Adamw

Yes we have been working on CAN support for the 2015+ STI.  I have most of it working, so far I have only tested on 3 local cars 2015-2018 and I think they are all NZDM models. On two of those cars everything seems to work correctly,  on the 3rd car the boost gauge seems to read incorrectly sometimes so I still need to go back and do more investigation on that.   
I will send you some test software to try for us as it would be nice to confirm that it works in a JDM model.  I will just get the software assembled into a more user friendly installer package and will private message you when I have it.  

Try dropping the write interval to 500.

Yes of course you will need a PC to set up the ecu.  You will ideally want coils with built in ignitors such as K20 or 1NZ toyota etc, connect the ground wires to the cylinder head, install a ignition switched relay to supply 12V to the coils, and run the trigger wires back to the ecu.  The coil on cyl#1 connects to Ign 1, cyl#2 to ign 2, etc.  In the ecu you will need to change ignition mode to direct spark, set up the dwell table appropriate for the coils and reset base timing and ignition delay.

You only need this if you get knock when you jump on the throttle pedal quickly.  Some engines are quite prone to this but I dont find I need to use it for most engines.  TPS delta is how quickly the throttle is moving.  You can log TPS delta to see typical values that your car exhibits, but as a rough guide a normal gentle gear change might have a TPS delta of around 15%/100ms whereas during a sprited drive you may have a TPS delta of about 60% during a gear change.  

Sorry I had to clean up my google drive a while ago, I must have deleted that video.  I think this might be it:  https://1drv.ms/v/s!AiYbYlZQuRHPxSh4fEj7PSJbW2lG?e=A1NDsE

Most analog tachos are very inaccurate.  Some have significant lag also, so they are less accurate under heavy acceleration than they are at lower acceleration.  There is a multiplier in the tacho settings that you can use to adjust its span but I would say your 300Rpm error at redline is probably already better than most.    

Tacho yes, speedo no.  Speedo is multiplex from the ABS unit.  

Does the V5/6 electronic VSS or the early RS legacy one fit in the V2 gearbox?  

Gracias, lo puse a funcionar!! Cambié la dirección de entrada de aire.

@tbase
I will first explain a couple of things as there are two different parts of the shift involved here.
First the unload for disengagement:
With a car on tarmac for example, to make an upshift you briefly cut ignition, this suddenly reverses the torque load on the dogs because you have gone from "engine driving the wheels" to "wheels driving the engine".  The dogs are easiest to disengage somewhere just as this torque reversal starts to happen. 
In contrast, on a low traction surface when there is significant wheel spin, when you cut ignition to make a shift all that happens is the wheels just slow down closer to the speed of the road surface - there is no complete reversal where the road surface is driving the wheels as it does in the first example.  So typically you need a longer cut to get enough unloading to allow the dogs to disengage.
Generally for this dog unload stage you want as much torque reduction as possible (usually 100% cut).
 
Now, engagement of the next gear:
What you are trying to achieve during engagement of the next gear is to have the dogs on the engine side turning at a similar speed to the mating dogs on the output side so that they have the best chance of the dog making it into the mating gap - and as gently as possible.  If the speeds are significantly different the dogs will just sit there grinding on top of each other or they will actually make it in with a very harsh impact, or they will make it in then bounce straight back out from  heavy impact.  With traditional gear shift strategies you had a fixed cut percentage that was present through the whole shift so it was a balance between getting enough cut to unload/disengage, but not so much cut that the input/output shaft speeds were vastly different.  This was especially an issue on rally cars where the disengagement might need a long cut, but then once disengaged the engine speed was much lower then it needed to be.
 
Related G4X improvements:
The disengage part of the shift (called dog unload) is now separated from the engagement part (called main shift stage), so you can have the 100% cut for best unload and less cut for the engagement stage where you just want enough cut to get the shaft speeds to be similar.   There is a shaft speed matching option, this uses the gear ratios and (optionally) driving wheel speed to calculate what engine RPM is needed to match the shaft speeds for engagement, it will vary the amount of cut to achieve the target.  Using the "gear ratio and speed" target calculation means if the wheel speed changes during the shift then that will be taken into account by varying the target RPM to match.    
I dont have a lot of personal gravel experience but probably know enough to offer some suggestions if you give more detail of the car.
 
