Weight loss: Width of the road? What about using moving average?

I’m new to beeminder, currently beeminding weight and fruits and veggies per day.

I just derailed on both for the first time! Extenuating circumstances of course (travel), but not THAT extenuating.

My question is about weight loss and peeking. I went from a blue dot. missed one weigh in, and then derailed the next day. At least I’m pretty sure I did. I think I deleted the offending datapoint to see if it undid my derailment, but then reentered it in the interest of weaselproofing and learning.

In retrospect, I definitely could have “peeked” at my derailing weight and not entered it and stayed on track, albeit skating the edge.

Really, though, I’m wondering about the width of the road for weight loss. I’m worried there’s a bigger range for my daily weigh ins than the yellow brick road can accomodate comfortably. And am I correct that the yellow brick road for weight loss no longer autowidens?

Especially if I’m off my routine (e.g. traveling), my morning weigh in can definitely fluctuate by more than 3-4 pounds. So for me to truly be “safe” day to day, I’m worried I need to be so far below the line that the safety buffer becomes misleading or meaningless.

I’m wondering why beeminder doesn’t just use the moving average line for weight loss to control for noise and narrow lanes to account for the lag? And has anyone set that up as a custom weight loss goal alternative?


The auto-widening guarantee only applies if you weighed in the day before. I think the key bit here is that you skipped a weigh in, which means that even if your last data point is in the green/blue, you can derail at your next weigh in.

1 Like

Oh, and


1 Like

So is there still a safe-day guarantee after the ‘death to auto-widening roads’ phase out talked about here?

I could potentially go from green to red in one day, and I currently interact with my goals using the android widget on my phone, which says I have 14 days until derailment (after a retro-ratchet), but that just doesn’t jive with my ‘hard cap’ number,

Certainly beeminding actions makes a lot of sense, but it seems like this whole platform grew out of weight tracking. And it’s super useful to track! I’ve personally lost reasonable amounts of weight just from the subtle behavior changes brought on by the awareness provided by daily weighing. Of course, one answer might be to beemind the action of weighing daily.

I’m starting to think I need a different accountability system for outcomes… maybe one that’s reward based for reaching the End Goal!

1 Like

Oh, huh I never saw that forums post (I think it is before my time). Hmmm now I am not sure, sorry! 0_o

Oh but I agree with you 100% that Beeminder is good for actions and bad for outcomes (unless you are very lenient with the goal settings and you also have an action-based goal for that outcome).

1 Like

Note that this is different from the auto-widening.

The width of the road lanes on a weight goal is determined by the variance of the data. If memory serves, each lane should accommodate about 90% of the day-to-day variation.

In my case, the lanes are about 0.9 kg wide, or 2 lbs in old money, which is a bit under 0.9% of my current mass . This calculated value should get more useful as you add more datapoints.


Also, weight being a lagging outcome indicator, don’t be too aggressive with your slope.

Prefer to beemind eating better (mostly plants — nice!), exercising more, getting enough sleep, and the contributing factors that might be more under your control.

Our beloved founders beemind their sugar intake, for instance. Other folks get right down into counting calories. Find a goal granularity that works for your particular psychology and helps combat your dietary weaknesses.

1 Like

Thanks for the reply! Some follow up questions and thoughts–
Is the yellow brick road actually defined as the 90% quantile of the rate-adjusted daily absolute variations in your data? The FAQ item refers to “90% variance in the goal stats” which I don’t see anywhere, and also says that this stat is “how wide the Yellow Brick Road should be so that if your true values were all right on the centerline, 90% of your actual values (with random fluctuations) would be on the road.”

I ask because in the forum thread announcing the death of auto-widening, @dreeves wrote “The can’t-lose-tomorrow guarantee will still obtain, in a sense, but only applies for flatlining [1]. The width of each lane of the
road is always equal to the current daily rate of the road”

As an aside, I’ll also point out that the roadwidth blog post does not reflect this change (whatever it is specifically), unlike many other beeminder posts and pages which direct the reader to updated info when stuff changes.

I’m also going to make one more plea for using the moving average as a beemindable metric for weight loss, even with the caveat that it’s probably more helpful to beemind actions than outcomes. Here’s my thinking.

Beeminder is an awesome platform with cool graphs and QS functionality, but it’s kinda (completely) broken for tracking weight loss, which is a bummer since it’s the chief use case for so many people, including myself. I want to use beeminder’s awesome features, like the lanes and the safety buffer! But with the noise in daily weight datapoints, these features are useless (and misleading!) because I might derail by surprise on any given day unless I’m so far in the green that I’m not paying attention to the road at all. And even if I become a premium member and make weight a free-bee, any “surprise” derailments will mess-up my nice pretty road graph.

