Major_Tom Disproves NIST Claims in a Number of Key Areas

Savitzky-Golay is a one-pass algorithm that calculates both velocity and acceleration with a single polynomial curve fit at each point. That is not what femr2 did.

So, according to the 1964 paper you cited, there is no reason to stop at the second derivative. According to your reading of the paper we should also be able to extract the third, fourth and fifth derivative......from a single application?


According to your understanding, is there any reason the authors can't just keep going past the second derivative?
 
Of course taking the second derivative also forces you to use a poly order of at least 3, which may or may not be desirable.
Even without enough interest as for having dug into the S-G filter details, from my understanding of it from reading the threads, that claim has raised my alarms.

For that claim to be sustainable, you'd need to show that two successive applications of S-G aren't equivalent (or "are more desirable") to fitting with a higher degree poly. I don't think that holds true. I presume that two successive applications will have equivalent or worse quality than using a higher poly order.
 
So, according to the 1964 paper you cited, there is no reason to stop at the second derivative. According to your reading of the paper we should also be able to extract the third, fourth and fifth derivative......from a single application?
Implementations vary of course, but there should theoretically be no limit upon which derivative is returned, though practical limits seem the norm.

For example, the implementation within OriginPro is limited to derivative order 9, as the polynomial order is limited to 9. Other implementations, such as Fredrik Jonsson's Command Line SGFilter have practical limits as derivative order cannot exceed polynomial order, and polynomial/derivative coefficients must be available.
 
Last edited:
Even without enough interest as for having dug into the S-G filter details, from my understanding of it from reading the threads, that claim has raised my alarms.
First, a quick correction...

Of course taking the second derivative also forces you to use a poly order of at least 2.
(Though I'd suggest always using a polynomial order higher than th maximum derivative order you're after)

Derivative order cannot exceed polynomial order, for obvious reasons.

For that claim to be sustainable, you'd need to show that two successive applications of S-G aren't equivalent (or "are more desirable") to fitting with a higher degree poly.
That's not the point here. The point is... You can't generate third order derivative, from a second order polynomial.

To answer your question, the result of two successive applications of SG (that being apply SG to displacement data and generate 1st derivative, then apply SG to velocity data and generate 1st derivative) is not the same as applying SG once and generating the second derivative.

Additional smoothing is being performed in the former, with two different sets of polynomials being generated for two separate sets of data.

Whether that is desirable is subjective. Iirc single pass was far too noisy for my purposes, and I wanted to minimise window width.

I presume that two successive applications will have equivalent or worse quality than using a higher poly order.
The data will have been processed more, as effectively additional smoothing will have been applied. That will of course affect both signal and noise.
 
Last edited:
First, a quick correction...
Yeah, I noticed that but nitpicking at it didn't sound productive.


That's not the point here. The point is... You can't generate third order derivative, from a second order polynomial.
My point, however, was that performing 2nd order poly filtering twice might well be equivalent to (or worse than) than performing 4th order poly filtering once with the parameters adjusted (not with the same parameters as you seem to imply later).


Additional smoothing is being performed in the former, with two different sets of polynomials being generated for two separate sets of data.
That bare statement does not in any way demonstrate that you could not achieve that additional smoothing (or a better one from some theoretical point of view [1]) by applying one SG filter with different parameters, including a higher degree polynomial, to the same data. Which was my point.

Footnote [1]: I mean based on some objective theory, such as uniformity of the attenuation of the spectrum, etc., which is not to say that for certain applications the result could be better with two passes than with one pass, but you're far from proving it is.


Whether that is desirable is subjective. Iirc single pass was far too noisy for my purposes, and I wanted to minimise window width.
Correct me if I'm wrong, but by performing two passes, you're NOT minimizing window width: in the first pass, each data point will be influenced by points at the window width distance (or more properly, at (window width - 1)/2 distance, if I understand what the window width parameter means here exactly); in the second pass, each data point will be influenced by points that were already influenced by the window width, so the effective window width is doubled, though with different weights, and thus with the possibility of introducing artifacts.
 
My point, however, was that performing 2nd order poly filtering twice might well be equivalent to (or worse than) than performing 4th order poly filtering once with the parameters adjusted
No, the second application of the filter is working with an entiely new set of data, each point of which is influenced by data from its own window of the original data.