@Ronague  I tune a couple of sports cars with geartronics paddle shift controllers on them so can speak from experience with both - I dont think geartronics have any real advantages over G4X now in terms of shift quality.  They do have an autoshift option (but pay extra to unlock) for paddle shift setups that we dont have yet and they have a couple of extra failsafes that we dont have yet such as dual track gear position sensor monitoring.  And in some ways I do kind of like their visual "barrel rotation" GUI thing with all shift settings related to "degrees of barrel rotation", but that's about all.  They can do a nice shift.  Negatives: Analog inputs are quite low resolution, they can only detect about 0.05V change and dither about 0.1V, logging is horrible, documentation is non existent, software is very buggy. 

I dont spot anything obvious.  Your lambda stays close to target under light loads quite often (but not always), say up to around 1/4 throttle.  It mostly gets lean as load increases.  I would start but monitoring fuel pressure.    

@tbase
I will first explain a couple of things as there are two different parts of the shift involved here.
First the unload for disengagement:
With a car on tarmac for example, to make an upshift you briefly cut ignition, this suddenly reverses the torque load on the dogs because you have gone from "engine driving the wheels" to "wheels driving the engine".  The dogs are easiest to disengage somewhere just as this torque reversal starts to happen. 
In contrast, on a low traction surface when there is significant wheel spin, when you cut ignition to make a shift all that happens is the wheels just slow down closer to the speed of the road surface - there is no complete reversal where the road surface is driving the wheels as it does in the first example.  So typically you need a longer cut to get enough unloading to allow the dogs to disengage.
Generally for this dog unload stage you want as much torque reduction as possible (usually 100% cut).
 
Now, engagement of the next gear:
What you are trying to achieve during engagement of the next gear is to have the dogs on the engine side turning at a similar speed to the mating dogs on the output side so that they have the best chance of the dog making it into the mating gap - and as gently as possible.  If the speeds are significantly different the dogs will just sit there grinding on top of each other or they will actually make it in with a very harsh impact, or they will make it in then bounce straight back out from  heavy impact.  With traditional gear shift strategies you had a fixed cut percentage that was present through the whole shift so it was a balance between getting enough cut to unload/disengage, but not so much cut that the input/output shaft speeds were vastly different.  This was especially an issue on rally cars where the disengagement might need a long cut, but then once disengaged the engine speed was much lower then it needed to be.
 
Related G4X improvements:
The disengage part of the shift (called dog unload) is now separated from the engagement part (called main shift stage), so you can have the 100% cut for best unload and less cut for the engagement stage where you just want enough cut to get the shaft speeds to be similar.   There is a shaft speed matching option, this uses the gear ratios and (optionally) driving wheel speed to calculate what engine RPM is needed to match the shaft speeds for engagement, it will vary the amount of cut to achieve the target.  Using the "gear ratio and speed" target calculation means if the wheel speed changes during the shift then that will be taken into account by varying the target RPM to match.    
I dont have a lot of personal gravel experience but probably know enough to offer some suggestions if you give more detail of the car.
 
@Ronague  I tune a couple of sports cars with geartronics paddle shift controllers on them so can speak from experience with both - I dont think geartronics have any real advantages over G4X now in terms of shift quality.  They do have an autoshift option (but pay extra to unlock) for paddle shift setups that we dont have yet and they have a couple of extra failsafes that we dont have yet such as dual track gear position sensor monitoring.  And in some ways I do kind of like their visual "barrel rotation" GUI thing with all shift settings related to "degrees of barrel rotation", but that's about all.  They can do a nice shift.  Negatives: Analog inputs are quite low resolution, they can only detect about 0.05V change and dither about 0.1V, logging is horrible, documentation is non existent, software is very buggy. 