Using the moving average would make beeminder more usable and helpful for weight loss because it would make use of beeminder’s current and excellent features.

1- It would make the lanes meaningful. If my moving average goes into the wrong lane, I know I’ve got to step up my “inputs” before I derail. But I also know I have a few days to do it. That’s what the lanes are for, right?

2- It would make the safety buffer meaningful. Same reason as above, but I want to emphasize it’s utility and current brokenness. The biggest single image/number on my graph is the color-coded countdown that says I have 13 days until I derail! That’s not at all true! I have a very good chance of derailing tomorrow!

3- It makes recommitting more sensible. When I derailed recently my graph re-railed at a weight that gave me a week off and 44 day buffer! That doesn’t make any sense given that the derailment was the result of a couple days of weirdness. With a moving average, both the one-week flat spot and the auto-recommitment based on the offending datapoint would actually make sense. And yeah, if I’m in the red when I weigh in the morning there’s probably nothing I can do about it, but that’s actually good because

4- Using the moving average would discourage crazy one-day weight loss gambits. We can all agree that a day of intense pooping and sweating is not the path to a healthy lifestyle, right? It’s waaayy better to derail and then get back on track over a week than to try to claw your way back onto the road over a few hours…especially since you’d pretty much have to stay dehydrated until you lost enough real weight to put you in a safe zone.

The argument against using the moving average outlined on the blog is basically this:
a) There’s always a “magic number” that you have to be under for any given day. and
b) Akratics who use beeminder will always skate towards the edge, and once they get there they’ll have an impossible magic number to meet. So
c) Conclusion: using the moving average for weight tracking won’t work.

But this argument has two problems:
1- It totally ignores how the moving average would jive so nicely with the existing beeminder features like lanes and safety buffers and auto-recommitment with a 7 day flat spot. Not using the moving average just feels like fighting the whole platform. And

2- While there’s always a “magic number” for any given day, beeminder defines the akratic horizon as one week (arbitrary, of course, but probably good enough). So, yeah, I might not be able to nudge my moving average much in one day, but I probably can change it meaningfully over the course of a week. And while I tend towards hyperbolic discounting, I’m still perfectly capable of prioritizing and planning at least a few days at a time if it’s something I care about (or there’s real money on the line). I’m akratic, not a fruit fly. And if I set my rate too steep, and it’s obvious upon derailment that my moving average was headed off track since day 1, recommitment is a perfect time to make an adjustment.

Ok. End brain dump. Now for a real question-- has anyone hacked a workaround with spreadsheets or IFTTT or something? And if so, can you share?


Agreed! And thanks!

From a “how do I actually do this” standpoint, those suggestions make a lot of sense. For reducing fatness, I’m currently trying to find the bare minimum, most frictionless things I can do indefinitely that will keep my trend moving in the right direction. I can count calories for a couple of months (and lose a bunch of weight doing it!). But I also know I really don’t want to do that for the next 50 years. It’s boring and time consuming and frustrating and imprecise.

But, I think I CAN weigh myself every day and keep track of my fruits and veggies for the foreseeable future, so we’re starting there. And fruits and veggies I’m using more as a proxy for ‘less junk food’ on the hypothesis that while my daily food intake isn’t constant, it’s at least finite. So more fruits and veggies will necessarily mean less “other”. Of course, I need to be wary of behavior that violates the spirit of the rule, like consuming 3-4 servings in smoothie or fresh juice form, leaving lots of room for junk food the rest of the day :slight_smile:

I need to add an exercise goal, but I haven’t figured it out yet. I wear a step counter, but I’m not sure how to structure the goal so that it’s forgiving enough (I won’t always have an hour to walk) without building up too much buffer…

I’m also considering "days with 10 or more pushups/situps/squats) at ~5 days a week. If I’m doing ten, I may as well do twenty. The challenge, as always, is getting started.

1 Like

Exceedingly Long Background Tale, with the TL;DR Bits Helpfully Bolded
I started tracking my weight using BM in November 2014. Even just tracking (as opposed to having a goal to move my weight in one direction or the other), I got bitten by the derailing for random fluctuation. So I decided on Strategy the First: put my “target” weight aka YBR up way, way high at a weight I’d never hit. (You can see Strategy the First in action between April and June 2015 on the graph.)

The problem with Strategy the First was that every time I looked at my graph, I felt SO SKINNY. Here was my actual weight, waaaaay down low on that graph, so far from that bad number up in yellow at the top, and there was allll this white space between my data point in the bottom third of the graph and the YBR. Yes, I’m just tracking, but I do want my Beeminder graph to subtly reinforce my desire to actually secretly lose another few kilos hey if it isn’t too much work. I don’t want it to subtly reinforce the sense that I’m way under my target and maybe even need to gain some weight before I expire from being too svelte.

