SpaceX progress towards a 100 launch year

[meekGee]

Sorry, it takes time LOL.

One week later, mighty complicated to take a pre-compiled Excel sheet and try to adjust to a clear exponential...

Well I deserve this criticism LOL.

An interesting differentiator would be to graph the "number of launches per 60 days"  sliding window style, over the last 3 years.

(60 days so we average out small logistics bumps)

Does it show the annual step-wise behavior eeergo is postulating?  Or is it continuously increasing?

Because if it doesn't, the logic of "for every 7-launch month they need a 9-launch month" is just part of the growth.  They'll launch 7/mo for 4 months, then 8/mo, then 9/mo, and this represents a 40% annual growth and that's not impossible.

[Robotbeat]

I just haven’t gotten around to a good way to include an exponential fit with a y offset yet. Brute force optimization using least squares doesn’t converge, at least with Libre. Probably I could just fit to the launch rate (derivative of an exponential is still an exponential).

[meekGee]

I just haven’t gotten around to a good way to include an exponential fit with a y offset yet. Brute force optimization using least squares doesn’t converge, at least with Libre. Probably I could just fit to the launch rate (derivative of an exponential is still an exponential).

You don't need an exponential fit to prove your point.

All you need to show is that the improvement in avg launch rate is continuous rather than annually step-wise.

If it's continuous, there's no reason to do the annual averaging eergo is arguing for.

If OTOH it's like he says, constant through the year and then jumps each January, then yes, he has a point.

[M.E.T.]

Currently SpaceX is looking on course for 19 launches by end of March. That translates to 76 launches for the year.

That’s awesome. Why is anyone trying to view that as a negative?

[eeergo]

Sorry, it takes time LOL.

One week later, mighty complicated to take a pre-compiled Excel sheet and try to adjust to a clear exponential...

Well I deserve this criticism LOL.

An interesting differentiator would be to graph the "number of launches per 60 days"  sliding window style, over the last 3 years.

(60 days so we average out small logistics bumps)

Does it show the annual step-wise behavior eeergo is postulating?  Or is it continuously increasing?

Because if it doesn't, the logic of "for every 7-launch month they need a 9-launch month" is just part of the growth.  They'll launch 7/mo for 4 months, then 8/mo, then 9/mo, and this represents a 40% annual growth and that's not impossible.

Again, please refraing from moving the goalposts.

All my statements attempted was to respond to Robotbeat's criticism of xyv's ANNUAL (that's *yearly*) plots showing a pretty clear linear trend, and his confident statement that *those plots* were instead best fit as an exponential.

I clearly stated in my later discussion with him that long-term (that's multiannual) trends are more likely different, although by eye it looks like a toss between a linear and an exponential increase so far, not having the numerical data that, for example, Comga has compiled. This is due to large discontinuities, both positive and negative, in the data, regardless of how you bin/average it.

I haven't postulated any "annual step-wise behavior", since there are larger "steps" elsewhere - so stop moving those goalposts too. Again, Comga has performed a similar exercise in averaging to the slightly different moving average window you propose (chosen at random as far as I can see, and for no good particular reason) of 2 months: he has employed a moving average of the last 10 launches, whenever they occurred, to yield the rate - which gives a pretty noisy curve and would approximate what you're proposing, and a moving average of the last 12 months. Neither show what the OP is scorekeeping -again, yearly rate-, and wrestling an exponential out of the so-derived 3-year-long cadence appears statistically dicey.

As for framing this as a negative: why are some of you obsessed about categorizing everything as good or bad? It either shows a behavior or it doesn't, no big deal about it. When you see a flock of birds out of the window, do you also try to judge if their rate of descent toward a tree canopy is "good" or "bad" according to your preset expectations?

[Robotbeat]

*sigh*

All that linearity over a short time period tells you is that you have some sort of underlying smooth function. Because literally any smooth function can be approximated locally by a line. This is just basic stuff.

You learn about the “local linearity approximation” in your first year of calculus. Basic stuff.

[meekGee]

Sorry, it takes time LOL.

