Awesome

Awesome

Want to see how it performs. I have 2 use cases for this:

1. Improved external perimeter corners as typically external perimeters are printed using slower speeds compared to infill
2. Improved overhang performance as again these are printed slower than the perimeter speeds

Number (2) has been disproven as a benefit as klipper has a micro stutter when changing PA, causing the reverse effect.

I am now testing (1) as the micro stutter shouldn't matter when transitioning features anyway as you're starting to print from a stand still anyway. Hopeful that this will be the major use case but will see following testing. If (1) is also disproven then I will shelve the feature as not needed and having an approximate PA being good enough for most cases.

I'll be updating the original post with my test results, but always welcome community testing as I only have 2 printers :P

This is awesome!
I thought it was more relevant to accel instead of speed.
But you testing seems indicate the speed affect more.

Do you use same accel for all the features in your tests?

This is awesome! I thought it was more relevant to accel instead of speed. But you testing seems indicate the speed affect more.

Do you use same accel for all the features in your tests?

Nope I'm using varying accelerations as below:

but I guess my accelerations are not sufficiently different between perimeters to show the effect of acceleration on PA. The only place I go high is internal solid infill. The rest are reasonably conservative (in order to remain within the klipper input shaper envelope).

The above PA profile was tuned for 3k accelerations. From the test I've done, higher acceleration does indeed affect PA but more likely than not it's a linear correlation. So what I suspect is happening in my test situation is that the faster speeds use lower PA anyway and that is good enough to prevent issues showing on the internal solid infill which also uses very high accelerations.

I wonder if this would also improve performance for scarf seams due to the change in speed and flow rate. Does the changes in this PR affect scarf seams? If so, I'll try some tests targeting scarf seams :)

So... Digging a little more:

https://klipper.discourse.group/t/pressure-advance-smooth-time-skews-pressure-advance/

From Kevin:
Smooth time impact on PA

"It may be possible to come up with an alternative smoothing system [the smooth time approach] to reduce the amount of over-extrusion during slow speeds. (Possibly by giving greater weight to the desired extruder position at low toolhead speeds.) It’s not immediately clear how to do that though."

"It seems an acceptable (smooth time) value is dependent on acceleration - the higher the acceleration the lower the smooth time should be in order to get acceptable results."

So it appears we actually have a double whammy and this (the effect of smooth time) is what we are compensating for. I may extend the feature to also consume an acceleration profile for a specific speed value and use the two curves to interpolate between.

For reference I have the smooth time set to 0.01 on my printer and this may be the reason why the curve is not ridiculously steep or not as visible as variations between different feature accelerations

More research is necessary.....

I wonder if this would also improve performance for scarf seams due to the change in speed and flow rate. Does the changes in this PR affect scarf seams? If so, I'll try some tests targeting scarf seams :)

No it won't as we must not be changing PA during a feature being printed as this causes proven scaring. So its a no go for this.

I'm also thinking it might make more sense to use volumetric speed instead of print speed here. At the end of the day, it's all about predicting and compensating for the pressure inside the nozzle. Wouldn't volumetric speed be more precise? If it works, it could make line width, layer height, and even nozzle size agnostic!

Test prints with a 10k acceleration target done.

At 50mm/sec there is no difference between the 3k and 10k acceleration PA test.

At 100mm/sec the optimal PA drops from 0.029 to 0.026

At 150mm/sec the optimal PA drops from 0.026-27 to 0.022

So there is for sure another correlation there.

For volumetric speed I need to do PA tests for the same speed and acceleration using different layer heights - that should keep the time dimension constant (speed/acceleration) while only looking at the flow rate.

To be honest I'm still trying to triangulate where that non linearity is happening from - if we understand that then we can model it. So far candidates are:
1. The smooth time parameter - this introduces a deviation from optimal PA that can be expressed as a delay in the PA reacting, causing gaps/blobs. This appears as different PA values against speed & acceleration (I think)
2. A non linearity of the PA depending on nozzle pressure that is not linearly proportional to speed. Extruder position of a given PA value is linear to the speed right now in Klipper - pa_position(t) = (nominal_position(t) + pressure_advance * nominal_velocity(t)) then smoothed over the smooth time and integrated.