(Let me be clear: this isn’t a Beeminder problem; it’s a mental thing in my head.)

So I turned to Strategy the Second: move the YBR down so that I am safely under it and in the green most of the time, but gaining any appreciable weight would put me on the YBR and out of the green and be motivating or at the very least a wake-up HELLOOOOO that I may be moving in directions unproductive. This also removed all that skinny-feeling plenty-of-room-for-ice-cream-and-booze white space between my weight and my “target” weight. As my weight dropped sustainably, I moved the YBR down to match. You can see Strategy the Second in action between July 2015 and June 2016.

Strategy the Second worked well until I went to the US for two weeks. (I’m from there, now living in Europe.) I came back 3 kilos heavier—which would have put me in the red and derailed me. I’d never imagined I might gain 3 kilos in those two weeks, but hey, I did.

Because my weight graph is purely for tracking and not a goal that I was slacking off on, I didn’t feel like derailing was “fair.” So I started logging the maximum safe weight every day, instead of my actual weight. The plan was that I’d drop the 3 kilos and be back into real-data territory pretty quickly. It was a little frustrating, though: this graph is my record of my weight, and here were bogus points on it, and when I looked back later it wouldn’t tell me how my weight had actually fluctuated. Even worse, I didn’t drop the 3 kilos as quickly as I’d hoped.

So I turned to Strategy the Third, which is a variant of Strategy the First that takes advantage of the mental cuing based on visual presentation: I put my “target” weight way up high again, but this time, I changed the max and min y values to display so that

(a) the “target” weight is off the top of the graph and thus invisible, so I don’t subconsciously see it as a value I’m doing much, much better than;

(b) the weight I consider ideal is vertically centered, so that my actual weights are in the top half of the graph, which exerts a subtle pressure to get below the center line aka get down to my ideal weight—and upward movement feels icky, rather than okay.

You can see Strategy the Third in action from June 2016 onward. Ideally, the YBR wouldn’t be visible on the graph, either, but the way I step-changed the target weight ended up doing that. Which is okay, I think; it makes a lot of weights blue instead of green, which can serve as additional motivation, now that the bogus “target” isn’t visible. (It also enabled me to retroactively correct my bogus weights to their actual values without triggering a derail…yay for me, but, uh, maybe not intended BM behavior?)

It isn’t how the weight graph is intended to be used, but for me, this more accurately accomplishes what the weight graph is intended to do, which is provide motivation to keep my weight down where I want it.


For weight-loss goals, yes, I believe so. The width of your weight road is based on the variability of your weigh-ins. In my case that’s just short of 1% of my mass.

Again, this has nothing to do with auto-widening per se. That’s the guarantee that if you’re in the blue with yesterday’s weigh-in you can’t derail on weight today. Also, @dreev’s post was a rant against auto-widening, not an announcement of actual change.

The moving average is a great indicator, but it’s a lousy beemindable metric, because of how the math works. It’s a very seductive idea, I think we’ve all been attracted by it, but the reality is not what you’d expect.

We’re in agreement that the current implementation of weight roads is not ideal.

I totally agree that the countdowns on weight graphs are ridiculous and unhelpful.

I did an experiment once of moving the road centreline to be on the moving average every time I derailed, and it was less helpful than I expected it to be.

(This is a recurring theme with any beeminding — the beautiful scheme we have in our heads often fails when it collides with idiosyncratic psychological realities, ceases to be effective after a period of time, doesn’t survive a holiday or busy period, or … )

We’re broadly in violent agreement, I think.

I applaud your starting with veggies and seeking out sustainable measures.

For keeping a sane safety buffer, I signed up to plan bee because of auto-ratchet. That lets me set an always-achievable slope without the risk of building up months of buffer. Sadly it does make for an uglier road, though after a year or so the yellow jags blend into a smooth-looking road. (And there’s also the premium perk of being able to edit your historical road to reflect the achieved slope.)

If you start counting individual exercises, check out @mary’s magic autoslope script, which pushes the pushups based on actual history.


Thanks for the reply Grayson! This sounds like about the best workaround with beeminding weight as-is…but it sure is a kludge! No offense to you, of course. If it works, it works!

Thanks for the clarifications, Philip! I also agree that auto-widening is probably bad, but variance in the width of the road is probably good. The problem remains that the range between daily variance in weight data-points is what makes this problem tricky-- With a regular, reasonable YBR you’re always at risk of a surprise derailment. But with auto-widening (or just too wide a road), you’re potentially setting yourself up for failure. Using the trend-line solves that problem, and I’m probably not going to be satisfied until I try using it for awhile… I’ll report back!