One week later, mighty complicated to take a pre-compiled Excel sheet and try to adjust to a clear exponential...

Well I deserve this criticism LOL.

An interesting differentiator would be to graph the "number of launches per 60 days"  sliding window style, over the last 3 years.

(60 days so we average out small logistics bumps)

Does it show the annual step-wise behavior eeergo is postulating?  Or is it continuously increasing?

Because if it doesn't, the logic of "for every 7-launch month they need a 9-launch month" is just part of the growth.  They'll launch 7/mo for 4 months, then 8/mo, then 9/mo, and this represents a 40% annual growth and that's not impossible.

Again, please refraing from moving the goalposts.

All my statements attempted was to respond to Robotbeat's criticism of xyv's ANNUAL (that's *yearly*) plots showing a pretty clear linear trend, and his confident statement that *those plots* were instead best fit as an exponential.

I clearly stated in my later discussion with him that long-term (that's multiannual) trends are more likely different, although by eye it looks like a toss between a linear and an exponential increase so far, not having the numerical data that, for example, Comga has compiled. This is due to large discontinuities, both positive and negative, in the data, regardless of how you bin/average it.

I haven't postulated any "annual step-wise behavior", since there are larger "steps" elsewhere - so stop moving those goalposts too. Again, Comga has performed a similar exercise in averaging to the slightly different moving average window you propose (chosen at random as far as I can see, and for no good particular reason) of 2 months: he has employed a moving average of the last 10 launches, whenever they occurred, to yield the rate - which gives a pretty noisy curve and would approximate what you're proposing, and a moving average of the last 12 months. Neither show what the OP is scorekeeping -again, yearly rate-, and wrestling an exponential out of the so-derived 3-year-long cadence appears statistically dicey.

As for framing this as a negative: why are some of you obsessed about categorizing everything as good or bad? It either shows a behavior or it doesn't, no big deal about it. When you see a flock of birds out of the window, do you also try to judge if their rate of descent toward a tree canopy is "good" or "bad" according to your preset expectations?

No goal posts are moving.

We're talking about the 100/yr goal, which is roughly 8/month plus starship.

Right now they are at 7, and clearly for every 7 there must be a 9 to reach the goal.

The question is about extrapolating from current performance to a whole-year prediction.

If we assume, like you suggest, that rates remain flat throughout the year, then the likelihood of reaching 100 is zero - the baseline is 84+starship for this year (up from 60) and that's that.

If OTOH we assume continuous increase, then over the course of time 7 wll become 8, then 9 etc.  If it happens at a 4 month rate or so, they'll hit 100.  That rate correspondence to an annual increase of 40%

[eeergo]

I feel a step back is needed:

[Directly answering to xyv's interesting scorekeeping plots with a linear increase sight-guiding trendline]I want to compare that with an exponential curve (that also gets to a 100/year launch rate). I think exponential curve is much more appropriate than linear here.
[...]
And actually it does show an exponential increase.
[...]
An exponential increase is the most natural and obvious one. Exponential increase is from making a whole bunch of incremental improvements that occur roughly stochastically over the year. [...]To make linear work requires just a bunch of overfitting and piecewise assumptions. It’s a crap model and actually over-complicated.

This ^^ is what I pushed back against. It is clearly, demonstrably, a false statement - as 2022's data shows.

I keep repeating: the long-term trend may be different than linear. Exponential, sigmoidal, polynomial, you name it. The YEARLY trend is LINEAR, to a very good approximation, for 2022. For 2023, it obviously is difficult to tell with only 20% of the year in, but so far doesn't break from that trend: it just shows a steeper slope.

The basic stuff your unmovable faith in your preconceptions apparently prevents you from understanding is that the "short time period" in this trend can still be a year. "Short" means whatever your prior allows you to consider it to mean. To go to an absurd case, planetary cooling is well-approximated by an exponential over geologic eras, yet you wouldn't fit a thousand-year period to such a curve, because it would be indistinguishable from a very straight line, for the same reason the horizon looks flat but Earth is spherical. FFS why would you be so hellbent on proving something that is evident from a casual look at the data?