For number (2) this is interesting. I would expect that

1. A higher PA is needed the higher the nozzle pressure.
2. Higher speeds would create a higher nozzle pressure due to higher consequential flow rate.
3. However what I'm seeing from my tests is the exact opposite - that a lower PA value is needed the higher the nozzle pressure for a given nozzle size

However from my experience with a 0.25 nozzle, I use a PA value there of 0.048 compared to the avg. of 0.030 or so for the 0.4 nozzle. Which makes sense as higher inherent pressure needs higher PA.

However look at this part of the curve here:

Maybe the reason it is flattening is because of the increase in pressure reducing the effect of smooth time on the needed PA compensation....

And maybe the reason the curve isn't flattening as much for higher accelerations is because the smooth time effect is larger than the pressure non linearity?

What I can tell for sure using experimental data so far:

1. PA has a relationship with speed
2. PA has a relationship with acceleration, but the effect appears to be near constant or slightly increasing above a certain speed.

We need more tests and I need more brains for this!!

So I’ve run a set of PA tests with varying flows (using different layer heights, keeping speed and acceleration constant) and there definitely is a correlation - lower flow needs higher PA or it looks that way.

It’s getting a bit late here so I’ll post results tomorrow

I ran a quick and dirty test, even though my printer is probably not the main audience for this feature as its a slightly modded ender 3, and I can see some marginal but promising results. My inner and outer walls have smaller/thiner gaps in between each other, there are smaller pinholes for my internal solid infill, and my initial layer is more consistent and no longer has a flattened out edge for the circle that makes the hole of the orca cube. I did also notice slightly worse overhang performance like you already mentioned.

I ran a quick and dirty test, even though my printer is probably not the main audience for this feature as its a slightly modded ender 3, and I can see some marginal but promising results. My inner and outer walls have smaller/thiner gaps in between each other, there are smaller pinholes for my internal solid infill, and my initial layer is more consistent and no longer has a flattened out edge for the circle that makes the hole of the orca cube. I did also notice slightly worse overhang performance like you already mentioned.

You shouldn’t have a difference in overhang performance as I’ve disabled the code that was causing the Pa to change before and after an overhang. Curious to what you’ve seen so I can fix it - any chance of a before/after pic for the overhang region?

Good morning everyone!

A bit of an update on flow vs PA.

Below is a graph plotting the different PA values as a function of flow rate.
Volumetric Flow speed here was changed using variable speed.

Volumetric flow speed based changed using variable layer height:

A couple of observations:

1. This graph tells us lower flow -> higher PA, higher flow -> lower PA.
2. The lowest point in that graph is the 50mm/sec speed -> 3.96mm3/sec flow rate
3. When reaching nozzle flow limits (for me its around 20mm/sec) PA appears to break down and reduce significantly
4. There is some correlation between the two methods (even though this is based on a sample of just 2 points) - the 12.93 flow using LH to generate it results in 0.023 PA for 10k accel. The 11.87 flow using speed to generate it uses similar 0.023 value. Same for the 15.86 and 18.83 point -> they both are around the 0.021 mark. Please bear in mind that I can only get so accurate with the PA. we are talking 0.01 delta's here which are a bit subjective too.
5. When using my 0.25 nozzle I noticed that my PA is around 0.42-0.45. This was calibrated at 140mm/sec which is 3. 12mm3/sec flow. This does show that this may actually work as a single PA curve across nozzles (?). The calibrated PA value and what I'm getting out of this model appears to be indeed very close for a lower flow nozzle even though the speeds are higher...

Community ask: It would be great if someone with a 0.6 nozzle could verify that their PA is correlated to nozzle size inversely. ie do you use a lower PA for a larger nozzle (with similar geometry) or not? What has the community observed?

PS. the only reference I've found related to PA and nozzle size scaling was here: https://www.reddit.com/r/klippers/comments/14dhj2y/comment/joq929v/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button and the user does indeed say that there is an inverse relationship -> larger nozzle -> lower PA which would validate the above results.

Converting the code to base the extrapolation on flow is a piece of cake (the interpolator doesn't care what values you put in anyway), so once we have some positive confirmation that this trend continues with larger nozzle sizes (reducing PA the larger the flow rate) I will convert the code from speed based to flow rate based which would have some very cool side benefits if the correlation holds:

1. One PA calibration and use it across all nozzle sizes
2. One PA calibration irrespective of Line width/Layer height

I use significantly lower pressure advance with 0.6mm nozzle compared to 0.4 when printing. I mainly print PETG, and 0.6mm nozzles have a PA value of 0.012 while for 0.4mm it's 0.045. So seperate PA curves for each nozzle size is probably still needed, probably due to the viscosity of PETG so it has a much harder time going though 0.4mm nozzles?