(not with the same parameters as you seem to imply later)
No, no limitation on using different parameters if you choose to.

That bare statement does not in any way demonstrate that you could not achieve that additional smoothing (or a better one from some theoretical point of view [1]) by applying one SG filter with different parameters, including a higher degree polynomial, to the same data. Which was my point.
You don't quite seem to understand. Increasing the polynomial order, with window size remaining constant, will tend to decrease the level of smoothing.

Footnote [1]: I mean based on some objective theory, such as uniformity of the attenuation of the spectrum, etc., which is not to say that for certain applications the result could be better with two passes than with one pass, but you're far from proving it is.
I'm not trying to prove it. You seem to be trying to do the opposite, whilst demonstrating you have some misunderstandings as highlighted above.

As I said, I chose to apply two single order derivative SG passes rather than 1 single order derivative SG pass and 1 second order derivative SG pass, as I wanted to minimise window width at all times, and second derivative SG passes were pretty darn noisy still.

By all means use a second order derivative SG pass on displacement data. If you use a fairly wide window and fairly low poly order, you'll end up with fairly similar trend results. There'll be differences in higher frequency fluctuations of course, but they are all pretty irrelevant when interpreting the graphs imo, as there is noise in the signal and I don't think specifying behaviour finer than trend is justifiable.

One way of gaining clear insight into any differences would be to test using Oystein's "god data", when he has time to generate it. Nudge :)

Correct me if I'm wrong, but by performing two passes, you're NOT minimizing window width: in the first pass, each data point will be influenced by points at the window width distance (or more properly, at (window width - 1)/2 distance, if I understand what the window width parameter means here exactly); in the second pass, each data point will be influenced by points that were already influenced by the window width, so the effective window width is doubled, though with different weights, and thus with the possibility of introducing artifacts.
You don't have to keep parameters the same between passes.

You're highlighting the misunderstanding you made above, when suggesting equivalency between 2nd order twice and 4th order once.

And, yes, each point in the second pass will have been influenced by all source points in its window. Similarly for data smoothed twice by any other method, such as running average with symmetric differencing.

I have performed numerous combinations btw, and with parameter changes as highlighted, the results are fairly similar. Similar trend. Meh.
 
Last edited:
No, the second application of the filter is working with an entiely new set of data, each point of which is influenced by data from its own window of the original data.
That doesn't disprove me. Just as an example: applying one IIR Gaussian filter to a signal twice with a certain radius (or different ones) is equivalent to applying one Gaussian filter with a bigger radius. The first pass generates an "entirely new" set of data, which is then applied a second pass, possibly with a different radius. Yet the resulting set of data is the same you'd obtain by applying the filter once with a certain bigger radius. (Note though that the "window" in this case is the whole data.)


You don't quite seem to understand. Increasing the polynomial order, with window size remaining constant, will tend to decrease the level of smoothing.
Of course, window size is one of the parameters I considered changing. I've outlined a reason why doubling it might get closer to the effect you obtain by applying the filter twice.


I'm not trying to prove it. You seem to be trying to do the opposite, whilst demonstrating you have some misunderstandings as highlighted above.
First, no, you're not trying to prove it, but you should if you want to address my objection.

Second, you have highlighted no misunderstandings. It seems the misunderstanding was on your side, though, as you implied a fixed window size while I didn't.


As I said, I chose to apply two single order derivative SG passes rather than 1 single order derivative SG pass and 1 second order derivative SG pass, as I wanted to minimise window width at all times, and second derivative SG passes were pretty darn noisy still.
You've lost me here. Unless you meant "rather that 1 single pass yielding both single order derivative and second order derivative".


By all means use a second order derivative SG pass on displacement data. If you use a fairly wide window and fairly low poly order, you'll end up with fairly similar trend results. There'll be differences in higher frequency fluctuations of course, but they are all pretty irrelevant when interpreting the graphs imo, as there is noise in the signal and I don't think specifying behaviour finer than trend is justifiable.
That sounds like an admission that my objection was sound. What this all led to is that you seem to be taking stabs in the dark without knowing what you're doing. That's what raised my alarms.
 
Last edited:
That doesn't disprove me.
You highlighted the "issue" in the final paragraph of your previous post.

Second pass is with a different dataset.

