{"id":"e926040f-7ca1-43c9-9d55-7dc4f156a034","arxiv_id":"2411.13257","paper_version":3,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A new probability framework shows the Sleeping Beauty 1/3 and 1/2 answers arise from two incompatible sets of principles, with no measure satisfying all four.","lead":"This paper builds a precise probability framework for the Sleeping Beauty problem and other cases where an observer has identical copies. It shows that the two main answers, 1/3 and 1/2, correspond to two different probability measures that cannot both satisfy all four natural principles.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.3 is sound, but its uniqueness is carried entirely by the normative PEI axiom (Eq. 2.4); if PEI is rejected, as [15,22,24] do, the Sleeping Beauty model retains a one-parameter family, so the Thirder measure is not forced by the Kolmogorov framework alone.","rationale":"The paper is a clean mathematical contribution: the proofs of Theorem 2.3 and Corollary 2.7 are valid, and the framework is explicit about which axioms are being used. The reader's weakest-assumption analysis is correct: PEI is the load-bearing normative premise. I do not see an internal inconsistency or a technical error that would overturn the accepted verdict. The concern is that the central theorem's uniqueness is conditional on a contested principle, and the paper itself flags this limitation. Because the paper frames its contribution as a comparative framework rather than a derivation of PEI from first principles, this concern does not change the verdict. The concrete test above isolates the role of PEI in the simplest possible model and confirms exactly where the philosophical disagreement lives.","tokens_in":19392,"tokens_out":26088,"duration_ms":289506,"concrete_test":"In the standard Sleeping Beauty model of Example 2.9, set a=PA({(0,1)}), b=PA({(1,1)}), so PN gives a+b≤1 and PA(Heads)=a. Enforce PI: PA(Mon|Tails)=b/(1-a)=1/2, hence b=(1-a)/2. Then enumerate PA(Heads)=a over a∈[0,1] with b=(1-a)/2. Enforce PEI: a/(a+b)=1/2, giving a=b, so a=1/3. Enforce PP instead: a=1/2, b=1/4. This one-parameter family shows that the uniqueness in Theorem 2.3 is precisely the content of PEI; the remaining normative question is whether the phone-call thought experiment justifies PA(·|S=x)=P(·|x∈X), which no theorem in the paper settles.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 2.3 is mathematically correct, but the theorem's conclusion is conditional on PEI. PEI (Eq. 2.4) asserts that the AO's centered information {S=x} and the CO's uncentered information {x∈X} yield the same FW-credences. This is not a consequence of Kolmogorov's axioms; it is a normative bridge between centered and uncentered evidence. In the proof, PEI is used to obtain the key ratio t_B/(q_B|B|)=s_x/Q_x (Eq. 2.14) and again to extend the measure to all F∈FW. Without PEI, even with PN and PI, the measure is not unique: in Example 2.9, PN and PI give the one-parameter family b=(1-a)/2, with PA(Heads)=a; PEI selects a=1/3, while PP selects a=1/2. The paper explicitly acknowledges that PEI is not derived and that [15,22,24] reject similar assertions. Theorem 2.15 only shows PEI is equivalent to PNFI under Assumption 2.12, so it does not make PEI compulsory. Thus the claim that Thirders and Halfers are precisely characterized is accurate only relative to accepting PEI; the mathematical framework itself does not adjudicate the philosophical disagreement.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a generalized Sleeping Beauty model on the product space Ω = Ω_O × Ω_A, with a finite set K of cells, a random occupied set X, and an auxiliary observation process Z. It defines four principles for an anthropic observer's probability PA: null sets (PN), indifference over occupied cells (PI), equivalent information with a classical observer (PEI), and Lewis's Principal Principle (PP). The central Theorem 2.3 shows that when the graph induced by joint occupancy is connected, PN, PI, and PEI determine a unique measure PE given by PE(F) = E(1_F X)/E(X), and Corollary 2.7 shows that all four principles are compatible only if |X| is deterministic given nonemptiness; otherwise one must choose between PE and a Halfer measure PL. The paper then studies conditioning on