1.0.19

Fixed

• The host-executable hold-back announced in 1.0.17 was not in the binary. The test lived in the diagnostic enumerator, which runs only at detach; the function that writes the fourteen bytes never had it. skip_exe was parsed, logged and echoed into the config, so a report showed skip_exe=1 while the executable was patched on the first scan. It is now in the loop that patches. 1.0.17's notes described behaviour that could not occur.
• CfgReload ran on the render thread every frame when the bridge never delivered one — a file open, parse and close inside the lock the game's own threads contend for. It was throttled on a frame counter that stops when delivery stops. The throttle is on the clock now.
• The game's own DLSS was suppressed on three frames into which nothing was written. BridgeWillDeliver is answered before the frame runs; BridgeFail left frame_ready standing until the third failure. It clears on the first now.
• The rebuild pause could never escalate. Its counter was zeroed on every resume, so a shape that never settles rebuilt eight times every three seconds for the whole session, silently.
• msaa_reported was cleared by the opposite condition, so a game alternating sample counts traded two messages forever.
• DLSS.Render.Subrect.Dimensions were swept into 1.0.18's mirror-only rule. They describe the region this feature is asked to process, which this bridge knows exactly. Restored.

Changed

• NGX results are named, not printed as hex: 0xBAD0000B, UnableToInitializeFeature.

Notes

• Baldur's Gate 3 exports NGX from bg3_dx11.exe too. With the executable held back its calls still arrive through the driver's loader and everything works — the first evidence that holding it back costs nothing.
• Three of these were one shape: a state reset at the wrong point of its own cycle.
• Verified on Baldur's Gate 3. The Elder Scrolls Online mitigation is still unverified there.