DKR-R v1.0.5 Beta 15 Playtest 14

HISTORICAL BUILD RECORD
The original findings and limits below belong to the named earlier build.
For current instructions, use docs/README.md and CHANGELOG.md in the
source repository. These old results do not qualify later packages.


REPAIRS
The launcher identified ROMs as dkr.us.v77/dkr.us.v80, but the host mod
catalogue and client resource preparation require us.v77/us.v80. A checked
conversion from the verified revision now connects these contracts without
changing existing ROM caches, installed mods or offline save paths. This
repairs vanilla and modded lobby preparation for all three backends.

Experimental Rollback now links separately isolated US v1.0 and US v1.1
(Rev A) complete replay engines. Guest functions, native bridge state,
input/EEPROM/Pak ownership and private audio helpers are revision-qualified.
The accepted lobby revision selects the engine; exact ROM and bootstrap
checks remain mandatory. No warning was bypassed, and no legacy backend is
silently substituted. Startup still follows the normal game intro/menus.

Each revision has a content-derived replay compatibility identity. Windows
and Linux built from the same closure must agree on this identity. All racers
must install this complete build and select the same supported game revision.
Different ROM byte orders are normalized as before. This does not add ports
for other regional revisions or support arbitrary executable ROM hacks.

HOST-MANAGED MODS
The host freezes the enabled character, legacy-track and Track Lab selection.
Clients review and consent before downloads. Content is reconstructed against
locally imported original ROMs, verified and prepared before lobby admission.
DKR-R never transfers ROMs or native executable code. Cached content is reused;
the separate online profile does not replace the client's offline selection.
Existing transfer, asset and memory budgets remain enforced. The expanded
character roster and its four animated stage positions are retained.

If preparation fails, a persistent modal now shows the Game Pak, backend,
preparation stage and full error. Dismiss it, correct the requirement and retry.
The same failure is recorded in runtime.log instead of disappearing in a
brief popup. Cancellation does not activate incomplete resources.

Online course preparation also accepts the bounded asset tables expanded by
larger character rosters. It uses the supplied vector length; retail pointer
scan bounds, signed asset-ID limits and section memory budgets are unchanged.

Rebuilt candidate: online staging now converts ordinary launcher paths to
the existing private long-path form before creating nested worker directories.
This fixes Windows "create_directory: The system cannot find the path specified"
failures during reconstruction. Normalized paths are also used for host/cache
queries and worker progress streams. No Windows policy change or cache reset
is required; retained mods, content identities and offline saves are unchanged.

Admission handoff correction: a rejected join now retains the authenticated
host settings-sync offer and reports the original rejection/connection reason.
The mod watcher no longer replaces it with a generic disconnect or clears the
existing one-click settings-sync UI. Settings-sync eligibility accounts for
host mods that will be downloaded later, without changing the client's actual
mod identity or skipping consent, checksums, reconstruction or runtime proof.
Client preparation stages no longer retain a previous host-ready label.
Track Lab's reader and network package boundaries now use native paths and
preferred separators, fixing nested payload names such as textures/0.bin
under Windows long paths and non-ASCII course directory names.
Its existing size limits and online content verification remain unchanged.

PRESERVED
Host Prediction and Strict Input Sync remain available and their gameplay
algorithms are unchanged. Offline save isolation, existing rendering/HUD fixes,
Track Lab/Blender support and independent local presentation settings remain.
No protected submodule or RecompiledFuncs/RecompiledPatches source was edited
manually; private payloads were generated into new Patch Pipeline directories.

PLAYTEST REQUEST
Use the same Playtest 14 package on every peer. Check US v1.0 and US v1.1
separately: vanilla lobbies first, then host-enabled characters/custom tracks,
client consent/download, character select, original/custom races and return
to the lobby. Include two-, three- and four-racer matches and two-player
Adventure/boss progression. After leaving, check that offline saves and mod
selections are unchanged. If a stage fails, retain the complete modal message
and newest runtime.log from every affected peer.

Verification is headless; real WAN latency, rendered gameplay and device-
specific performance still require team playtesting. No private ROM, game
checkpoint, save or crash dump is included in either distribution.
