I just finished re-reading the whole thread and I've got one more scenario to try:
SB wakes up and feels that it is 50% likely that heads was flipped. Then she is told that what day it is and asked again. What impact does hearing this information have on her credence?
(I don't know whether it makes it easier or harder, but you can assume that she knows beforehand that she will be told what day of the week it is after she is first asked for her credence).
Here you have a halfer SB, so hearing it is Tuesday kills the heads possibility and credence, so hearing it is Monday doesn't change or add anything to her, her credence is still 1/2 for heads.
It is more interesting if she is a thirder, so prior to being told the day she thinks it is 1/3, then on hearing it is Monday, she updates her credence on conditional probabilities:
She goes from prior probabilities P(HM)=1/3,P(TM)=1/3,P(TT)=1/3 to
P(H∣M)=P(HM)/(P(TM)+P(HM))=1/3/(1/3+1/3)= 1/2.
So your variation is interesting because it reveals that hearing “Monday” is not neutral evidence unless you already chose your prior over centered awakenings.
The thirder says:
before hearing the day: three centered possibilities, equally weighted,
after hearing Monday: eliminate Tuesday,
leaving Heads-Monday and Tails-Monday equally likely.
The halfer says:
before hearing the day: the coin is still 1/2
hearing Monday adds no bias toward Heads,
hearing Tuesday kills Heads.
That explanation sounds consistent, but still seems odd to me. Let me poke at it from one more direction.
Let's wake sleeping beauty up and ask her what credence she gives to the idea that it is Monday.
I'm guessing from what I saw above, that halfers will say "3/4" but anything other than "1" will do.
We then ask "If I told you it was Monday, what would be your credence that it is heads?" and "If I told you it was Tuesday, what would be your credence that it is heads?"
We can then derive a value that should match her credence that it is heads:
my credence/posterior is 1/2; my wager posture is 1/3. we revisit the assumptions under which the two numbers are typically believed to coincide and find they don't hold for SB.
I'd love to hear a statement of those assumptions. I strongly suspect that such conditions exist, but I'd like to disagree that they apply in this particular case, but can't reasonably do so because I don't know what the assumptions are.
I'd also like to revisit the situation that I mentioned earlier (on 2020-08-17).
The procedure there was that she is woken up on both days, but is only given a piece of paper to record her credence if is Monday or if tails was flipped.
What should she write down as her credence?
It feels to me as though halfers should agree that it is 1/3. (Because the information as to whether she was given a sheet of paper is available each time she wakes up.)
The question I would like to as is:
How does the presence or absence of the event of "SB waking up and going to sleep without any memories being formed" affect her perception of what happens during the time that event isn't happening?
I'm not saying that such an event can't affect her perception, but my imagination is not coming up with anything. (A possible starting spot. When she wakes up, she is in a state of uncertainty as to whether she will be given a sheet of paper, which then resolves. But how does that matter? And why would it matter if being told that it is Monday doesn't matter?)
here's a variant of the problem: consider 1000 mail boxes, the sender flips a coin, if it lands on heads, he sends a letter to the first box, if it lands tails, he sends a mail to all of the 999 other boxes. you happen to be the owner of one of the boxes and receive a letter, what is the probability that the sender's coin landed heads?
kocbftn2, here are my thoughts:
I own one of the 1000 boxes, but I don't know which one (or how the position numbers were assigned); so with only this limited info I believe (or guess-timate?) there is a 0.1% chance I have the first box and a 99.9% I have a box that is not the first. I know the odds of a fair coin landing H or T in a fair toss to be done in the future is 50% H and 50% tails by definition... And I received a letter, so either the coin landed H and I have box #1 or the coin landed T and I have a box that is not #1 (a box numbered 2 or higher, up to 1000)... And... I feel like somebody looking for Schrödinger's cat because I can't decide if the fact that I got a letter (which I would get in any case, if the coin landed H or T) gives me any information about how the coin landed... And I'm wondering how other people would feel about your question and another similar question with one (or 2) changes, what if the question is changed to state "if coin toss lands T then send NO letters (& if H send letter to box #1)" and what are the odds if you received a letter? What are the odds if you didn't receive a letter?
the fact that you received a letter doesn't give you any new information about which box you own. the only new information it gives you is that a coin was flipped, which means that according to the information you gain from having been sent a letter, the probability of a heads and tails outcome is equal. without receiving a letter, knowing that you own one of the mailboxes gives you 99.9 percent probability of owning one of the 999 other boxes as opposed to the first box, but it does not give you any information about the outcome of a coin toss that should occur. so neither owning one of the mailboxes nor receiving a letter actually allows you to update the probability of a coin toss outcome, which by default is 50 percent
*I meant: "(which I Could get in any case, if the coin landed H or T)" - meaning: Either 1) the coin landed on H and I have box #1 and I got a letter, OR 2) the coin landed on T and I have one of the nine-hundred-and-ninety-nine (999) boxes that got letters.