Also, it's your assertion. If you don't believe or trust what I'm telling you, I suggest you look into it yourself.

Of course, window size is one of the parameters I considered changing.
You said...
That bare statement does not in any way demonstrate that you could not achieve that additional smoothing (or a better one from some theoretical point of view [1]) by applying one SG filter with different parameters, including a higher degree polynomial, to the same data.
...yet now when told that using a higher degree polynomial will tend to decrease smoothing, you say "of course". Switcheroo ?! :)

I've outlined a reason why doubling it might get closer to the effect you obtain by applying the filter twice.
Try it and see. You've switched from "equivalent" to "closer to" by the way...

Second pass is with a different dataset.

First, no, you're not trying to prove it, but you should if you want to address my objection.
Why don't you ? Why should I spend time further addressing your unsubstantiated assertion, when I've already told you...

Second pass is with a different dataset.

Second, you have highlighted no misunderstandings.
I beg to differ.

It seems the misunderstanding was on your side, though, as you implied a fixed window size while I didn't.
Nonsense. I specifically stated you're free to use whetever parameters you like when you suggested I'd implied something I hadn't.

You've lost me here.
It would appear so, though you really shouldn't be lost. It's very simple.

Unless you meant "rather that 1 single pass yielding both single order derivative and second order derivative".
No, I meant what I said...

I chose to apply two single order derivative SG passes rather than 1 single order derivative SG pass and 1 second order derivative SG pass, as I wanted to minimise window width at all times, and second derivative SG passes were pretty darn noisy still.

My first pass resulted in velocity data from displacement data.

My second pass resulted in acceleration data from that velocity data,

Even if acceleration data is generated from displacement data with a second derivative SG pass, to produce velocity data and acceleration data still requires running the SG filter twice, albeit with both passes using the same source dataset. Once to generate the velocity data, and once to produce the acceleration data.

You only get one dataset output per pass, not two.

That sounds like an admission that my objection was sound.
To coin a phrase "you seem to be taking stabs in the dark without knowing what you're doing" :)

What this all led to is that you seem to be taking stabs in the dark without knowing what you're doing.
Nope. Choices made during the process of actually doing the task (what I did), rather than slightly wandering assertions based upon a seeming want to continue the trend of criticism whilst not doing the task (what you're doing).

As I said, by all means do it and see. I've already told you the outcome, but hey...

That's what raised my alarms.
What "raised your alarms" was...

"Of course taking the second derivative also forces you to use a poly order of at least 2." (where 2 was corrected from an erronious 3)

An obvious point I've had to spell out for you.
 
Last edited:
By all means use a second order derivative SG pass on displacement data. If you use a fairly wide window and fairly low poly order, you'll end up with fairly similar trend results. There'll be differences in higher frequency fluctuations of course, but they are all pretty irrelevant when interpreting the graphs imo, as there is noise in the signal and I don't think specifying behaviour finer than trend is justifiable.
Rather than waste more of my time engaging in fairly tedious ping-pong with opinion based assertions from others, and as I'm a thoroughly super chap, have a peek.

One way of gaining clear insight into any differences would be to test using Oystein's "god data", when he has time to generate it. Nudge :)
:eusa_drool:
 
Last edited:
You said...
That bare statement does not in any way demonstrate that you could not achieve that additional smoothing (or a better one from some theoretical point of view [1]) by applying one SG filter with different parameters, including a higher degree polynomial, to the same data.

...yet now when told that using a higher degree polynomial will tend to decrease smoothing, you say "of course". Switcheroo ?! :)
I've highlighted in yellow the part that proves you're strawmanning. A bigger window size could compensate for the unfiltered noise due to using a higher degree polynomial, with the net result of having an equivalent smoothing.


Try it and see. You've switched from "equivalent" to "closer to" by the way...
Since the beginning I used both possibilities simultaneously. At some point I may have used only one; if so, it was merely for the sake of brevity. My original statement was 'aren't equivalent (or "are more desirable")'. In my second message I used "might well be equivalent to (or worse than)". Etc.


Second pass is with a different dataset.
Yes, the derivative of the filtered version instead of the filtered version. Not a big difference for my objection.


