Billy Mitchell's Direct Feed Recordings are MAME Generated

[VON]

Can we use the footage from the IVGHOF announcement to demonstrate his Junior play was also MAME-generated?

[bh]

After seeing Wes Copeland's argument against Mitchell's run, I decided to figure out just how unusual it is.

I started by transcribing all of the blue barrel and flame smashes from barrel stages on Mitchell's 1,050,000 game. Here's my spreadsheet with timestamps. I hope this is accurate. The video has serious quality problems and much of the time I was stuck inferring scores from digit rollover in the blurry total score. I welcome any efforts to replicate, extend or amend this transcript.

What I found is that the score distribution in the 1.05m run is significantly shifted toward the high range. There are fourteen 300 pt blue barrel smashes and thirty-four 800 pt smashes. On average his blue barrel smashes were worth 570 points and his flame smashes 477 points. Between these two types of smashes he scored 78,500 points over his runs. The odds of this happening fairly are about 1 in 360 (99.73th percentile), or about once per 450 hours of near perfect play.

I suspect a transcription of variable scoring events in rivet and conveyor stages will push the odds into absurd heights.

Thanks goes to @othercriteria for double checking my math, and Wes Copeland and the folks on #mame for patiently answering my questions.

[xelnia]

After seeing Wes Copeland's argument against Mitchell's run, I decided to figure out just how unusual it is.

I started by transcribing all of the blue barrel and flame smashes from barrel stages on Mitchell's 1,050,000 game. Here's my spreadsheet with timestamps. I hope this is accurate. The video has serious quality problems and much of the time I was stuck inferring scores from digit rollover in the blurry total score. I welcome any efforts to replicate, extend or amend this transcript.

What I found is that the score distribution in the 1.05m run is significantly shifted toward the high range. There are fourteen 300 pt blue barrel smashes and thirty-four 800 pt smashes. On average his blue barrel smashes were worth 570 points and his flame smashes 477 points. Between these two types of smashes he scored 78,500 points over his runs. The odds of this happening fairly are about 1 in 360 (99.73th percentile), or about once per 450 hours of near perfect play.

I suspect a transcription of variable scoring events in rivet and conveyor stages will push the odds into absurd heights.

Thanks goes to @othercriteria for double checking my math, and Wes Copeland and the folks on #mame for patiently answering my questions.

I'll definitely dive into this data soon, but it's worth reading this thread on CAGDC about the bugs involved in expected smash values for variable scoring items (blue barrels, fireballs, pies). In short, the EV of a fireball is 450 points and the EV of a blue barrel or pie is 525 points (except for one pie that will be a forced 300).

[Mitch Mitchell]

Incredible catch finding the loading differences, thanks for the hard work and concise post.

[Arbee]

We're not talking about game state at all, simply the manner in which MAME composes the contents of the frame it displays vs original hardware.

DK hardware's video generator reads the contents of VRAM as it composes the signal going to the monitor.  That means the scan lines at the start of the scan display the state of VRAM from an earlier point in time than the scan lines at the end of the scan.   The Z80 keeps drawing stuff into VRAM while the video raster is going.

Old MAME does not emulate this hardware at all, it simply composes the screen in it's entirety at one time-point in the vblank.   This means the image does not contain any changes the Z80 made to the VRAM during that 1/60th of a second spread of time, only a 'freeze frame' moment in time.

This isn't a lack of emulation in old MAME, it's just that on older PCs (and current Raspberry Pis) sampling the generated bitmap more than once per frame was extremely slow, because it involved flushing the PC's caches of everything they knew about the Z80 state each sample.  Nowadays we can get away with that.

Unfortunately the left-to-right vs right-to-left thing is weird because we're just showing what the Z80 is drawing, so it makes me think we're gonna need a logic analyzer on a PCB to understand what's really happening.

[Sock Master]

This isn't a lack of emulation in old MAME, it's just that on older PCs (and current Raspberry Pis) sampling the generated bitmap more than once per frame was extremely slow, because it involved flushing the PC's caches of everything they knew about the Z80 state each sample.  Nowadays we can get away with that.

Unfortunately the left-to-right vs right-to-left thing is weird because we're just showing what the Z80 is drawing, so it makes me think we're gonna need a logic analyzer on a PCB to understand what's really happening.

