NFS III original local CLI route, 2026-11-13
The production CLI supports the existing local original-media route through native renderer enumeration. No private provisioning is needed for this checkpoint. The original catalogue installer, first setup registration and second setup launch succeeded; the personally reviewed native chooser and opened dropdown show Software Renderer and Direct3D Device 1. Renderer selection, race, native Pause/Continue and reviewed driving remain UNPROVED. A fresh pre-selection sample was 1,497,788,416 bytes, below the required strict 1,500,000,000-byte floor. Selection was not submitted; the live CLI was quit promptly. This is a local capacity blocker, not evidence of a missing production process/native-control contract.
Evidence: scratch/runs/20261113-nfs3-local-original-renderer-driving-0800/.
Reviewed images are chooser/initial-frame.png and chooser/options.png.
The source identity is the recorded base-commit.txt plus
relevant-source.patch; frozen.json records 138 path/hash pins, including
all133 existing media entries, their manifest, original executable and current
production module. Original EXE SHA-256 is
0defab3eeb22ee4b6e0007a4d5b26a99d868008ba77e2b9bd3ef770e924548ad;
module SHA-256 is
04c534192ee3fe39e388536c2ce710c12c47a5cd2ca4d95c19f9d00b1e8f0a17.
No source/media archive or media copy was made.
The route uses unchanged test/run.js and registered apps:
--app=nfs3_device_data --reg-export=...runs the original catalogue installer, exits0, and exports118 ordinary registry entries.--app=nfs3_device_setup --reg-import=<catalogue export>runs original setup registration, exits0, and exports121 entries.- A second
nfs3_device_setuplaunch imports that exact unedited native export and displays the chooser. Its registered working directory isC:\Setup\3dSetup. No CreateProcess child was needed for these launches; child-process runtime semantics remain untested here.
All three use --no-build --real-ticks --d3d-worker --live-present, live
loopback control, a330-second runtime limit, and an independent360-second
watchdog observed armed before each launch. Fresh prelaunch decision samples
were1,504,456,704, 1,503,838,208 and1,501,777,920 bytes. Each launch reread the
frozen pins. Registry snapshots were passed unchanged using production
export/import. The registered setup entry supplies only existing installation
DisplayName metadata; native setup itself wrote the default softtri values.
Those native writes do not establish a reviewed renderer choice or race.
Before the sole dropdown click, an actual live observer bound PID955817, HWND0x10001, instance/exports, memory/buffer, encoder/module/worker/control, canvas/window and shared command buffers. The production D3DIM encoder reported ready with zero fallbacks and zero draws. This establishes setup transport readiness, not game-device rendering. The ordinary CLI click invokes balanced renderer mouse-down/up. No fresh capacity sample was recorded immediately before that dropdown click; the next fresh sample caused the pre-selection refusal. Do not treat this helper as a qualified finite driving controller. The retained browser full22 controller has browser owner and 1024x712 frame assumptions; this CLI uses640x480 and requires an explicit, focused actual-consumer adaptation before driving. Its former proofs were not promoted to this different consumer.
Captures directly encode the existing composed canvas without invoking the
CLI PNG command: that command calls presentForRead, which can force DX
presentation. There was no extra Present, guest readback, clock/config/guest
state/AI patch, fake installed state, automatic route or blind input chain.
Only one balanced dropdown click occurred. No key, renderer-row choice, OK,
Pause, Continue or driving input was sent.
An initial helper invocation used an obsolete Node flag and exited9 before
loading the runtime; its actual wait and stderr are retained in catalogue/.
The flag was removed for the first actual native launch. Read-only inspection
requests after first setup had already exited hit connection refusal; no
uncertain input was replayed. These are helper/read-only failures, not new
production defects. No production code changed, so no locked build is needed.
All actual native runs and independent watchdogs have actual parent waits,
code0 and both stdout/stderr EOFs. The chooser's normal quit acknowledged
quitting:true; no owned runtime remains. Box ID NONE, state
NOT_CREATED, expiry null, stop receipt N/A. Protected Pawn0730,
LMarbles0650 and recovery data were untouched. No cleanup, capacity loop,
new box, public deployment, agent or other worker was launched.
Next: with adequate actual capacity, continue through this local production
route, select the reviewed native Direct3D device and commit it through native
OK, then transfer its unchanged export into registered
nfs3_chosen_renderer. Establish actual game-device identities and qualify
the actual CLI finite input consumer before any driving. Personally review
two settled full native Pause/Continue frames and a stable world timer before
one finite interval of at least1500ms, with independent UP/Escape and an
observed held frame before UP. No bend/lap claim is supported by this run.