You shouldn’t have a difference in overhang performance as I’ve disabled the code that was causing the Pa to change before and after an overhang. Curious to what you’ve seen so I can fix it - any chance of a before/after pic for the overhang region?

It's probably because I didn't spend much time making sure that the PA values were the best ones to use and using an old and crappy filament. I'll do some more through tests once I have some good filament.

I know I need to make a calibration test for it as it takes a while to run 16 or so PA tests to get a good profile.

in the mean time also check out the wiki page for the feature - it explains in detail how to run it.

https://github.com/SoftFever/OrcaSlicer/wiki/adaptive-pressure-advance

oh im so happy to see this PR. great work ! i have been having this issue for a while. this will be awesome :)

As from what I learned, printing speed is not the problem but acceleration.

Bambu's Pressure Advance Algorithm does in contrast to Marlin's Linear Advance not modify print acceleration but only extrusion (as far as I know).

Print acceleration also has to be reduced the higher you set the K value, to give the changes in extrusion speed enough time to compensate for the filament compression that happens between the feeder and the tip of the nozzle. This is why they've added this to Linear Advance. They had realized that modification of only the extrusion speed is not enough to compensate for higher filament compression, caused by soft filament and/or higher speeds respectively higher acceleration. I've read this in the comment to the commit where it was accomplished.

High K values are needed for flexible filament like TPU. I need depending on filament brand, shore hardness, nozzle temperature, hot end (E3D Obxidian e.g. is totally different from the stock hot end in this regard), nozzle diameter and layer height a K between about 0.15 and 2.5(!!!). This only works if I also reduce acceleration (and also jerk) radically.
E.g. for Recreus Filaflex 60A I have to reduce acceleration to about 200mm/s² and jerk to 3 otherwise PA would have no effect and also it would jam the extruder, but that's a different issue.
TPU
I'd suggest to just add some parametrizable curve function to the filament profile which modifies K depending on nozzle diameter and current layer height, because all the other parameters (printing temperature, hot end ...) usually seldomly change and also reduce acceleration the higher K gets to simulate what Linear Advance does.
This would also be easier to configure for the user than the current monstrosity ;) in OrcaSlicer 2.2.0 Beta.

The algorithm looks at two parameters:
Volumetric flow rate
Acceleration

speed is used as a proxy in order to run the calibrations but the model is based off volumetric flow rate.

the higher the layer height, the greater the flow rate for the same speed and similarly with nozzle sizes. So I believe you can use the same model for this. You also have the option to completely disregard acceleration or flow rate from the calibrated model - if you set the values to be all the same the model ignores them.

I`m so curious if now with this features will have better seams when using 0.6 nozzle and 0.25 layer height

Possibly. My seams have been absolutely perfect with it enabled but then I don’t have a 0.6 nozzle to test it. Below is with a 0.4

it looks fantastic! nice job! what filament is that it looks beautiful!
I will test it with the .4 and black petg, this is what i have loaded, and after when i will switch nozzles i will post the results. thanks for the amassing work!

The algorithm looks at two parameters: Volumetric flow rate Acceleration

In my understanding it's not the flow rate that affects the K parameter as the PA algorithm is designed to have a K parameter that's independent of the flow and thus the printing speed, but the resistance of the the filament to being pressed out of the nozzle.
I tested this by printing lines with different width but using the same nozzle diameter which does increase the flow rate but didn't influence K. It is mandatory that it works this way, because otherwise the Arachne wall generator would produce an utter mess.

Every parameter that increases that resistance causes a higher filament compression inside the extruder (between the feeder and the ground where the molten filament is laid on, like the plate or the previously printed layer). These parameters are mainly the liquidity of the molten filament which depends on the nozzle temperature, the nozzle diameter and the layer height, where with wider nozzles and higher layers K quickly converges towards zero.

PA tries to compensate for the extrusion latency caused by this filament compression. As I understand, it's not a physically correct implementation but only an approximation, but does not depend on the flow rate. The only issue is that the faster you print, the more visible will the defects from a not perfectly calibrated K parameter become, so optimally you should calibrate K using the maximum printing speed you will be using. This is also what Marlin's Linear Advance documentation essentially is saying ("Fast Printing Speed and Slow Printing Speed should be significantly different or the K-factor effect will barely be visible.").