[alugobi]

[Robotbeat]

You’re gonna make me follow through on this, arent you? LOL, well fair enough. But not right now.

[meekGee]

I feel a step back is needed:

[Directly answering to xyv's interesting scorekeeping plots with a linear increase sight-guiding trendline]I want to compare that with an exponential curve (that also gets to a 100/year launch rate). I think exponential curve is much more appropriate than linear here.
[...]
And actually it does show an exponential increase.
[...]
An exponential increase is the most natural and obvious one. Exponential increase is from making a whole bunch of incremental improvements that occur roughly stochastically over the year. [...]To make linear work requires just a bunch of overfitting and piecewise assumptions. It’s a crap model and actually over-complicated.

This ^^ is what I pushed back against. It is clearly, demonstrably, a false statement - as 2022's data shows.

I keep repeating: the long-term trend may be different than linear. Exponential, sigmoidal, polynomial, you name it. The YEARLY trend is LINEAR, to a very good approximation, for 2022. For 2023, it obviously is difficult to tell with only 20% of the year in, but so far doesn't break from that trend: it just shows a steeper slope.

The basic stuff your unmovable faith in your preconceptions apparently prevents you from understanding is that the "short time period" in this trend can still be a year. "Short" means whatever your prior allows you to consider it to mean. To go to an absurd case, planetary cooling is well-approximated by an exponential over geologic eras, yet you wouldn't fit a thousand-year period to such a curve, because it would be indistinguishable from a very straight line, for the same reason the horizon looks flat but Earth is spherical. FFS why would you be so hellbent on proving something that is evident from a casual look at the data?

Can you explain what you mean by "the yearly trend is linear"?

Because whether it's exponential or linear within the year, I can't see how you're supporting the prediction that 100 launches/year is out.

In fact it'd be difficult to differentiate between the two cases within a single year, given the noisy discrete data like we have.

But it doesn't matter which it is, as long as it is not constant.

...unless you meant to say "constant" instead of "linear", and then maybe you have a point - if the data supports that.

[Redclaws]

I said this before, but I’ll say it again:

It’s nonsense to model something like the F9 launch rate with complex equations and expect it to have much predictive value.  It’s clearly controlled by many complex and interlocking factors of various kinds, many of which probably aren’t representable by any sort of smooth function at all.  Range shutoffs and capacity, payload availability, etc.  They’re going to be deeply discontinuous - straight lines, with straight line jumps or drops.  Mixed with some functions that are perhaps linear or exponential.  But what equation can model *in a usefully predictive way* the shift between Starlink generations and its effect on payload availability?  It’s certainly not going to be linear or exponential.  It’s going to be all herky-jerky, stops and starts, sudden changes, etc.

[alugobi]

Can you explain what you mean by "the yearly trend is linear"?

Because whether it's exponential or linear within the year, I can't see how you're supporting the prediction that 100 launches/year is out.

I don't recall that he, or anyone, is asserting that.  Some of us have expressed skepticism that they'll make 100, but I don't think anyone has flat out said "no way". 

[meekGee]

Can you explain what you mean by "the yearly trend is linear"?

Because whether it's exponential or linear within the year, I can't see how you're supporting the prediction that 100 launches/year is out.

I don't recall that he, or anyone, is asserting that.  Some of us have expressed skepticism that they'll make 100, but I don't think anyone has flat out said "no way".

Well the thread has meandered, but I believe it started with the analogy of a grade point average, and how if you keep getting 8/10, your odds of getting 9/10 at the end of the semester are greatly diminished.  Or some such.

[Robotbeat]

Just want to point out that an exponential function is actually the simplest way to model the thing where you improve slightly (but continuously) in capability over time. Piecewise linear requires a bunch more assumptions which are pretty arbitrary.

Exponential is actually extremely simple. Obviously things are going to be discontinuous and stochastic when we look at one launch to the next but the overall trend can be considered pretty simply.