It seems the misunderstanding was on your side, though, as you implied a fixed window size while I didn't.
Nonsense. I specifically stated you're free to use whetever parameters you like when you suggested I'd implied something I hadn't.
:eye-poppi
You don't quite seem to understand. Increasing the polynomial order, with window size remaining constant, will tend to decrease the level of smoothing.
Ahem...

I didn't suggest window size to remain constant. If you knew that and yet made that objection, then you were strawmanning again.


You only get one dataset output per pass, not two.
Ok, that's an implementation detail. The algorithm is the same even if run twice because of limitations in the software. By "pass" I understood producing a new dataset from the input dataset which would be fed as input to the next pass. But we're in agreement here. Your switching of terms is what confused me.


What "raised your alarms" was...

"Of course taking the second derivative also forces you to use a poly order of at least 2." (where 2 was corrected from an erronious 3)
Wrong. What raised my alarms was: "which may or may not be desirable", which was right after that sentence and you've omitted. You were implying that raising the degree may not be desirable, yet you apply the method twice which may well be equivalent to raising the degree, and thus you may not know what you're doing. I'm still unconvinced.


Which one is better: 1 pass or 2 pass? What degree and window size did you use for 1 pass? And for each pass in the 2 pass version?

ETA: And by the way, it would be interesting to see 1 pass SG overlaid with tfk's acceleration plot.
 
Last edited:
So, according to the 1964 paper you cited, there is no reason to stop at the second derivative.
Correct.

According to your reading of the paper we should also be able to extract the third, fourth and fifth derivative......from a single application?
Correct, but I may be misinterpreting what you mean by "a single application". I'm interpreting that to mean a single pass over the raw time series data. For more detail, including the definition of "a single pass" as that phrase is used by mathematicians and computer scientists, see my responses to femr2 below.

According to your understanding, is there any reason the authors can't just keep going past the second derivative?
Savitzky and Golay use third derivatives in several of their examples, provide tables for computing first through fifth derivatives, and derive the equations for arbitrarily high-order derivatives in their Appendix I.

Having asked me these questions, I assume you won't mind if, in the spirit of public service, I respond (succinctly) to some of the other posts in this thread.

LOL. Exactly how are you going to get a velocity plot and an acceleration plot by running the filter once (without recoding it so that multiple derivative orders are output) ? :rolleyes:
Let N be the degree of the best-fit polynomial, and let W be the number of points in each window.
Code:
for each window:
  for i = 1 to N:
    compute derivative i
That's one pass over the time series data, with N W-sized convolutions at each data point.

If both N and W are small, those N W-sized convolutions will probably take less time than reading a single data point.

If N is small but W is large, it might be faster to find the best-fit polynomial via standard minimization of the sum of the square errors, followed by evaluation (at the center of the window) of every non-trivial derivative of the best-fit polynomial.

That's what my implementation does. It delivers exactly the same results as standard Savitzky-Golay (up to roundoff error).

With modern computers, and time series as short as the ones we've been discussing in this subforum, it's beyond silly for femr2 to worry about the speed of the Savitzky-Golay algorithm.

You can run the filter on displacement data and get velocity data. You can run the filter on displacement data and get acceleration data. We want both.
Easily done. Run the Savitzky-Golay algorithm over the displacement data, generating both velocity and acceleration data in that one pass (as described above).

Implementations vary of course, but there should theoretically be no limit upon which derivative is returned,
Correct.

...derivative order cannot exceed polynomial order,
Incorrect.

Polynomial functions are C∞ (smooth, infinitely differentiable).

Of course taking the second derivative also forces you to use a poly order of at least 2.
Incorrect.

(The second derivatives of linear and constant-valued functions aren't very interesting, but they do exist.)

Derivative order cannot exceed polynomial order, for obvious reasons.
Incorrect.

That's not the point here. The point is... You can't generate third order derivative, from a second order polynomial.
Incorrect.

To answer your question, the result of two successive applications of SG (that being apply SG to displacement data and generate 1st derivative, then apply SG to velocity data and generate 1st derivative) is not the same as applying SG once and generating the second derivative.
Correct.

Whether that is desirable is subjective.
Correct.

Iirc single pass was far too noisy for my purposes,
We can't know whether that's correct until femr2 explains his purposes.

That bare statement does not in any way demonstrate that you could not achieve that additional smoothing (or a better one from some theoretical point of view [1]) by applying one SG filter with different parameters, including a higher degree polynomial, to the same data. Which was my point.