observations, proves an equivalence between PEI and a 'no future information' condition, identifies an improper-conditioning error in Hartle–Srednicki, and applies the framework to examples on the probability of life, a two-zone universe, and the cosmological constant.","tokens_in":19709,"tokens_out":11912,"duration_ms":122974,"significance":"The main theorem is clean and its proof is transparent, and the paper explicitly separates mathematical consequences from normative principles. The precise comparison of PE and PL within one Kolmogorov space is a genuine contribution, and the critique of Hartle–Srednicki in §3.3 is mathematically valid. The cosmological examples usefully show where the two measures agree and disagree. The main limitation is acknowledged in the text: uniqueness is conditional on accepting PEI, so the contribution is a comparative framework rather than an unconditional derivation of Thirding from Kolmogorov's axioms.","major_comments":[{"comment":"The equivalence between (PEI) and (PNFI) as stated needs an explicit nondegeneracy assumption. PNFI uses the conditional PA(G|S=n), which is defined only when PA(S=n)>0, while PEI constrains PA(·|S=x) only when PA(S=x)>0. Under Assumption 2.12 and PN alone, PEI does not force PA(S=n)>0 for every n: for example, with M=2 and q1=q2=1/2, the measure PA defined by PA(G×{1})=P(G) and PA(G×{2})=0 satisfies PN and PEI vacuously for x=2 but fails PNFI for n=2 because the conditional is undefined. Adding the condition PA(S=n)>0 for all n≤M (automatic if PI is also assumed) repairs the theorem and does not affect Theorem 2.3.","section":"§2.2, Theorem 2.15 and Definition 2.14"},{"comment":"The proof begins 'As PA satisfies (PN), (PI), and (PEI) it is equal to PE,' which invokes the uniqueness part of Theorem 2.3. That theorem requires the graph (K,E) to be connected, but Corollary 2.7 does not state this hypothesis. If connectedness is intended as a standing assumption of the section, it should be stated explicitly in the corollary; otherwise the proof needs a short direct argument. The conclusion itself appears to remain true in the disconnected case, since disconnectedness forces |X| to be 0 or 1 almost surely, but this needs to be said.","section":"§2.1, Corollary 2.7"}],"minor_comments":[{"comment":"The displayed formula P(SR|HER) = f(p,M)/(f(p,N)+f(p,N)) has a repeated term in the denominator; the preceding Bayes computation gives f(p,M)/(f(p,M)+f(p,N)). Please correct the typo.","section":"§3.3, Eq. (3.12)"},{"comment":"The sentence 'If Y ≥ 0 then E(1/Y ) ≥ 1/E(Y )' is not the inequality actually used. The displayed bound relies on E(Yθ/(X0+Yθ)) ≥ E(Yθ)/E(X0+n0), which follows from independence and Jensen applied to 1/(X0+n0). Please repair the sentence.","section":"§4.3, proof of Lemma 4.2"},{"comment":"After introducing α, the symbol κn (or κ_n) is used both for the binomial parameter times n and for n^{1−α}; please distinguish these quantities explicitly to avoid confusion.","section":"§4.3, notation"},{"comment":"The step labelled 'a further application of (PEI) proves (2.8)' would benefit from one line of detail, since (2.8) is the formula used throughout the rest of the paper.","section":"§2.1, proof of Theorem 2.3(b)"}],"recommendation":"minor_revision","confidential_remarks":"The paper is within scope for a mathematical probability journal with interdisciplinary applications, and the philosophical caveat about PEI is handled honestly in the body. The two technical issues in Theorem 2.15 and Corollary 2.7 are local and fixable. I see no need for further external review."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Martin Barlow has written a careful, genuinely useful paper. The new thing is not the size-biased weighting—he credits that—but the axiomatic setup on Ω = Ω_O × Ω_A: four named principles, uniqueness Theorem 2.3, impossibility Corollary 2.7, the PEI/PNFI equivalence in Theorem 2.15, and the restriction property for PE. The proofs are clean and the main line checks out. The Hartle-Srednicki correction in Section 3.3 is a real service; equation (3.11) double-counts observers exactly as Barlow says, and his 1/1001 answer for M=1, N=1000 is the right intuition.