Old MAME ran the Z80 for 50000 cycles (or however many Z80 cycles it executed in 1/60th of a second), then composed a full, complete frame to display, then ran the Z80 for another 50000, then composed the next frame, then another 50000 cycles, next frame, etc.  Simple as that.

Newer MAME interleaves the Z80 and composition of the screen at much finer intervals so there is real-time interaction between the Z80's actions and what appears in the frame.   MAME composes the screen in the opposite direction from actual DK hardware.   Reverse the order and it'll more closely match actual DK hardware.

[Scoundrl]

Does it still draw opposite the arcade hardware no matter the screen orientation 0/180 or 90/270?

Is it possible this cab was using a none sanyo monitor and the inverter they were using (or the inverter they would have used to get the direct feed) is the cause of this?

[Sock Master]

Does it still draw opposite the arcade hardware no matter the screen orientation 0/180 or 90/270?

Is it possible this cab was using a none sanyo monitor and the inverter they were using (or the inverter they would have used to get the direct feed) is the cause of this?

MAME renders the frames the same way no matter which orientation is selected.  At this time, no version of MAME regardless of settings, rotation or monitor will render frames the way actual DK hardware does.   Now that they are aware of the issue, I'm pretty sure MAME dev will fix it in future releases of MAME.

The color inverter does not affect rendering order, just inverses the colors.   DK hardware will still render like DK hardware even if using a normal monitor and a color inverter.

[ChrisP]

Since I weighed in on this thread, I just wanna say one thing, for the record.

In light of some reading and thinking and talking that I've been doing over these past few days, much of which is all new (at least to me), and in light of how the situation continues to develop, I'm just gonna say this: if I were to write those posts now, they would not be nearly so charitable towards Billy. And I'm coming to regret having expressed myself with the "it all worked out for the best" softballs.

I don't remember the order of the "five stages of grief" but I'm pretty sure I'm now at anger. 

[TheSunshineFund]

So who was the first to surmise that Billy was using MAME?  I know there was mention of a Hank.ppt at one point a long while back, was that part of his deck?  Just interested for historical purposes of all this impressive work.

[LMDAVE]

I hear ya, Chris. It's hard not to feel a little anger or confusion on this one. Confusion to me being, what more evidence is needed? Too many things need to be explained from Billy's side.

I'm at the stage of "why?". I hate to do any type of speculation, but if I had to try to put my own guess on why these tapes were made, this would be my guess. I think maybe the first tape was made from a standpoint of "if it could be done", could a tape be made with MAME to show a million point game. And maybe it was done and just filed away, never intending to be used (maybe joked about it to others that "I have this huge score on tape"). Then the emergence of Steve Wiebe comes along, and obvious he wasn't going away, so the decision to reveal that tape came out. The Fun Spot delivery. "This is a score they've been waiting to see for a long time and I held it for far too long" -Billy to Doris

That particular tape was 40-50K over Steve's score. Maybe then after the submission passed the eyes of the referees and everyone else, that became clear that this can be done again. Each subsequent tape after that was the famous 1,100 over the WR then dump the game.

That sort of taunting was accepted because it was just Billy being Billy, but to find out afterwards that those scores are fraudulent (allegedly), that's the part that hard to get over for those of us that have many years invested in this particular game.

[Xermon54]

I agree with Dave. At first, his MAME submission were probably done for entertainment purpose, just for messing around, or to see how much points could be done, etc. But then, once he saw that his score got accepted, he saw an opened door for continuing his cheater, up to a point of no return.

Also, if Twin Galaxies wouldn't have accepted those direct feed score in the first place, Billy wouldn't had been encouraged to continue his cheating. It's like an olympic athlete taking steroids but never get tested positive in competition, he'll obviously keep on cheating.

[DonkeyKongGenius]

I actually find myself indifferent towards this matter. It isn't that I don't care, in that I would like to know if someone cheated or not, but I guess I have no real emotional connection to Billy so whether he or anyone else has evidence presented in a case against them, well, I just say... O.K. thanks for the information.

[TheSunshineFund]

I have no real emotional connection to Billy

Dat sig doh.

I think it's pretty difficult not to have one for a great majority of us.  I would have never rediscovered classic games if I didn't see KoK and realize that people still played these games, etc, I would have never went up to Funspot if I didn't see that movie, met a lot of new great friends.  It's definitely a mix of  ,  and 

[YesAffinity]

Guys,

I would like to submit some additional evidence.  As the conversations continue, things just aren't adding up for me, from the technical perspective.  I'm not able to reconcile either scenario, given what is currently known - neither that the performances WERE absolutely mame, nor that they WERE absolutely true arcade hardware.

I think first and foremost, it is of extreme importance, because so much emphasis is being placed on the visual evidence of the gameplay, that the likely environment of capturing of the footage be replicated as best as possible.

Considering technology that was likely used to capture the video, without going into the hardware that was actually producing the gameplay, it appears that:
-The 1.047M video is a camera pointed at a screen
-The 1.05M video is a vcr recording of a direct feed

If there is actual knowledge contradicting these assumptions, and of how the footage was recorded, in part or in whole, please provide that information.

In both assumed scenarios, you would effectively have a 60 hz screen being captured by a 29.97fps recording device.  Even if the assumptions aren't exactly right, this is still the most likely scenario for the time period.  60hz screen being recorded with a 29.97fps recording device.

So, I think it's important we replicate that environment for analysis.  Comparing a 60fps capture of a 60hz signal is not apples to apples - twice as many frames being captured in this scenario.

Below is a video that as closely replicates the original recording scenario as I am able to without going to the thrift store for a VCR and some crappy old VHS movie to record over.  My cab is generating a 60hz signal, my video processor is at 480p/60, my capture card is at 29.97fps, OBS is at 29.97fps.  I also have my webcam at 29.97, capturing the screen (60hz).

https://www.twitch.tv/videos/226969998
I will let anyone who wants, to draw their own conclusions.  In some instances, like on L=1 rivets, there is no "lead-in" drawing of the screen.  One frame is the stacked monkeys, the next screen is the entire rivet level populated on the screen.

I will happily provide the non-streamed recording file to anyone that wants it for further analysis and comparison.

Now this alone still doesn't answer all questions, but I think it objectively aids the argument that the recordings were not provably MAME beyond the shadow of a doubt.  And I emphasize OBJECTIVELY.  I simply want answers, like I think everyone here, and I believe is the intent of the original analysis/OP.

There is also potentially some gray area being created in the visual representations from the differences between the following scenarios.  They could be meaningless, but they could also be affecting the visual results.  Based on my assumptions above of the original recording environment:
-1.047M tape - analog screen, analog recording (external to the screen)
-1.05M tape - analog generation, analog recording
Compared against:
-My direct feed - analog generation at gameboard, digital capture
-Jeremy's camera phone - analog screen, digital video recording

There are still things that don't jibe with me from a technical aspect of both sides of the conversation:
1) As implied but not outright stated in my earlier post: a MAME setup of the nature required to achieve the alleged result would be a technological challenge even now, but more so at the time.  If it were any game other than a Nintendo game with inverted video, it would be much more easily achieved.  Because of the inverted video, it's an added layer of complexity.  For at least one of the tapes (live performance in front of others), we would be talking about a modified cabinet with the ability to playback an .inp file in a format that jibes with a monitor that requires inverted video, and to send video to another monitor/device for recording that may or may not require inverted video.
2) If it was truly arcade hardware, was the tape footage rotated?  For the 1.047M footage, again, this appears to be a camcorder or some external recording device taking video of a screen.  That screen could easily have been physically rotated and oriented for the camcorder/etc. recording.  The apparent VCR footage, however, isn't so easily explained away, unless it appears in its native non-rotated orientation on the actual tape, or it was post-processed.

All this is NOT to say that the original analysis is wrong.  I do think, however, that the means of comparison that are ultimately leading to the statement "it was MAME!" are not painting a picture that is cut and dry.  At the same time, I'm only looking at the few samples from the original footage that are provided here.  If EVERY new level has been looked at in both pieces of footage, and it can be said without a shadow of a doubt that there is absolutely no partially-present "lead out" screens from stacked monkeys and no partially-present lead in screens of the start of the level, effectively there is absolutely no "sliding door" effect in the footage, well that is significantly more conclusive.  And maybe this statement was implied in the OP, if Jeremy or SM can make that statement based on what they've reviewed, again, that is significantly more conclusive.

I know the original tapes are supposed to be in transit to authoritative bodies for further review, and hope my suggestions help in some way to further the conversation and provide clearer answers for some of us that may now be looking at the situation through a wider lens.