The Conclave Process

Guanyi Chen; Itai Benjamini; Shuyang Gong; Zhangsong Li; Zhenhao Cai

REVIEW 2 major objections 4 minor 12 references

The Conclave Process

T0 review · 2 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper proves that in a voting model where the probability of voting for a candidate is proportional to the α-th power of that candidate's previous vote count, the absorption time is (1+o(1))·2 log log n / log α for every α>1, and that

desk verdict A genuinely interesting new process, but the 1<α<2 regime rests on a false inequality in Lemma 5.6 that I don't see how to patch from the text. read the letter →

arxiv 2607.22324 v1 pith:4FGTGMIQ submitted 2026-07-24 math.PR math.CO

Itai Benjamini , Zhenhao Cai , Guanyi Chen , Shuyang Gong , Zhangsong Li This is my paper · ORCID

classification math.PRmath.CO MSC 60J1060K35

keywords reinforcedvotingprocesspower-lawreinforcementabsorptiontimelog-logconsensusphasetransitionPoissonapproximationwinnerselectionlevel-setmethod

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper studies a stochastic voting process in which n cardinals vote repeatedly and the probability that a candidate receives a vote is proportional to the α-th power of that candidate's current tally. It establishes a sharp transition at α=1: for any α>1, consensus is reached in about 2 log log n / log α rounds with probability tending to 1, whereas for α<1 the process typically needs at least exp(Ω(n)) rounds. It also identifies exactly when the eventual winner becomes predetermined: if 2^{1/k}<α<2^{1/(k-1)}, the unique leader after round k is the eventual winner with probability 1−o(1), while the round-(k−1) leaders are not. A sympathetic reader cares because the model captures a general reinforcement mechanism — success begets success — and the results show that such reinforcement produces ultra-fast consensus and a sharp predictability transition.

What carries the argument

The load-bearing object is the pair (maximum vote count M_t, relative gap δ_t between maximum and second maximum) plus, for 1<α<2, the level-set counts |A_t(j)| = #{i : X_i^{(t)} = M_t − j}. The core mechanism is the deterministic-looking recursion 1−δ_{t+1}≈(1−δ_t)^α once the leader has a large enough lead, together with a level-set lemma that controls how many candidates sit at each distance below the maximum. These estimates feed a recursion for the number of candidates at each level set; the balance between enumeration gain and expected-value loss determines the critical round k and the initial gap δ_0=(log n)^{-1+o(1)}, which via Proposition 4.4 yields the constant 2/log α.

What would settle it

Run the voting process with α=1.9 for n=10^4,10^5,10^6 and track whether the eventual winner was among the first-round leaders; Theorem 1.6 predicts this probability tends to 0, so any flattening above 0 refutes the phase transition.

Watch

Extended reading notes

Core claim

The central claim is a complete asymptotic classification of the absorption time and winner identity in the power-law reinforced voting process. For α>1 the process reaches full consensus in (1+o(1))·2 log log n / log α steps; the proof decomposes the dynamics into a pre-heuristic phase in which leaders at successive rounds migrate through level sets of the vote distribution, and a heuristic zone in which the maximum and second maximum evolve deterministically via the map 1−δ'≈(1−δ)^α. For 1<α<2, the level-set analysis shows the winner is selected at round k exactly when α lies between 2^{1/k} and 2^{1/(k-1)}, with all earlier rounds' leaders eliminated; for α>2 the winner is fixed already a

Load-bearing premise

The exact thresholds 2^{1/k} and the constant 2/log α rest on the level-set estimates of Lemma 3.8 being accurate to a (log n)^{o(1)} factor; if those Poisson count estimates fail by more than that, the phase boundaries and log-log constant shift.