\n\nThe soft spot is the one the paper itself flags. PEI, equation (2.4), is doing all the work. It is not a consequence of Kolmogorov's axioms; it is a normative bridge between the classical observer's uncentered information and the anthropic observer's centered information. Without PEI, Example 2.9 leaves the one-parameter family b = (1−a)/2, and Theorem 2.3's uniqueness collapses to a free parameter. The paper is honest about this—it notes that [15,22,24] reject similar assertions, and Remark 2.2 allows that different real-world experiments may justify PEI differently. So the Thirder answer is characterized, not compelled. I do not count this as a flaw, because the stated contribution is a framework that lets competing positions be compared precisely, and that is true. A reader hoping for an unconditional resolution of Sleeping Beauty will be disappointed, but the paper's language is appropriately careful.\n\nMinor quibbles: Theorem 2.15's equivalence relies on Assumption 2.12, which is mild but does real lifting; the paper says so. The Four Beauties example is nice, but its 'common sense' reading presupposes PEI. Section 4's cosmology examples are illustrative rather than fitted. None of this threatens the main contribution.\n\nI would send this to a serious referee. It is a genuine mathematical framework for a literature that mostly argues informally, and the Hartle-Srednicki correction alone justifies publication. The paper is aimed at probabilists, philosophers of credence, and cosmologists doing anthropic reasoning. I would cite it.","headline":"A clean, honest framework for the Sleeping Beauty/anthropic inference debate; the uniqueness theorem is real but rides on the normative PEI axiom, and the Hartle-Srednicki correction is a solid concrete payoff.","tokens_in":20236,"tokens_out":2017,"would_cite":true,"duration_ms":21351,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60A05","60A10"],"pacs":[],"model":"deepseek-v4-flash","headline":"On a single Kolmogorov probability space, the Sleeping Beauty problem has a unique answer—the Thirder measure—forced by three natural principles.","keywords":["Sleeping Beauty problem","anthropic inference","size-biased distribution","Kolmogorov axioms","principle of equivalent information","credence","self-locating belief","multiverse"],"falsifier":"Run the Four Beauties experiment (or a Sleeping Beauty variant) with participants who can communicate fully while awake: if, after A and B have shared everything they know, A's credence that she is the Waker differs from B's credence that she is the Waker, then PEI fails and the unique-measure conclusion collapses. The paper itself shows that the Halfer measure $P_L$ predicts exactly such a divergence (A gives 1/4, B gives 3/4), so the experiment separates the two frameworks. A purely mathematical falsifier would be a connected $(K,E)$ for which the explicit measure (2.8) violates one of (PN), (PI), (PEI).","tokens_in":19197,"feed_emoji":"🛌","tokens_out":10093,"duration_ms":91536,"temperature":0.7,"pith_summary":"The paper attempts to show that the Sleeping Beauty problem and related anthropic inference puzzles can be placed on an ordinary Kolmogorov probability space, and that within that space a precise uniqueness theorem settles the main debate. It builds one space that contains both the objective randomness (which world is actual) and the observer's location (which cell or day), and then imposes three principles: a mild null-set condition, an indifference condition among occupied cells, and a principle of equivalent information between a classical observer and an anthropic observer who share all their information. The main theorem states that these three principles force a unique probability, which in the original Sleeping Beauty problem gives 1/3 for Heads—the Thirder answer. A fourth principle used by Halfers, the Principal Principle, is shown to be incompatible with the other three unless the number of observers is deterministic. If the framework is right, the philosophical debate is replaced by a transparent mathematical statement of which axioms one accepts.","feed_headline":"Three principles force the Thirder answer in Sleeping Beauty","feed_subtitle":"A single Kolmogorov space shows the Halfer answer must drop the principle that equal information gives equal probability.","key_machinery":"The load-bearing object is the product