If you dont want to do as the tuner suggested which would be perfectly fine, then you should use the expansion connectors.  You will have to find your own +12V supply but one of the expansion connectors has 5 spare DI's on it and signal ground.  Ground is actually not critical for a digital signal either - you could use chassis ground if easier.  
C29 is not used by the link ECU, and most users dont bother with the VTEC oil press switch on DI2 either.  
Note also DI 1 is connected to both Pin C29 and Expansion pin 8, so if using Expansion 8 ensure C29 is disconnected.

Map attached.  Note the forum host is currently sometimes removing the file extension from attachments.  If you download this file and windows doesnt recognise it then make sure it has the .pclr on the end.  
Trigger offset may be around 0 or 142, try 0 first and check with a timing light, if you dont see the timing mark then try 142.  I cant remember if this triggermode  had the offset built in or not in the final version (test map was from testing with beta firmware).
VQ35HR.pclr

I would start with an update rate around 1-2 at idle, ramping up to 10Hz at higher RPM.  Get the gain as high as you can without overshoot or oscillation.  Start with 1.0 right across, you will probably be able to go significantly higher than that but that should be a good place to start.  

There were very few G4+ JZX plug-ins as they were released just around the time we changed over to G4X.  However the pinout is identical and the G4X manual has a more complete pinout so its probably better to use that anyhow. 
I will attach the map and G4+ manual.     
Toyota JZX100 G4+ Xtreme Plugin.pclr JZX100.pdf

Not to take anything away from you post, I love to see these ideas coming through - but Im just mentioning this in case you werent already aware of the "quick trim" function that has been around since the G4, it will do this same basic calculation and apply the result to the fuel table for you (you still need to manually type in the measured value though).  If you have a log open and the fuel map on the same page, click on the cell in the fuel map that you want to adjust, then hit the M key, it will grab the target lambda from the lambda target table, you enter the measured lambda from the log and it will apply the correction to that cell for you.

There is no MAF signal sent out on the CAN bus with the stock ecu as far as I remember, but there are several torque related channels sent from the ecu to the DCCD, and the factory ECU uses the MAF as one of the inputs to approximate engine torque.  The Link ecu instead uses the "torque management" function to generate the approximated engine torque, (air mass is already calculated in modelled fuel equation).  All of the torque related channels that are generated by the link ecu are very similar to the OEM ecu and I tested this for quite a few months in a large range of operating conditions.  Our base map will be close enough for most mild road engines.  Even with the stock engine and ecu the engine I found the torque variable is already maxed out at less than WOT so increasing engine torque or power above stock should not make a lot of difference to how the DCCD behaves either.  

Ok, if it was a link ECU you wouldnt need to change the ecu at all if you were keeping the vvti loom, you would only need to change a couple of wires at the engine end of the loom and some settings in the ecu.  So we would not have any need for such an adapter. 
Try Mark panic at Panicmade.com, he knows his Toyotas well and does lots of adapter harnesses for a huge range of them, far more than you see on his website.  

This is a forum for support of Link products.  If you have a problem with a haltech dash you should be seeking support from them.  

Sorry I missed your last post.  Yes those coils would be fine as they have internal ignitors.  

Ah, I dont think that is intended, I ask the firmware team to take a look when they are back from holiday.  

Yeah that looks quite usable.  Trig 2 arming threshold probably wants to be down around 0.2/0.3V to still be reliable when cranking slow on a cold morning or partially flat battery etc.  

It looks like it has never been calibrated or tuned as it just has our base map offset values in there.  Go to VVT>cam angle test, set it to calibrate and it should do an automated calibration and turn itself back to off when done.  Then check if the VVT is working.  Your exhaust target table only has zeros in it so you will need to put some negative numbers in it to confirm it is working.  Then both targets will need to be tuned.  

Yeah that should be correct.  The only thing I can think of is check that the Error high and error low settings are set correctly.  I have seen some cases where they had been set reversed - ie they had error low at 5V and error high at 0V so the input was always in error mode.  
Temporarily set error low to 0.00 and error high to 5.00 to disable fault detection.
If still no good from there I would connect a multimeter instead of the ecu to see what that shows.