Editorial extensions

If this is right

For any α>1, consensus time does not grow with n beyond log-log; even astronomically large electorates finish in a handful of rounds, matching the observation that real conclaves end quickly despite n≈100.
The threshold sequence λ_k=2^{1/k}: for α between λ_k and λ_{k-1}, the event of the eventual winner is determined in round k, so the later a leader emerges, the smaller α is; round k−1 gives no information about the winner.
At α>2, the conditional constants differ: if the first-round leader is unique, T≈2 log log n/log α; if not, T≈(1+α/2) log log n/log α.
For α<1, the process is essentially frozen at exponentially long time scales; any majority is exponentially unlikely, so absorption time ≥ exp(Ω(n)).
The empirical vote distribution after k rounds is close to a universal Poisson-mixture law μ_k, with partition function close to n z_k, so early-round dynamics are asymptotically distributionally tractable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

An immediate testable extension is to replace x^α by a general increasing reinforcement function μ and ask whether the thresholds become the points where an iterated map μ^{∘k}(m)/m crosses 2; the paper's techniques suggest such a generalization.
The log-log law implies that the same model could serve as a stylized explanation for fast consensus in winner-take-all elections or in ranking algorithms, since a few rounds of proportional-to-power updates concentrate an enormous field into a single leader.
The paper's conjecture that at the exact critical α=2^{1/k} the winner is decided at round k+1 rather than k could be tested numerically where the bounds are tight; a Monte Carlo estimate of P(W∈L_k) at α=2^{1/k} would settle it.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Quick take: this is a genuinely interesting process and the paper has the right shape — a sharp phase diagram and a winner-determination ladder — but the proof of the 1<α<2 branch has a load-bearing error that I don't see how to patch from the text. If you pick this up expecting the full theorem, you will find the core lemma (5.6) applies Lemma 3.8(3) in a regime where its hypotheses fail.

What's new and good: Definition 1.1 (power-α multinomial reinforcement) is natural and, as far as the citations go, new. The α=1 reduction to Wright-Fisher/Kingman is standard and correctly credited. The α<1 exponential lower bound (Theorem 1.5) is clean and easily checked — the p_i ≤ 1/(1+3^{-α}) < 3/4 bound does the job. For α≥2, the two-step analysis via occupancy and Poisson coupling is plausible and has no obvious gaps. The absence of fitted constants or self-citations is a real plus; the thresholds 2^{1/k} and constants 2/log α, (1+α/2)/log α are derived, not tuned. The simulations are honest.

Now the soft spot, and it's a big one. In Lemma 5.6, proof of (5.28), Case α^t<2, the authors need Lemma 3.8(3) for v=j. That lemma requires v ≥ 4S(λ,m) with S = √λ exp(√log λ) √log m. In the intended application, λ=O((log n)^{α^t+o(1)}), log m = (log n)^{2-α^t-o(1)}, so S = (log n)^{1-o(1)}. But j = (log n)^{α^t-1+o(1)}. Since α^t<2, α^t-1<1, so j ≪ S for any α^t bounded away from 2? Actually even if α^t close to 2, j ~ (log n)^{1-o(1)} as well; compare to S ~ (log n)^{1-o(1)} times exp(sqrt(log H))? Actually S ~ (log n)^{1-o(1)} so j could be comparable if α^t close to 2, but they claim 4S < j in (5.54). The algebra (5.54) is simply wrong: the right side should have √log m in the numerator, not the denominator. So the hypothesis fails.

Maybe this can be fixed: perhaps the authors meant to use a different S or a different level-set lemma. But as written, the lower bound for |A'_{t+1}(j)| and hence the induction (5.28) is unsupported. Since Proposition 5.5 and the sharp constant 2/log α for 1<α<2 rest on this step, the main theorem is not established for that range. That's not a minor gap.

If the authors can correct (5.54) or find an alternative bound, the paper would be significant. As it stands, my advice: send to a competent referee for α≥2 and α<1 parts, but tell the referee to focus on Lemma 5.6 and not let the polished intro hide the error. The α=1 literature summary is worth citing.

Referee Report

2 major / 4 minor

Summary. The paper introduces a reinforced voting process on n candidates in which each voter in each round chooses a candidate with probability proportional to the α-th power of that candidate's previous vote count. The main claims are: for α>1 the absorption time satisfies T=(1+o(1))·2 log log n / log α with high probability, with a conditional variant for α>2; for 2^{1/k}<α<2^{1/(k-1)} the round-k leader is the eventual winner with probability 1−o(1), while round-(k−1) leaders are not; for α<1 the absorption time is at least exp(Ω(n)) with high probability; and the empirical vote distributions in constant rounds are close to an explicitly defined sequence of Poisson mixtures. The proof is built around a Poisson-coupling framework, level-set estimates for maxima of i.i.d. Poisson variables, a multi-step induction in Lemma 5.6, and a final 'heuristic zone' analysis in Proposition 4.4.

