named in the source we
+   documented in the plan.*
+5. **`rcv_protocol fail during iserver handling, xmits=0`** — the iserver/boot
+   handshake failed with zero successful transmits, and the game aborts.
+
+## Conclusion
+
+The failure is **precisely** the plan's predicted Phase 0 outcome: the game
+runs unmodified on Windows 10 and reaches the VPX board's link protocol, failing
+only because no board answers at port 0x150. Two architectural assumptions are
+now empirically confirmed rather than inferred:
+
+- The host↔board interface really is the polled link adapter at **0x150**, and
+  the game's failure is a clean protocol read of `0xffffffff` — exactly what an
+  emulated C012 FIFO must instead answer correctly.
+- The **RIO-on-COM1** control path executes at boot and will be satisfied by
+  DOSBox-X serial passthrough (Phase 5).
+
+## Reproduce
+
+```
+cd emulator
+dosbox-x\mingw-build\mingw\dosbox-x.exe -conf baseline.conf
+```
+
+(The game runs, prints the above, and pauses. `image/` must contain the
+BTRAVINE build + `32RTM.EXE`/`DPMI32VM.OVL`.)
+
+## Next: Phase 1
+
+Build a DOSBox-X I/O device claiming ports `0x150–0x157` that **logs every read
+and write** and returns configurable C012 status/data, so we can capture the
+game's exact boot conversation (reset → transputer monitor → i860 segments →
+version handshake) from the *shipped binary* and confirm the register map
+against `LINKIO.C`. This requires building DOSBox-X from source (MSYS2 mingw64
+is present); the device is a self-contained `IO_RegisterReadHandler` /
+`IO_RegisterWriteHandler` module.
diff --git a/emulator/baseline.conf b/emulator/baseline.conf
new file mode 100644
index 0000000..2cf2dcf
--- /dev/null
+++ b/emulator/baseline.conf
@@ -0,0 +1,26 @@
+[sdl]
+output=opengl
+[dosbox]
+memsize=32
+machine=svga_s3
+[cpu]
+core=normal
+cputype=pentium
+cycles=20000
+[serial]
+serial1=disabled
+serial2=disabled
+[autoexec]
+mount c "image"
+c:
+set L4CONTROLS=RIO,KEYBOARD
+set L4TIMER=
+set BLASTER=A220 I5 D1 H5 P330 T6
+set VIDEOFORMAT=svga
+set TEMP=c:\
+set DPLARG=/tranny~.\vrendmon.btl~/i860~.\vrnostex.mng~/device~0x150~/video~svga~/pipes~1~/qual~0x14~/system_tex~0~
+32rtm.exe -x
+btl4opt.exe -egg test.egg
+32rtm.exe -u
+echo GAME-DONE-RC=%errorlevel%
+pause