There’s basically two ways people have been talking about the data in this thread:
1) Launch rate is a constant for each year. Last year was 61 launches per year, and so far this year it is just 80-85 launches per year. They will not likely make improvements to launch rate during the course of the year (those were all made around the new Year), so the overall projected launch total will therefore most likely be no different than the current trend of 80-85 launches per year.
2) The launch rate changes very gradually over time, not abruptly at the start of a new year. Simplest way to model this is just a gradual compounding improvement in launch rate. So last year they got 61 launches but the launch rate started lower than that and ended higher as they added capacity. The launch rate so far this year of 80-85 launches per year is roughly the same as it was near the end of 2022, maybe slightly improved, and they’ll make additional improvements gradually over the course of the year (so will end at a rate somewhat higher than 100 per year) and look to be on track to end up at around 100 launches this year with a similar rate of improvement.

Option #2 is the claim that the annualized launch rate improvement will be around 64%, enough to go from 61 launches last year to 100 launches this year and the current 80-85 launches per year annualized rate is right on track with that.

[xyv]

More activity than I could have hoped for in this thread.  Everybody has triple downed on why I am not making projections or curve fitting - noisy sparse data.  Don't get me wrong, I have curve fits in the same spread sheet (linear if you must ask  ) but the entire point of this was to have a place for every body to score keep at a glance if they are over or under the "line". 

And yes, a linear fit intersects at 85 right now and also shows noticeable seperation from the 2022 plot.

[Comga]

Sorry, it takes time LOL.

One week later, mighty complicated to take a pre-compiled Excel sheet and try to adjust to a clear exponential...

Well I deserve this criticism LOL.

An interesting differentiator would be to graph the "number of launches per 60 days"  sliding window style, over the last 3 years.

(60 days so we average out small logistics bumps)

Does it show the annual step-wise behavior eeergo is postulating?  Or is it continuously increasing?

Because if it doesn't, the logic of "for every 7-launch month they need a 9-launch month" is just part of the growth.  They'll launch 7/mo for 4 months, then 8/mo, then 9/mo, and this represents a 40% annual growth and that's not impossible.

I agree: Sigh 
This graph gets updated with each launch, but it seemed excessive to post it each time.
(That could be over 100 times this year! )
Attached is one from today, updated after the launch on February 9.
You wanted averages over 60 days.  This graph averages over 365.
Crandles57 averages over the last 12 launches.  This graph averages over the last 10.
YMMV 
The latter averaging is noisier than the longer time average.
Looking at that pace, it has been increasing erratically.
It went from 20 launches per year to 40 and back again in 2021
It reached 50 in early 2022 and around 70 in late 2022.
Recently it's been around 85, but it has been over 90.
It could drop below 82 if CRS-27 doesn't launch as scheduled, but could rise back over 90 if they get the next three launched as currently scheduled.
In other words, this is still noisy.
But it's piecewise constant AND kind of exponential.
edit: And many of those steps are remarkably close to the start of the years, meekgee, but that is neither here nor there.
To me, the bottom line is if SpaceX can get the next "bump up" in rate, with some inovation at which we can only guess.
If they do, they will reach the round number benchmark of 100 in a year, 1E2 in 1E0 Earth orbits.
If they dont, they won't. 

Edit: spelling

[alugobi]

There are 51 launches on the manifest list.  16 Have launched, total 67. 

If we stipulate that those 51 are solid, are there 33 more out there somewhere?

[DanClemmensen]

There are 51 launches on the manifest list.  16 Have launched, total 67. 

If we stipulate that those 51 are solid, are there 33 more out there somewhere?

Only four(?) of the 51 are Starlink. I thought the SpaceX strategy is to launch Starlinks when other customers do not need the launch slots, and it is not clear that those launches would show up on the manifest yet. So, just magically throw in 33 more Starlink launches. 

[joek]

Only four(?) of the 51 are Starlink. I thought the SpaceX strategy is to launch Starlinks when other customers do not need the launch slots, and it is not clear that those launches would show up on the manifest yet. So, just magically throw in 33 more Starlink launches. 

They did ~34 last year. Expect that will increase this year. So throwing in an additional ~33 sounds reasonable (more like minimal), even if they don't (yet) show up on the manifest.