As I said above, I tested with both dynamic and normal core with the same result. Incidentally, DOSBox's (and Boxer's) default cycles=auto and core=auto settings result in normal core and 3000 cycles for most Sierra games, but Gabriel Knight's protected mode kicks it up to dynamic/max.

(GOG.com's rerelease of the CD version uses dynamic/max and appears to be patched to at least 1.0a. I found several GOG support forum posts about the police station timer bug on Day 6, but none concerning the Day 5 freeze. I haven't played far enough with it to see if it exhibits the bug myself.)

I'm sourcing an original floppy version of the game now, and will play through to the relevant point and post the saves. My previous test copy had 'dubious parentage', and the Sierra binary patches applied to it with a few warnings about newer/modified files being left unpatched; that might explain why the NRS patch I applied afterwards didn't fix the bug. In any case, that means its saves aren't trustworthy.

Using the v1.0 floppy version of the game, installed from eleven high-density 3.5" floppies, the attached save was made at the end of Day 4. After restoring, click the ? icon on Grace to Request Research on both available items, exit the conversation choices, then Day 4 will end and Day 5 will begin. Note that other versions of the game, including the v1.0b patched floppy version, will probably refuse to restore from this save.

Shortly after Day 5 begins the phone rings, and it's at this point that the game can seem to hang in DOSBox. In 0.74 and SVN with cycles=auto, DOSBox is using max cycles because of the DOS4GW extender, and the game gets stuck where Gabriel does not proceed to answer the phone; however, it is possible to press the spacebar and skip the phone call sequence and continue with the game. When using fixed cycles of 6000 or so, the phone call sequence proceeds normally.

Thanks. I am very aware of how version specific that SCI saves can be. My copy is version 1.0, so I should be able to use them if I reinstall without patching. I never use auto core or cycles with any SCI games before the later SCI32 games, where even the interpreter is 32-bit. Those don't really have the speed bugs of the earlier games and can have too slow load times with my standard starting point of cycles for SCI games.

I haven't located a 1.0-floppy version of the game yet, so couldn't load your save, but I did locate a more trustworthy 1.0b floppy version (which had not yet been updated with GK1_NRS) and tested that with my previous set of savegames.

This version exhibited the Day 5 bug at max cycles and at 8000, but didn't at 6000 (normal vs dynamic core made no difference). So, behaviour consistent with your 1.0 version.

Here's the kicker though: updating this version of the game with the files from the GK1_NRS patch made the bug occur at any CPU speed: rather than fixing the bug at any CPU speed, it actually regressed.

The testing was done in Boxer 1.2.2, which is equivalent in emulation to DOSBox 0.74.

Keep in mind that the NRS patch had not been thoroughly tested. He had just released it when Endroz first released his installers. He removed them from his site shortly afterwards so as not to compete. A couple of the NRS patches do introduce other issues, which is why the ScummVM team block the usage of them when running them in ScummVM. Another thing to be aware of is that application of patches (official and fan made) to the wrong version of the SCI games can break the games. This is why I have version detection built into many of my installers.

I see, so Endroz's installer for GK1 actually uses different replacement scripts than the GK1_NRS patch? Are those available separately anywhere?

I'm pretty sure my new (working) copy of GK1 was patched to 1.0b and no further; given its behaviour, I suspect my previous copy of the game had been pre-patched with 1.0b and GK1_NRS and I'd simply re-applied all the patches to a game that already had them without realising. How would one tell for sure which (official) version the game is?

In any case I'll make Boxer 1.2.3 auto-configure GK1 for 6000 cycles, since this avoids the Day 5 freeze in unpatched versions (along presumably with the later timing bugs, which appear to be more forgiving).

No, they are not available separately and his installer only installs the Win16 version of the game. I once tried to copy over the missing DOS files to an installation from his installer, but it was too buggy to be playable. Curious, as both interpreters use the same script files.