During my testing I've never experienced the flow rate depending problems that caused this pull request and I was one of the first who embraced LA and later PA where other people laughed at me when I was trying to tell them that it's not possible to print a line that maintains a homogeneous cross section throughout its whole length without using and properly configuring one of those two algorithms.

So IMHO what we need is an implementation that accounts for just nozzle diameter and layer height and a function which reduces acceleration (and maybe also jerk) when K increases to simulate what Marlin's Linear Advance does but is missing in Pressure Advance so that PA will have enough time to compensate for very high filament compression (high K values) which happens when printing TPU and other flexible filaments.
I currently have to maintain one filament profile for each filament, nozzle diameter and layer height combination and many process profiles for TPUs with different softness just to use lower acceleration values depending on shore hardness.

It’s Polyterra fossil grey PLA. It’s the most revealing filament I have for imperfections as it catches the light with any uniformity variation!!

here you go these are the pictures:

0.16 LH/0.4 nozzle / 3k external accel / 150mm/sec external perimeters. Voron 2.4 350 with Galileo 2 extruder and Voron revo plus ERCF v2

This has been merged. Further discussion should take place in the Orca Discord server.

I work with a Bambu P1S printer, have recently bought a roll of transparent PETG from R3D, and decided to go through the tedious process of calibrating adaptive pressure advance. I have already calibrated flow before dealing with PA, and used the pattern test for easier PA inspection. After ~4 hours of printing and the adaptive PA model fed into the slicer, I created a simple triangular test model to evaluate the effectiveness of my PA model. To my frustration, it appears that static PA yielded better print results.

Transparent & glossy filaments are difficult to visually inspect. To pick my PA value, I held up the part against my black phone screen, find a light spot, and basically look at the "shadow" of the part that I'm inspecting. I start with the larger PA values, where wall separation creates holes that the light could shine through. Then I look at the edge of the 90° turn, and find the one with a slight indentation like the one in the picture. I select this shape in particular because I have printed a 1cm^3 cube separately, and the bottom few layers look exactly like this sideways.

My filament has a max volumetric speed of 23 mm^3/s, and my slowest volumetric speed is around 5mm^3/s. Also I would like to cap my acceleration to 4000 during normal printing. Therefore, I have tested flow 5.5, 11, 22. The lowest acceleration that could be tested is calculated with v^2=2ax, and I used x=25mm given the herringbone length is around 30mm. Here is the PA values that I have come up with.

Besides the above 14 lower precision PA values, I also tested flow 15.96 and acceleration 5000 (the default PA pattern settings) with 0.001 step length, and got a PA value of 0.036.

For evaluation, I created a testing model by copied the dimensions of a single herringbone pattern, and connected the open endings to create a closed loop, shown below.

Then, I put 6 of the triangles in the same plate, with varying layer height and acceleration, to simulate actual print conditions. Print speed is the same across the triangles.
For easier identification I numbered them 1 to 6 (see pic). Triangle 1-4 have varying layer height and same acceleration (1500).

• Triangle 1: model height 1.21mm (0.32mm * 3 + 0.25), flow 21.56 for outer wall, 21.56 for gap infill
• Triangle 2: model height 1.05mm (0.20mm * 4 + 0.25), flow 14.78 for outer wall, 21.56 for gap infill
• Triangle 3: model height 0.85mm (0.12mm * 5 + 0.25), flow 9.27 for outer wall, 12.67 for gap infill
• Triangle 4: model height 0.73mm (0.08mm * 6 + 0.25), flow 6.32 for outer wall, 8.28 for gap infill

On the second row, triangle 5 have acceleration set to 500, triangle 2 set to 1500, and triangle 6 set to 4000. These three have the same layer height (0.20mm). I have set all other slicing parameters to match the ones used in the default PA pattern test.

In order to verify the effectiveness of adaptive PA, I printed the triangles with 1) adaptive PA and 2) static PA = 0.036.
I've photographed the 90° angle and the left 45° angle under 30x magnifying glass. The right 45° angle is affected by seam, thus I am not showing it here.

1st 90° is not sharp enough, 6th 90° has severe wall separation

constant PA 90° ones look more consistent

6th one with adaptive PA is not sharp enough; constant PA 45° ones look more consistent; (The blue stuff is my marker)

Also, my evaluation project: (I have custom G-code, select your own printer preset or use at your own risk)
PETG_adaptive_PA_evaluation_KrisBigK_20250223.zip

Apparently static PA works better in this circumstance. Is it because of a bad PA model? Do Bambu firmware support this at all? Any help would be appreciated.