1 Like

Agreed—and I don’t think this approach would necessarily work for any other goal I have. Possibly this works because I have a lifetime inhabiting this body and watching it fluctuate with pregnancy, aging, etc. and my weight feels very tangible. I’m aware of it all the time (not in a bad way; just: wherever I go, my body is with me). And there’s a whole host of cultural/societal/whatever feels and thinks imprinted in my brain that make me care about keeping my weight “low enough.” All my other goals are more abstract, in that sense—no one but me cares whether I ever learn Mandarin, forex, and unless I move to China, failing to learn it won’t follow me around all day, every day, wherever I go.

It’s actually been pretty fascinating to notice the variance in my akraticism (akraticity? just “akrasia” seems to miss the “state of being” flavor I’m going for here) across my goals—I guess I don’t “need” Beeminder for weightminding. On the other end of the spectrum, I’ve just had to start a work goal to get through a project I cannot seem to make myself do (which has a long enough deadline that I will get myself backed into a very unpleasant corner if I don’t find a way to work on it a little bit every day). I think Beeminder is going to be what saves me, on that one.


The blog post against the moving average (http://blog.beeminder.com/movingav/) is in large based on the notion that you’re supposed to be on the edge every day. But I don’t want to live on the edge (been there, didn’t like it). My goal is to have a 7 days buffer on all goals. So beeminding the weight moving average (or any moving average) is perfectly “legitimate” in my view.


Same here, though I’ve had to train myself to do it. The color scheme is great for getting me into the green, but there’s no visual reward cue for getting that one-week buffer. Life has thrown enough curve balls at me in the past year, however, that I’m motivated to build up at least a week of green so I can safely adapt the YBR slope when a curve ball hits.

It’s also nice for non-curve-ball slacking—being able to take a day off now and then if something great comes up (say, a friend invites you to the beach for the day). That beemergency-day feeling of AAAHHHHH I’VE GOT TO DO SOMETHING is nerve-wracking as a strategy, for me. Using 7 or 10 days of buffer as the “edge” still gives me a value to stay above, but without the frayed nerves.

I set up most of my goals so that doing them every day will slowly build up buffer, maybe an extra day a week. That means that if I ever do need to skip a day, I can crawl back to safe land just by picking back up. The flip side of that is using retroratchet to keep the buffer from growing excessive.


This. I’ve only been beeminding for a month, but I want to build up a slow buffer that I need to retroratchet occasionally.


This topic just popped up in my Beeminder digest and I thought I’d weigh in with some ideas.

I’ve never used this, and it requires a premium plan, but I’m surprised nobody’s mentioned custom aggregation methods yet.

You could use truemean or median to basically be beeminding an average (albeit not a moving/biased one), even though you’re entering the actual instantaneous value each time (in other words, you just enter the weight you see on your scale, without having to calculate averages).

I saw in another post you setup a Rube Goldberg-style hack for this, and so long as that works, it’s arguably better, because it beeminds a moving average instead of an average, which may make more sense because your goal is to eventually be around X weight, not that your average weight since you started tracking is Y weight.

Maybe this usecase will inspire the addition of a xdaymovingaverage aggregation method where users can put say 10dmovingaverage and the result will be automatically converted for you. I think it’d be nice if this option was available (for weight loss goals only) to non-premium users, as a way for newbees to use Beeminder for weight loss.

P.S. You should try a do-more goal of drinking more water every day. Say one or two 16.9 fl oz bottles a day. I’ve found that often when I think I’m starving, I’m just dehydrated! You can also try to have a rule where you drink a glass of water before every meal. IIRC, the science behind that intervention is fairly compelling (vs. the science behind needing 64oz of water a day is simply wrong).

Some other potential rules

  1. Eat slower. In my case, I don’t have the time or patience for super-long meals, so what I do is eat a smaller portion, and then go do something else that is distracting, even though I feel like I could eat more. I work, read, walk, etc. Usually I find that after 20-30 minutes, I’m no longer hungry. Of course, if I’m still hungry, then I take that as a sign of genuine hunger and don’t feel bad about eating a little more.

  2. Eat more protein (protein itself, not “meat”). The science is pretty clear on this: increasing protein leads to decreased calorie intake, as well as (slightly) better body composition independent of calorie decreases. One way to do this is to use protein bars or shakes (yes, even though they’re processed foods). Another way is to eat lean meat (chicken, skipjack tuna, shrimp).

  3. Eat more veggies, as you’ve been doing, but with a particular emphasis on those that have a lot of volume. For example, 100 calories of broccoli is very filling for the calories, especially if you season it in a way that doesn’t use heavy cream-based sauces.