...
Although {in one way (but not in other ways of thinking about it)} it seems that since I got a letter it is 999 times more likely that my box is NOT box#1 and thus 999x more likely the coin landed T - mostly I think/feel that is a type of statistical illusion - like something Kurt Gödel would create (like the visual illusions Escher created - the staircase loop that goes up (&/or down) forever.
Neato! Thanks for sharing kocbftn2, Carroll ★ at 2020-08-17, William Fraser, and everyone else!
In other words: I agree with kocbftn2's above reply to me (sent 5 hours ago, which I didn't see until I clarified and expanded my reply to them, directly above/ a minute ago).
I think this one is more straightforward than the original SB problem (because "given that SB is awake" is hard to pin down formally and nothing here is ambiguous).
Define the following random variables:
Heads: heads was flipped.
Box1: I own box i.
BoxK: I own a box that is not 1.
Letter: I received a letter.
The assertion that I want to make is:
P(Heads given Letter) = 0.001
Do we agree that the definition of "P(A given B)" is "P(A and B) / P(B)"?
I think there is no disagreement with the following:
P(Heads) = 0.5
P(Box1) = 0.001
P(BoxK) = 0.999
P(Letter) = 0.5
Heads is independent of Box1 and BoxK.
I assert that the conditions on when letters get sent imply:
P(Heads and Letter and Box1) = P(Heads and Box1) = P(Heads) x P(Box1) = 0.5 x 0.001 = 0.0005
P(Heads and Letter and BoxK) = 0
Thus:
P(Heads and Letter) = P(Heads and Letter and Box1) + P(Heads and Letter and BoxK) = 0.0005 + 0 = 0.0005
and finally:
P(Heads given Letter) = P(Heads and Letter) / P(Letter) = 0.0005 / 0.5 = 0.001
Yes I do agree:
Here the asymmetry is 999:1, and—crucially—under Heads you might not receive a letter at all. That makes "I got a letter" genuine, discriminating evidence: it's 999× more likely if Tails.
This is why the 1/1000 answer feels almost unavoidable here, while the analogous "thirder" answer (1/3) in classic SB remains controversial.
The Halfer Objection (and Why It's Weaker Here)
A halfer might argue: "The coin is still fair; nothing about my letter changes that objective fact."
But that objection loses force because you might not have been in the observation pool at all. In standard SB, Beauty is guaranteed to wake up at least once regardless of the flip—so some argue her waking teaches her nothing. Here, if Heads, 999 out of 1000 box owners observe nothing. Your observation that you did get a letter is informative in a way even halfers usually accept.
Upshot for Sleeping Beauty
This variant supports the thirder intuition: conditioning on your own existence/experience within the outcome can shift your credence away from the prior, even when the prior is 1/2. The mailbox version just amplifies the effect to make it unmistakable, very good variant, did you come on it yourself ?
here is the problem with the 1/1000 answer, if i am the sender and run the process many times, each time with a different mailing list, the number of heads and tails will end up evening out at about 1/2 ratio, which aligns with the 1/2 answer, not the 1/1000 answer. that gives a reason to think that the true probability of heads is 1/2, but the proper betting odds from the perspective of the recipients is 1/1000.
I agree that the number of heads and the number of tails will be roughly equal. And the number of times you do and don't receive a letter will be roughly equal. But during each iteration of the process, there will still be a strong correlation between a given person receiving a letter and the coin landing tails.
On one of the videos I watched recently had an argument which temporarily weakened my believe in thirderism (it might have been linked here, but I couldn't find it this morning). I'm going to list it here, followed by the "flaw" that I found in it:
On Sunday morning, you offer SB a contract that says "If the coin came up heads, I'll pay $11 for this sheet of paper on Wednesday" for $5. Since the coin is fair, this is a good deal, so she buys it.
Then, when you wake her up, you offer to buy it back from her for $4. If she believes that the odds that it came up heads is only 1/3, then she should be willing to sell it for $4.
The two objections this raised to the thirder position are:
This violates the reflection principle (Which here I interpret as "If you are definitely going to believe something in the future, then you should believe it now.")
She can be taken advantage of to buy something for $5 and sell it for $4, which puts the Dutch Books shoe on the other foot.