Significance. If the main theorems hold, the paper gives a sharp and surprising phase transition at α=1, with the same exponent α controlling both the consensus time scale and the round at which the winner is predetermined. The paper is also commendably non-circular: the thresholds λ_k=2^{1/k}, the constants 2/log α and (1+α/2)/log α, and the recursion (μ_k,z_k) are derived rather than fitted, and the simulations in §1.2 are consistent with, not calibrating, the theory. The α<1 proof in §6 is short and convincing, and the α=1 identification with Wright-Fisher/Kingman coalescent is standard. However, the entire 1<α<2 branch, including Theorems 1.3 and 1.6 for all k≥2, rests on Lemma 5.6 and the level-set estimates of Lemma 3.8; one displayed inequality in the proof of Lemma 5.6 is algebraically inconsistent with the definition of S(λ,m), and the affected lower-bound estimate is load-bearing. The sharp results for 1<α<2 are therefore not established as written.

major comments (2)

[§5.2.4, Eq. (5.54) and Lemma 3.8(3)] The proof of the lower bound in (5.49) applies Lemma 3.8(3) to the i.i.d. Pois(H_t(λ)) variables on A_t(λ), requiring 4S(H_t(λ),|A_t(λ)|) ≤ j. The displayed inequality (5.54) claims 2S(H,|A|) ≤ exp(√log H)·√(H/log|A|). But by the definition (3.7), S(H,m)=√H exp(√log H)√log m, so this inequality is equivalent to 2√H exp(√log H)√log m ≤ exp(√log H)√H/√log m, i.e. 2 log m ≤ 1, which is false in the regime where |A_t(λ)| is large. In the intended application, log|A_t(λ)| ≈ (log n)^{2−α^t+o(1)} and √H_t(λ) ≈ (log n)^{α^t/2+o(1)}, so S ≈ (log n)^{1+o(1)}, while j ≈ (log n)^{α^t−1+o(1)} with α^t−1<1 for α<2. Thus the required condition 4S≤j fails badly, and Lemma 3.8(3) cannot be used to obtain (5.56). Since (5.28) is the inductive engine of Lemma 5.6 and Proposition 5.5, the proofs of Theorems 1.3 and 1.6 for 1<α<2 are incomplete as written.
[§5.2.4, around (5.52)–(5.56)] The transfer from the Poisson levels A'_{t+1} to the multinomial levels A_{t+1} is asserted via (5.31)–(5.32) and the relation A_{t+1}(λ)⊂∪_{−8≤u≤8} A'_{t+1}(λ+u). This relation is plausible but is not proved in detail, and the constants 8, 1/2 are not derived from the coupling. More importantly, the lower-bound part of the transfer uses the invalid inequality (5.54); without a valid lower bound on |A'_{t+1}(λ,j)|, the lower bound in the definition of E'_{4,t+1} is unsupported. This is the same load-bearing failure as in the previous comment, but I flag it separately because it is the exact step that feeds the induction for all t≤m_0.

minor comments (4)

[Title] The title contains a spacing typo: 'CONCLA VE' should be 'CONCLAVE'.
[Remark 1.2] The assertion that the two-thirds-stopping-rule variant has absorption time (1+o(1))E[T] for α>1 is stated without proof. If this claim is not needed for the main theorems, it should be labeled as a conjecture or moved to an 'open problems' remark; as written it appears to be an unproved additional result.
[Eq. (5.54)] Even if the inequality in (5.54) were reversed, it would not help the proof, because the required condition for Lemma 3.8(3) is a lower bound on j, not an upper bound. The authors should re-derive the condition carefully and either correct the display or replace the argument.
[Notation, §5.2] The superscript notation n^-_t, n^+_t, N^-_t, N^+_t is easy to confuse with exponents and with negative numbers. A different notation, such as n_t^- and n_t^+, or n_{\mathrm{low},t}, n_{\mathrm{high},t}, would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's derivation is self-contained; thresholds and constants are derived from the model, not fitted or imported from self-citations.