Bambi doesn’t behave the same way as klipper for Pa. I think it adapts internally already but I may be wrong. Hence this feature is not needed for Bambu printers.

Here are some of my observations on the Bambu PA implementation, and how it could be improved:

https://www.youtube.com/watch?v=zZjimJ6Gp7w

And also this issue I opened up with them because the extruder seems to skip steps before reaching an ideal PA value in some cases:

bambulab/BambuStudio#2922

He, @igiannakas
I believe I may have found an inaccuracy in the previous dynamic PA implementation. After running more extensive tests using the PA tower method, it seems that PA is primarily influenced by layer height and acceleration, while the impact of volumetric flow might be indirect. I’ve opened a new issue with detailed test results and observations
#8878

The PA K value, which is defined as the compression length of the filament inside the extruder between feeder and nozzle tip per velocity unit in mm/(mm/s) depends on everything that makes it harder for the feeder to push it through the nozzle and thus compresses it more and also on the tightness of the tubes where it has to go through after leaving the feeder, because the wider the tube the more the filament can be bent, which also counts as compression.

This in essence is the softness of the filament, the print temperature, the properties of the extruder like how the hotend is built, the nozzle diameter and last but not least the layer height.

But it neither depends on print speed nor acceleration and thus also not on the flow. This becomes clear when looking at the definition of K which is as I've already mentioned above the filament compression lenght per velocity unit in mm/(mm/s).

The reason why it may seem that it depends on speed and acceleration and thus flow is, that the faster and the higher accelerated you print and the higher the K factor of the filament, nozzle diameter and layer height combination is, the more the feeder has to compensate for the filament compression. This unfortunately puts too much strain on the extruder motor when the combination of acceleration, speed and K gets too high.
I'm not an expert on stepper motors but maybe the stepper driver keeps the movement of the motor within its limits or the motor begins to loose steps or whatever but either way PA will not work anymore and further increasing K will not have the desired effect.

The proper thing to do, would be to lower acceleration and probably also jerk when K gets higher to give the PA algorithm and thus the extruder motor enough time to properly compensate for the filament compression.
This is what Marlins Linerar Advance (LA) does, at least since LA15, because one of their developers recognized this fact and implemented the automatic reduction depending on K within LA in firmware.

To produce the same result I forked this Repo and implemented the automatic reduction of acceleration and jerk inside OrcaSlicer using hyperbolic curves based on values I've figured out by experimenting on my Bambu Lab X1C.
This works perfectly and also lets me e.g. print even quite soft TPU, which naturally needs very high K values, depending on softness of up to about 2.0(!) at quite high max speeds, because the automatic reduction of acceleration and jerk not only "repairs" PA for high K values but also prevents the extruder from clogging when printing flexible materials.

I’m not sure we are getting extruder skipping or slipping since we are decreasing Pa to compensate for filament compression / path constraining when printing at higher flows / accelerations.

Also in klipper setup the variance in Pa is not huge - closer to 0.035-0.028 or so which doesn’t indicate a “break point” in the system. I would have guessed a much bigger delta would happen if there was slippage.

Also the Pa is lower the more back pressure is encountered which to me indicates that the extruder is over compensating with higher accels. If the above hypothesis was correct it would need to increase the Pa value with higher accels as more compensation would be needed. Also because more Pa is needed with lower accels we can’t really further reduce accels as this would create a feedback loop further needing an increase in Pa!!

fyi Bambu uses a different algorithm vs klipper for Pa that based on experiments done over in the discord by users does not benefit from adaptive PA. Probably they are implementing a Pa adjustment internally in the firmware.

@igiannakas you may of course be right that the Bambu PA implementacion is more correct than Klippers but then it's a Klipper Problem ans should be fixed there. I know it's funny, me writing this after having fixed the acceleration issue for high K values also inside OrcaSlicer (in my personal fork and I intentionally haven't pushed a PR since I know that it might be considered being an opinionated approach and my function curves might only be optimally tuned for the Bambu Lab X1C), since I cannot modify the Bambu firmware 😅. But otoh Klipper is open source.

Fact is, K must not depend on velocity or acceleration, if it does, PA is not correctly implemented.

If your adaptive PA implementation is needed for Klippers obviously not correctly enough implemented PA it's of course useful although it's imho too elaborate to be utilized by the average user since you have to do lots of time consuming calibrations to fill the table with enough data points for interpolation.