And the rebuttal that I came up with was that she isn't really being given the choice of selling it for $4 (I.e. "SB trades ownership of the contract for $4"). She's really being offered the choice of "If SB still owns the contract, then she trades ownership of the contract for $4." which is a bad trade.
If she is to be awakened once (coin is heads, 50%) and she says she says heads, she will be right. There is a 50% chance of this happening, assuming she will be answering heads.
If she is to be awakened twice and she says tails (which we can assume she would do both times) she will be wrong twice. There is a 50% chance this will be happening, assuming she will be saying heads.
So saying "heads" will make her right i 1/3 of the possible cases. Conversely, saying tails will make her right in 2/3 of the possible cases.
Heads and tails are equally likely, but the frequency of her being asked depends on the coin. The very fact that she is being asked and does not know if she has already been asked the day before gives her additional information. It may not seem like it, since she cannot remember, but it does. She will be right more often if she says tails.
At what point the amount you see in the transparent box could change you from a one-boxer into a two-boxer, would it be $500 000, or $900 000 ?
If sleeping beauty is able to have rational thinking while asleep during Tuesday, she knows she is sleeping on a Tuesday, what is her credence for heads then? If you think it is 1 for P(H∣daydream) as she knows otherwise she would not be asleep.
Will you agree on this equation P(H) = P(awake)P(H∣awake)+P(daydream)P(H∣daydream), then would you agree also on P(awake)=3/4 and P(H)=1/2, this makes:
Dvd Avins, on 2026-04-21, wrote: "Heads and tails are equally likely" - and I think we all agree on that. However, before Dvd wrote that, Dvd wrote (in the same reply): "So saying heads will make her right in 1/3 of the possible cases. Conversely, saying tails will make her right in 2/3 of the possible cases." and I'm not sure I fully agree with that - what if when SB says "heads" we immediately (re)ask "can you please repeat that so I can make sure I heard you correctly?" and she says again "heads" (but we don't re-ask when she says "tails", except on Tuesday per the explanation/question/ original-setup). The original SB question states it is a "fair coin" so by definition it is a 50/50 chance of H or T - and thus answering either H or T is a 50% chance of being correct if the person answering has no additional info - no mastter how many times they are asked... Yes? No?
The key sentence is ** if the person answering has no additional info **, and all the debate is on that, does being awake for sleeping beauty give her additional info? Does dreaming and knowing it give additional info? Does receiving a letter give additional info? Thirder think so.
Yes, so SB has no additional info, so their most accurate answer would be: since I didn't see the coin toss, and I have no additional information indicating how it landed AFAIK, I can only say that from my perspective there is a 50% chance the coin landed on Heads and 50% chance it landed on tails... However, if we want to talk about gambling and attempting to make money, then Dvd Avins is on to something. How much will SB be paid if she guesses correctly? Will she be paid the same if it lands Heads and SB guesses Heads (once) and if it lands Tails and SB guesses Tails twice? Or will she be paid more (x2?) because she answered correctly twice?
SB doesn't have enough information to know how the coin landed, and we don't have enough information to know how to bet/gamble/answer! lol
...and since SB was told all the details of how the experiment will be done, she could state: "if I was asked this same question yesterday, or will be asked this same question tomorrow, the coin was tails, and if nobody will not be asking me this question tomorrow then the coin landed on heads." But that is only stating what SB had been told (how the experiment would be performed); so redundant. The setup strongly implies she can't ask what day it is when she gets interviewed {quote: "she will not be able to tell which day it is or whether she has been awakened before"}... But maybe she can ask: "What the heck are you trying to prove with this experiment?" :D
From the point of view "What would be rational?" I'd probably recommend taking both boxes at about $800K. But as a member of "Tribe 1-box" I'd probably hold out for anything less than $1M as a political statement, even though there is a chance that the referee would incorrectly calculate that I'd fold and thus fail to put the money in when it got to $950K, for instance.
I am going to flip a coin. If it comes up heads, I will roll a d6. If it comes up tails, I will roll a d20.
Whatever number "n" I roll, I will ask SB for the next "n" days what credence she gives to the event that the coin came up heads -- no amnesia or uncertainty about what day it is this time!
I think the choice should depend on the credence you give the Oracle to be good, call that p, and what’s in in transparent box Tb, as a one-boxer you should estimate your ev as p.1000000+(1-p)Tb, this is greater than one million if Tb>= 1000000.
I just came up with a variant on the mail box example. There's a raffle with 1000 tickets, sold for $1 each. The officials will flip a coin and draw a number.
If the coin lands heads, the person whose number was drawn will receive $999.
If the coin lands tails, the remaining 999 people will receive $1 each.
You get a letter in the mail that says on the outside "You were a winner in the raffle, your prize is enclosed".