Footnote [1]: I mean based on some objective theory, such as uniformity of the attenuation of the spectrum, etc., which is not to say that for certain applications the result could be better with two passes than with one pass, but you're far from proving it is.
Correct.

The Savitzky-Golay algorithm is noted for its flat frequency response, which derives from its rectangular convoluting function. In pgimeno's next paragraph, quoted below, he sketches a proof that two iterations of the Savitzky-Golay algorithm can't be equivalent (on all inputs) to using a rectangular window of convolution. That means iterated SG doesn't have the same frequency response as the Savitzky-Golay algorithm. If there are objective reasons for wanting a flat response, as can happen, then iterated SG would be objectively worse than Savitzky-Golay differentiation.

Correct me if I'm wrong, but by performing two passes, you're NOT minimizing window width: in the first pass, each data point will be influenced by points at the window width distance (or more properly, at (window width - 1)/2 distance, if I understand what the window width parameter means here exactly); in the second pass, each data point will be influenced by points that were already influenced by the window width, so the effective window width is doubled, though with different weights, and thus with the possibility of introducing artifacts.
Correct.

(To simplify his explanation, pgimeno assumed the window widths were the same for both iterations, but his point would hold even without that assumption.)

No, the second application of the filter is working with an entiely new set of data, each point of which is influenced by data from its own window of the original data.
pgimeno's point appears to have sailed over femr2's head.

By all means use a second order derivative SG pass on displacement data. If you use a fairly wide window and fairly low poly order, you'll end up with fairly similar trend results.
Correct (for this particular time series).

And, yes, each point in the second pass will have been influenced by all source points in its window. Similarly for data smoothed twice by any other method, such as running average with symmetric differencing.
Again, femr2 seems to have missed the point.

Even if acceleration data is generated from displacement data with a second derivative SG pass, to produce velocity data and acceleration data still requires running the SG filter twice, albeit with both passes using the same source dataset. Once to generate the velocity data, and once to produce the acceleration data.

You only get one dataset output per pass, not two.
Incorrect.

As Wikipedia explains:
In computing, a one-pass algorithm is one which reads its input exactly once, in order, without unbounded buffering.
With that definition, which is the definition used by mathematicians and computer scientists, femr2's assertion is objectively incorrect.

Rather than waste more of my time engaging in fairly tedious ping-pong with opinion based assertions from others, and as I'm a thoroughly super chap, have a peek.
Once again, femr2 has failed to describe his parameters and methods in enough detail for others to replicate his results.
 
Last edited:
These pissing contests in this forum can go on for years. All one gets as a result is a bunch of urine.

Simple solution: Produce results and compare them.


W.D. Clinger, please produce a result and show it. Just one actual result will do.



This is how the SG filter has been applied to the extraction of acceleration trend from position data in physics related studies by other people:



Application in physics:


http://www.google.com/url?sa=t&rct=...=dSW1Di0VUK-m2EumZg-S8Q&bvm=bv.42080656,d.aWc


http://www.google.com/url?sa=t&rct=...=G_bciIC80W5VIHXoiS1Q_w&bvm=bv.42080656,d.aWc


http://www.google.com/url?sa=t&rct=...=v6b8RgOp2wF-gZW692HI7Q&bvm=bv.42080656,d.aWc


Savitzky–Golay and the Butterworth filters are often mentioned together.


Please demonstrate some actual results. Let us experiment and compare results. Much more interesting (and fun) than rooster posturing and accumulating urine.


Femr2, does W.D. Clinger have access to the raw position data? If so, let us see a demonstration with actual results.


>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>.

From the first link:

III.C. Computing Acceleration from Position via Numerical Differentiation
In order to determine acceleration from the videogrametrically-measured position data, we chose the approach of
differentiating the position data twice, using a simple numerical algorithm. While differentiation is an inherently
noisy operation, this approach has been used successfully in the past to compute velocities from experimentally-
determined position data.
For instance, Alexander and Colbourne [12] differentiated experimentally-determined
position data for several locations on a human limb segment, in order to obtain linear and angular velocities of the
limb. While no examples reporting the use of videogrammetric position data to obtain acceleration information
were found in the literature, there is no technical reason why this computation cannot be accomplished, as long
as appropriate numerical processing schemes are used to mitigate the noise amplified by the successive
differentiations.
.