Correct that’s the theory. That PA tuning should be independent.

klipper implements a linear function to calculate the extruder movement based on speed. Also implements Pa smooth time which basically creates an error rate / deviation between ideal (calculated) extruder movement and actual movement - to prevent the extruder from performing too high too fast moves as extruder movements as a result of PA are unbound.

There is a fork of klipper where smooth Pa is being trialed and the general idea is to get feedback from the slicer implementation to basically identify the core variables and implement in klipper eventually.

What we are seeing though is in reality the slope of the linear model changes based on acceleration and acceleration and flow with a reducing factor / so a linear adjustment is likely not possible. Maybe with anything that changes the back pressure on the nozzle (flow, accel, layer height)

Probably because the filament gets more rigid the faster you print and the closer you get to the hotend melting ability.

My theory is that a higher flow -> more solid filament in the hotend -> less Pa needed to retract and advance it as the amount of molten plastic is reduced.

For the deviation based on acceleration I’m really not sure - but maybe a similar story as with the above as we are basically reducing the time window where the adjustment can happen (?).

But to do so we need data. And to figure out why the linear model is not good enough. And also how to test for it. Hope that makes sense?

@igiannakas sure, this makes much sense!

I actually didn't intend to criticize your code but to make people aware that there is a PA issue when printing high K filaments at high speeds which has its roots in extruder motor limits and that this is not only my observation but has already been tackled by Marlins Linear Advance algorithm.
My workaround for this can be found in my OrcaSlicer fork but honestly I cannot tell if the acceleration and jerk values my easy to use feature, it's just ticking a checkbox in the process profile, generates, work equally well on other printers than my Bambu X1C.

I could only compare Bambus PA against Marlins LA (the Prusa implementation) since a Bambu X1C and a Prusa MK3 (which I've given away recently) are the only printers I had, for experimenting.

@igiannakas Here you will find evidence for what I'm trying to explain. LA in Marlin FW handles it in contrast to Klipper and Bambu PA.
https://marlinfw.org/docs/features/lin_advance.html

I know :) have read it. Marlin doesn’t have the smooth time concept. Klipper uses smooth time as a substitute for limiting extruder movements to the jerk limit.

Btw Bambu is completely unbound on the extruder acceleration. I’ve had instances where I saw extruder stalling when pulling filament too fast out from the extruder during multi material changes. See here: #3094

maybe they have capped it for Pa, not sure; but for regular print moves jerk limits are not respected.

@igiannakas but "smooth time" seems not to be working well for high K values which you need e.g. for TPU but also PETG, PCTG and CPE profit much from slightly lower accel and jerk.

There seem to be more than just the K parameter for use with the M900 GCode. Isn't one of them responsible for configuring the "smooth time". Maybe OrcaSlicee should try to modify this with increasing K values?

For klipper, smooth time is a function to reduce and spread extruder motion over a specific time window.

Typically in klipper you want almost as close to 0 smooth time as possible (ie infinite jerk in marlin), to minimise delta between ideal extruder motion and actual motion. It was introduced as Bowden extruders needed to move too much (due to their high Pa) hence smoothing that motion over time acted as an effective acceleration cap on the extruder.

Basically the higher the Pa and higher the print accelerations, the bigger effect smooth time has.

for tpu / petg material in klipper you can mitigate the stretching of the material by printing with slower extruder transitions (by reducing printing accelerations) - which means lower extruder movements as a result of the needed Pa value to maintain ideal extrusion timing.

You can also further smooth that by applying extrusion rate smoothing in orca, which creates slowdown segments beyond the firmware commanded ones allowing Pa enough time to adjust.

there has been chatter about introducing a test to calibrate smooth time but nothing has come to being yet. Basically you want to reduce smooth time till the gap in a corner observed when you have borderline high Pa aligns with the edge of the corner.

too high smooth time will move that gap before the corner.

I don’t have as much insight on the marlin implementation, and much less so on the Bambu implementation so can’t offer any meaningful advice other than what I’ve seen with my X1C before I sold it :)

@igiannakas question: in Orca, the acceleration control is set by commands such as SET_VELOCITY_LIMIT ACCEL=1000, which as far as I understand applies to all axes including extruder when there is no AXIS argument. isn't this not ideal for PA which sometimes require higher accel on E vs XYZ?

The set accel command doesn’t apply to E in klipper at least from what I understand. That is also the reason Pa smooth time exists - to cap acceleration on the extruder.