space $\\Omega = \\Omega_O \\times \\Omega_A$ built from the objective random worlds $\\Omega_O$ and the observer-location set $\\Omega_A = K \\cup \\{\\partial\\}$, with $S$ marking the anthropic observer's cell. The mechanism is the Principle of Equivalent Information (PEI), equation (2.4): $P_A(F|S=x) = P(F|x\\in X)$, which asserts that an anthropic observer in cell $x$ and a classical observer who learns $x\\in X$ must agree on every objective event $F$. In the proof of Theorem 2.3, PEI is what forces the ratios $P_A(S=x)/P(x\\in X)$ to be constant across the graph $(K,E)$, and connectedness turns this into a single constant $\\lambda = 1/\\mathbb{E}(X)$. The resulting identity $P_E(F) = \\mathbb{E}(1_F X)/\\mathbb{E}(X)$ is the whole argument in miniature: the available principles act only through the random number of occupied cells, reweighting objective probabilities by the observer count.","core_discovery":"The central claim is that once the observer's location is placed on an equal footing with the objective world, the ambiguity in the Sleeping Beauty problem disappears. On the space $\\Omega = \\Omega_O \\times \\Omega_A$, the principles (PN), (PI), and (PEI) determine a unique probability $P_E$ whenever the graph whose edges are pairs of cells that can both be occupied is connected. The measure is explicit: $P_E(F \\cap \\{S=x\\}) = \\lambda P(F \\cap \\{x\\in X\\})$ with $\\lambda = 1/\\mathbb{E}(X)$, equivalently $P_E(F)=\\mathbb{E}(1_F X)/\\mathbb{E}(X)$, i.e., the number of observers acts as a Radon–Nikodym weight. Applied to Sleeping Beauty this gives $P_E(\\text{Heads}) = 1/3$, identifying $P_E$ as the Thirder measure. The paper further shows (Corollary 2.7) that if one adds the Principal Principle (PP), then a probability satisfying all four principles exists only when $|X|$ is deterministic conditional on $X \\neq \\emptyset$; the Halfer measure $P_L$ satisfies PN, PI, and PP but violates PEI, which is why it can disagree with $P_E$.","pith_inferences":["If PEI is accepted as a general rationality requirement, the framework reframes anthropic decision theory: any agent with multiple copies should evaluate decisions under the size-biased measure $P_E$, and many copy-related paradoxes in sequential decision problems would be traced to the choice of measure rather than the conditioning rule.","The identity $P_E(F)=\\mathbb{E}(1_F X)/\\mathbb{E}(X)$ offers a compact observer-weighted expectation that could be applied directly to fine-tuning arguments: a multiverse theory is automatically more probable for an observer to the extent that it produces more observers, with no separate 'we exist' conditioning step.","The restriction property suggests a testable consistency check for anthropic models: conditioning an observer-weighted posterior on being in a subset $X'$ of locations must coincide with recomputing the size bias from $X'$; any model that violates this will produce internally inconsistent inferences after further location information arrives.","One could attempt to derive PEI from an exchangeability or Dutch-book argument; if such a derivation succeeds, the Sleeping Beauty controversy would be resolved in favor of Thirders on purely decision-theoretic grounds."],"forward_implications":["In the standard Sleeping Beauty problem, the unique $P_E$ assigns $P(\\text{Heads})=1/3$; the Halfer answer $1/2$ is exactly the measure $P_L$ that satisfies PN, PI, and PP but violates PEI.","No probability measure can simultaneously satisfy PN, PI, PEI, and PP unless the number of observers given $X\\neq\\emptyset$ is constant (Corollary 2.7).","$P_E$ has the restriction property: learning that the observer lies in $X'\\subset X$ yields the probability obtained by building the model with $X'$, so subset conditioning is consistent with size biasing.","For sequential experiments with a future randomization, PEI is equivalent to the principle of no future information: the anthropic observer and the classical observer agree on all events that lie in the future.","In multiverse models with many observers and many independent components, the predictions of $P_E$ and $P_L$ converge in the large-volume limit, so for cosmological constant estimates the Thirder/Halfer distinction becomes numerically negligible."],"supporting_citations":[{"why":"It