V.A.ii. Determination of Acceleration via Numerical Differentiation of Smoothed Tracking Measurements
Prior work done by Kienle et al. [15] suggests that numerical filtering could be used to improve the results of
obtained when videogrammetry-based displacement data are twice differentiated to obtain accelerations.
Filtering prior (rather than subsequent) to differentiation was chosen because high-frequency noise components
caused by inherent measurement uncertainty tend to be significantly magnified by the differentiation process.

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

From the second link:

The first step of the marker displacement data analysis consists in the time histories filtering using a
Savitzky-Golay polynomial filter (Sgfilter). Successively, after double derivative to calculate the acceleration, a
band Pass filter is applied to the acceleration time history. The choice of the most appropriate filtering parameters
has been made comparing the results from different combinations of the parameters. E.G. the figure 3 shows the
effect of the SGfilter with different polynomial degree and same windowing width.
 
Last edited:
Yeah, these pissing contests and posturing are useless. They don't help MT achieve his goal of __________.
 
These pissing contests in this forum can go on for years. All one gets as a result is a bunch of urine.

Simple solution: Produce results and compare them.

What would be the criteria for comparison? Without a clear objective for the data one could compare by best use of color in a graph. Femr2 has already said his methods were used to "suit his purpose" although he has not actually defined what that is.
 
Last edited:
Yeah, these pissing contests and posturing are useless. They don't help MT achieve his goal of __________.
MT's current goal in this thread is to avoid supporting his claim that he has "disprove[d] NIST claims in a number of key areas".

So the pissing contests are not useless to M_T - they are helping him evade the challenge to support his claim. The pissing contests are not simply useless to the rest of us. They are a positive hindrance to progressing the topic of this thread.

M_T has steadfastly resisted requests to:
1) Identify two or more areas where he claims to have disproved NIST; OR
2) Publish his criteria by which he identifies "Key Areas".

All the off topic piss contests about "prove femr wrong version the umpteenth" are helping M_T to achieve his goal of not supporting his claim.

So the alliance of friends and foes joining forces to assist M_T to run away from supporting his claim is.....err.... "interesting"? :rolleyes:
 
Echoing ozeco's concern for the thread topic: It's been a while (about 10-12 days) since Major_Tom ignored these question:

The second point on the list, extracted from the 2002 transcripts:

2) The need for the NIST investigation to produce concrete, specific details of the collapses. The importance of avoiding vague, general conclusions. The importance of avoiding conclusions of a generalized nature. The need for a comprehensive, specific and complete understanding of the chain of events that led to collapse. The key, the fundamental purpose, lies in the details. The need for the investigation to center around producing detailed proposals for improvements in building and fire codes. It is essential to learn exactly what happened to these structures. The comprehensive nature of the study is essential. The most critical thing is the need to understand the collapses in much better detail.
...

:confused:

What exactly do you assert?

That NIST didn't produce concrete, specific details of the collapses?
That NIST came only to vague, general conclusions? The importance of That NIST came only to conclusions of a generalized nature?
That NIST didn't arrive at a comprehensive, specific and complete understanding of the chain of events that led to collapse?
That NIST didn't provide details?
That NIST's investigation didn't center around producing detailed proposals for improvements in building and fire codes?
That NIST didn't learn exactly what happened to these structures?
That the NIST study wasn't of a comprehensive nature?
That NIST didn't understand the collapses in much better detail?
 
These pissing contests in this forum can go on for years. All one gets as a result is a bunch of urine.

... From the first link:

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

From the second link:
Like your conclusion in your "book". A yellow paper of nonsense and BS not related to 911. You did not disprove any key area NIST stuff, except in your mind. Did I miss the list?


Posting papers which you have no idea why SG filter was used; cutting and pasting from the text. Which mean you can't characters the noise the filter is used to remove. What is the noise? Did you read the papers? lol, this is better than your "gravity collapse illusion".

engineer, engineer, engineer,

darn, you are still here.


You quote-mine the paper to get what you think is useful, and ignore the rest. I would listen to the mathematician, and stop trying to be an engineer/mathematician until you finish school.
 
Last edited:

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom