What are you afraid of?

• Or rejected for trying something different

Antidote:

• Talk with another few tech people/managers (Sankar, FI dude in AZ, Ellen?, Don Domes?). Confirm that it's seen as a strength and not as a weakness

Particularly with a number of employees, etc

Antidote:

• Deal with that when you get there! You haven't even taken the first step yet!
• Take time off / work on a different project!

Frequently I notice that I am wrong when talking with my collaborators at work or robotics. That I had one perspective going in, and came out usually agreeing with the other person's perspective and changed my own.

How right do I have to be in order to ship something?

How do I continue to have these people checking over my work without fully hiring them?

• Motivation is high for about 2-3 weeks, then on the first few missteps I feel unable to get anything done. Malaise.

What is the goal of these projects anyway? Profitability? Fun for me? Fun for others?

https://mbuffett.com/posts/maintaining-motivation/

• Even more flexible schedule
• More motivation, hopefully, they are my own projects. Don't have to convince myself to work on them. They fit in my head (for better or worse), and allow me to learn more varied stuff and more importantly, understand the tradeoffs to integrate them.
• When else are you going to be able to work on your own projects?

It's not about tricking potential sales leads, it's about transparently deciding if it's worth spending the next 5-10 years of your life working on one particular thing. oof

While the statistics provided by Statbotics are helpful, they are using only alliance-wide data and tend to be susceptible to bad match schedules and thus misrank teams. Hence many FRC teams also have scouts that are taking time-series based data for individual robots during a match on their phones with various touch inputs. Here is AEMBOT's, it's nice! https://github.com/AEMBOT/AEMScout. Another dashboard: https://www.chiefdelphi.com/t/curie-division-scouting-alliance/462271

For a variety of reasons teams don't tend to collaborate on this scouting data:

• Different teams have slightly different statistics they are looking for / have invested into their own app already and are used to it
• If a not great team gets a great match schedule and ends up a top ranked team going into selections, they tend to rely on ranking or statbotics data, which leaves diamonds in the rough for the taking and is a strategic advantage if you have the students / data. Sharing this data broadly nullifies that advantage.

But this brings bad externalities:

• At most competitions, the top (10?) teams have 6-12 scouters per team who are taking detailed notes during every match on every robot, rotating out if possible. This is a stressful task, and not very enjoyable.
• There is a lot of redundant data being taken down by those ~60 scouters at a small event, ~120 scouters for PNW champs.
• Because it's done manually, there are data issues. Leading to the strategy leads to not trust the data and instead use their own hunches.

What if we had shared, ideally automatically done time-series stamped scouting data?

• Most new-to-FRC students could be spending their time analyzing robot logs or spending more time in the pits exploring robot designs, or thinking about statistics and the weightings of an ideal robot for the elimination alliance! Or strategy.
• The strategic advantage of scouting data would go away. If you explicitly only want to do things that win a blue banner, that's a downside.
• Of all the things that FRC teams need to do in order to be competitive on the field, should fielding a scouting team of 6-12 people be one of them?

Answer questions like:

• When was Bear Metal on defense? And additionally, how effective were they?
• How reliable is the robot?
• Why did this upset happen? Did a team break, or maybe strategy wasn't great! (blue alliance stole 1323's fast passed notes faster than we could put them in) https://www.statbotics.io/match/2024new_qm64
• Is it more effective to have a robot play defense than to have it try to cycle? At what point does that tradeoff hold true?

Other requirements:

• How quickly is video data uploaded to youtube / put onto TBA?

• Note detection on PhotonVision uses a slightly customized YOLO v5 model. Should work well for tagging robots hopefully.
• Jeremy Howard / fast ai examples are using Resnet34? to distinguish between dog breeds. https://docs.fast.ai/quick_start.html#computer-vision-classification
  • His notes on limited sample training. https://www.fast.ai/posts/2016-11-17-not-all-the-same.html. Uses data augmentation (rotation, resize, etc). https://keras.io/guides/transfer_learning/
• Some of the issue with the video is that there is significant motion blur. Thankfully blind deconvolution is a thing (https://yuzhikov.com/articles/BlurredImagesRestoration1.htm), but it's not even blind thankfully! We know from what pixels changed the rough speed of the object, the "blur" amount then is just how long the shutter is open in the direction of travel. Rotation is a little tricky though, but that is what gradient descent is for.
  • Here's another paper: https://webee.technion.ac.il/people/anat.levin/papers/deconvLevinEtalCVPR09.pdf
• Can do corner tracking for motion, but also for mapping back to which robot it is in a cheap way. (if you have an unfolded sort of 3d image of each robot, priored based on what robots competed in that match)
• If doing difference detection, maybe want to do background subtraction? (for shadows). There are some tools for that: https://docs.opencv.org/3.4/d1/dc5/tutorial_background_subtraction.html

End goal is to show a scaled version of reality. Expected outcomes:

• While you can bet the farm on a few stocks (and need to in order to outperform), in aggregate you can't get magic outperformance of the index.
• Index funds take no effort and guarantee average, that's ok here!

• Insulated from the real stock market, yet inspired by it.
• Shares are traded among the class, prices are set based on these transactions.
• Dividends are paid periodically, making value investing a thing
• Throw some "cool" brands (and "bad" brands?) in there? (Instead of A,B,C,D) so that "castles in the sky" is also a thing, which is indistinguishable from actual value too. "whatever the market will bear"
• Give each student $25 in real money to invest? (paid out at the end)
• Add some "Life" elements, such as random expenses or bonuses that show up for each person?
• A die roll for each day determines small dividend movements up and down for that day. If, say, a middle-ish number is hit, you pull from the good and bad events pile. These can affect dividends right now, or dividends a few days in the future.
• How to implement an index fund? I think I can do it realistically. Have it sort of act like market maker. Pay a spread / premium based on what stocks you need / don't need. Will have tracking error!
• Kalzemeus has a great set of writeups on implementing this for his failed startup, Starfighter / Stockfighter.
  • https://www.kalzumeus.com/2015/08/20/designing-and-building-stockfighter-our-programming-game/ (halfway through)
  • https://www.kalzumeus.com/2015/10/30/developing-in-stockfighter-with-no-trading-experience/

Wonder what Khan Academy has put out on this.

Matt had a good idea too, compress a real time period of stock market from 1 month down to 1 day. So long time period effects.

• Can maybe host the database with github pages. Then have one central admin account and server that handles transactions and pushes the update periodically to Github pages, but still allows clients to run / analyze while "offline".
• Thwart hacking by running all transactions in HTTPS and assign a unique hash cookie from each person's phone (can't guess based on someone's name)

Change my mind!

• I'm fairly emotionless. Ok with seeing assets drop 30%. I sold my bonds and bought more stocks in 2020's downturn.
• I'm young. It's still funny money to some extent instead of life savings. I can see it being a different story for someone older.

• Interest is tax deductible
• Only acquire stable assets when you need it. Overnight interest rates will be low at that point too.

I-bonds. They rate reset, correct? High value now, 7%.

ERN article on asset allocation.

Downturn of a year or two, then recovery

Prolonged downturn (decade?)

Here's a thought experiment for you!

How would you listen to an audio signal if you could only look at it at a maximum of 100 times a second with only a yes/no output (best-case scenario of every neuron in the body)? Look at it...differently :)

Great animation: https://www.youtube.com/watch?v=0jyxhozq89g, but more specifically, imagine a tube/wire resonating, similar to a spectrogram: https://www.youtube.com/watch?v=EGeTawy4pDo

I can distinguish distinct impulses up until about 40 Hz, and then I just "hear" pitches. Crazy, huh?!

There's a similar deal Given that the maximum firing rate of a neuron is 200 times a second, how do we humans hear frequencies above that?

Another puzzler for you, why is it that insects that fly are so fast? Are they smarter than people, or is there another reason?

Slow mo guys damselfly eating smaller fly really fast: https://www.youtube.com/watch?v=dOyc98tV5kA

Given that the propagation speed of an action potential in a neuron in the body is 20-80 m/s and it takes 1msec for a signal to cross the gap between two neurons (a synapse), and the maximum firing rate of...

How many fans would you need?

• From trying some Lasko 20" from home depot, a lot of them.

Fan	CFM	Watts	$
Lasko 20 in white	1800	83	$18 https://www.homedepot.com/p/Lasko-Save-Smart-Energy-Efficient-20-in-3-Speed-White-Box-Fan-with-Built-In-Carry-Handle-3733/100405665#product-overview
Lasko Power Plus 20 in black (better) Look into adding ring on cheap one? Says improves velocity 20%	2000	112	$30 https://www.homedepot.com/p/Lasko-20-in-Power-Plus-Box-Fan-B20540/300202629#product-overview
Mainstays walmart box fan 20 inch Stackable more easily too! (nothing on top)	2400	??	$20 https://www.walmart.com/ip/Mainstays-20-3-Speed-Box-Fan-FB50-16HB-Black/136335464
Mainstays walmart high velocity fan 20 inch	5800	??	$40 https://www.walmart.com/ip/Mainstays-20-inch-360-Degree-Pivot-High-Velocity-Steel-Floor-Fan-Black/846634950
Harbor Freight 30 in shop fan	9000?	216	$100 on sale, $150 not on sale https://www.harborfreight.com/30-inch-pedestal-shop-fan-47755.html

And if you want to crank the asset location discussion up to 11, the real answer is it depends! Spreadsheet provided though! https://earlyretirementnow.com/2020/02/05/asset-location-do-...

The author also is talking about taxable brokerage accounts too, for a total of three different tax styles (after-tax "taxable" brokerage, pre-tax "Traditional" retirement account, after-tax-with-no-future-taxes "Roth" retirement account), but given enough reads I think you can figure out what you need to.

It's also good to know the magnitude of the difference in final portfolio value on the different strategies. If it's <1% I personally don't worry about it, especially if it takes more work to implement.

So who’s right? Conventional wisdom or the skeptics? Long story short: they’re both wrong! You can easily construct examples where either conventional wisdom or the skeptics prevail. So neither side should claim that their recommendation is universally applicable. The asset location decision depends on… Your expected rates of return, Your expected tax rates, Your investment horizon. Yup, you heard that right, it’s possible that you want to go either one way or the other depending on the horizon. Though, this is not really a separate case but really only a result of asset allocation drift. Accounting for that, we’re back to the two cases, but more on that later!

HI Nolan, thanks for sharing this article, though it's a LONG and confusion one. I am sure you read through and have some thoughts or take-away from it. To me, without doing all the calculations, if your overall equity/bond allocation is 70/30, then having >70% equity in RIPC and >30% bond in 401k would probably make up a good ball park of asset allocation(70/30) for risk tolerance and for tax-efficiency purpose, regardless of the three factors you listed above. Do you agree? In the article, a taxable account is irrelevant to our discussions here and makes the calculations so much complicated... Thanks again!

Good point Lillian!

You have a multi-layered question, and there's a few elements:

• Asset allocation (70/30 like you're saying). Risk vs. reward.
• Retirement savings vehicles (traditional vs. roth vs. taxable). Primarily based on your tax brackets during working and retirement and expected returns and taxes from those investments.
• The combination of the two (what assets go into what vehicles)

As I read through it again I agree it is long and confusing. But I would argue it's mostly because the underlying topic is actually long and fairly nuanced, as the author hints at. I would doubt anyone that says it is simple![1] Unfortunately for some situations (same tax brackets going in and coming out = high earners and high retirement spenders) there is no difference, but for others (high earners looking to retire early with minimal expenditures) the difference in outcomes is large, measurable in years of your life spent working! (https://www.madfientist.com/traditional-ira-vs-roth-ira/)

I wish I had a simulator to point to [2], but at this point you either have to do the spreadsheet/code yourself (kinda fun once you ramp up on it) or hire a retirement planning professional who knows the ins and outs of the topic and has good software to back them up. I don't have the answers for everyone, just a lot of anecdotes and not a lot of motivation to be an expert financial planner :)

To your taxable point, it is actually better in some situations to contribute to taxable over a Roth since you can do tax-loss harvesting and can withdraw earnings at any time. https://www.gocurrycracker.com/roth-sucks/

[1] I couldn't believe all of the implicit assumptions that were made in this Dave Ramsey article: https://www.daveramsey.com/blog/traditional-401k-vs-roth-401...

• Excess money from savings over the 401k limit would just get spent and not saved.
• Paying taxes when withdrawing money feels/is different than taxes on contributions?! (they are the same thing!)
• All retirees will be in the 22% tax bracket on retiring (>$80K of income for married couples for the marginal bracket, but for the overall tax rate this is actually much higher!)

[2] An excellent NY Times simulator on whether to rent vs. buy: https://www.nytimes.com/interactive/2014/upshot/buy-rent-cal... Something that's a good start/concept for asset location tradeoffs: https://observablehq.com/@russelldmatt/ira-tax-stuff

https://www.smbc-comics.com/comic/lions

• I like the infinity scroll mouse, right? Lost remote control for it. It's heavier too.
• But now my brain sees that lighter is better for gaming. Aaand all of a sudden it doesn't want a heavier mouse. Hahahaha.

• Just lost an eBay bid at the last second for an old-style MacBook Pro whose keyboard I kind of like. For whatever reason, I don't really feel like buying it anymore, despite feeling intense "hole burning in pocket" feelings before. Maybe later?
• I bought it a year or two later on Craigslist, the feeling came back. Aaand it was cool, but not cool enough to build a custom PCB around it. Since the laptop itself was so old, I couldn't easily install working remote control software that would work with work VPN, etc. So frustrating. Killed 10 hours.

• Why couldn't I get enough of barbecue chips two months ago but now can't really stand them?! Weird. It'll probably come back soon.
  • <6 months later> Nope!
  • <Dec 2021> Nope!