supplies the conditional-probability argument (1.1)–(1.3) whose strategy the proof of Theorem 2.3 explicitly adapts.","marker":"[12]"},{"why":"It introduces the SSA/SIA distinction and notes that the $P_E$ law is the size-biased distribution of $|X|$, connecting the theorem to established anthropic terminology.","marker":"[7]"},{"why":"It uses an assertion close to PEI in a mathematical analysis of Sleeping Beauty, serving as the precedent the paper points to for the key principle.","marker":"[27]"},{"why":"It rejects conditioning on events such as $\\{\\text{Mon}\\}$ for some probability laws, the exact operation PEI requires; the contrast motivates the principle.","marker":"[14]"},{"why":"It also rejects assertions similar to PEI from an asynchronous-systems perspective, marking the main philosophical obstacle the paper addresses.","marker":"[15]"},{"why":"It provides the Four Beauties example, used to exhibit the sharp disagreement between $P_E$ and $P_L$ and to frame the challenge events for Halfers.","marker":"[24]"}],"fun_headline_variants":["Equal footing for observers yields definitive Sleeping Beauty answer","Observer's location as objective: Thirder wins","Three principles pick one Sleeping Beauty probability","Radon-Nikodym weight gives Thirder in Sleeping Beauty","Halfer fails when observer is part of the model"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the Principle of Equivalent Information (PEI): an observer who wakes in a known cell and an outside observer who knows that cell is occupied must assign the same probability to every objective event, because they are assumed to have the same information.","fun_headline_variants_meta":{"raw":{"variants":["Equal footing for observers yields definitive Sleeping Beauty answer","Observer's location as objective: Thirder wins","Three principles pick one Sleeping Beauty probability","Radon-Nikodym weight gives Thirder in Sleeping Beauty","Halfer fails when observer is part of the model"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001215,"raw_usage":{"total_tokens":4943,"prompt_tokens":831,"completion_tokens":4112,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":447,"completion_tokens_details":{"reasoning_tokens":4038}},"tokens_in":447,"tokens_out":4112,"duration_ms":29535,"temperature":1.0,"reasoning_tokens":4038,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:38:39.639729+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the Four Beauties experiment (or a Sleeping Beauty variant) with participants who can communicate fully while awake: if, after A and B have shared everything they know, A's credence that she is the Waker differs from B's credence that she is the Waker, then PEI fails and the unique-measure conclusion collapses. The paper itself shows that the Halfer measure $P_L$ predicts exactly such a divergence (A gives 1/4, B gives 3/4), so the experiment separates the two frameworks. A purely mathematical falsifier would be a connected $(K,E)$ for which the explicit measure (2.8) violates one of (PN), (PI), (PEI).","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the conditional-probability argument (1.1)–(1.3) whose strategy the proof of Theorem 2.3 explicitly adapts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It introduces the SSA/SIA distinction and notes that the $P_E$ law is the size-biased distribution of $|X|$, connecting the theorem to established anthropic terminology."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It uses an assertion close to PEI in a mathematical analysis of Sleeping Beauty, serving as the precedent the paper points to for the key principle."},{"cited_title":"Gr¨ omping","cited_arxiv_id":null,"evidence_quote":"It rejects conditioning on events such as $\\{\\text{Mon}\\}$ for some probability laws, the exact operation PEI requires; the contrast motivates the principle."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It also rejects assertions similar to PEI from an asynchronous-systems perspective, marking the main philosophical obstacle the paper addresses."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It provides the Four Beauties example, used to exhibit the sharp disagreement between $P_E$ and $P_L$ and to frame the challenge events for Halfers."}],"review_version":1}