full rationale

The paper introduces a new stochastic model and derives its absorption-time and winner-identification asymptotics mathematically. No parameter is fitted to data and then renamed as a prediction: the only model parameter is α, which is fixed in advance, and the asymptotic constants (2/log α, (1+α/2)/log α, λ_k = 2^{1/k}) are obtained by analyzing the evolution equations, e.g., from comparing exponents in Eq. (2.1) and from the recurrence (1−δ') ≈ (1−δ)^α. The recursive measures µ_k and constants z_k in Theorem 1.8 are defined to match the limiting Poisson dynamics, but Theorem 1.8 is a convergence theorem proving closeness of the empirical distribution to those measures; this is not circular because the statement is nontrivial and the definitions do not presuppose the convergence. The references contain no works by the present authors, so there is no self-citation chain. The simulations in §1.2 are illustrative consistency checks, not calibration inputs. A possible technical flaw in Lemma 5.6 (as suggested by a skeptical reading) would be a correctness issue, not a circularity issue. Overall, no step in the derivation reduces by construction to its own inputs.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The process has exactly one modeling input, the exponent α, which is held fixed in every theorem and never tuned; the thresholds λ_k, the constants 2/log α and (1+α/2)/log α, and the recursion (1.10) are outputs of the analysis, not fitted inputs. Proof-architecture constants such as (log n)^{101} and '10 rounds' in Proposition 5.4 are not free parameters. The single unproven auxiliary claim is Remark 1.2's two-thirds-rule assertion.

assumptions (5)

standard math Chernoff–Bennett concentration bounds for Binomial and Poisson variables (Lemmas 3.1–3.2, after [3])
Used throughout (e.g., Lemmas 3.7–3.9, 4.5, 5.3) to control fluctuations of vote counts; textbook results.
standard math Stirling's formula and Poisson point-process thinning used in the multinomial↔Poisson coupling (Lemma 3.12)
The key coupling bounding |ζ_i−η_i|≤4 and #{i: ζ_i≠η_i}≤2√n log n is built on a Poisson point process on R+ with intensity n (Section 3.3); it underpins Lemmas 5.1, 5.6 and Theorem 1.8.
standard math McDiarmid's inequality (Lemma 3.10)
Used to control TV distance between empirical distributions and their laws in the proof of Theorem 1.8.
domain assumption Definition 1.1 dynamics: each voter independently samples a candidate with probability ∝ (previous vote count)^α, rounds independent given the state
This is the object of study; all theorems are conditional on this model, which is postulated, not derived. The real-world conclave connection (Remark 1.2) adds an unproven bridge.
ad hoc to paper Remark 1.2: the two-thirds-stopping-rule variant has absorption time (1+o(1))E[T] for α>1, asserted without proof
Unproven claim connecting the mathematical absorption time (all n votes) to the real-world two-thirds rule; if false, the real-world interpretation weakens, but the theorems themselves are unaffected.

how reviews work

0 comments

Cite this review

Pith. "Pith review of The Conclave Process." pith.science (2026). https://pith.science/paper/4FGTGMIQ

@misc{pith2026260722324,
  author       = {Pith},
  title        = {Pith review of: The Conclave Process},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4FGTGMIQ}},
  note         = {Machine review of arXiv:2607.22324}
}

abstract

We introduce a stochastic model for the papal conclave in which $n$ cardinals vote repeatedly among themselves until one cardinal receives all the votes. In each round, the probability that a cardinal votes for a given candidate is proportional to the $\alpha$-th power of that candidate's vote count in the preceding round. For $\alpha=1$, the model reduces to the Wright-Fisher model and is dual to Kingman's n-coalescent. We reveal a sharp transition in the absorption time $\mathcal{T}$ at $\alpha=1$. It was known that when $\alpha=1$, $\mathcal{T}$ is typically of order $n$. We prove that for $\alpha>1$, it drops to order $\textit{loglog n.}$ In contrast, for $\alpha<1$, $\mathcal{T}$ is typically at least $\exp(\Omega(n))$. We also prove a sharp phase transition in the identity of the winner when $\alpha>1$. For every positive integer $k$, if $2^{1/k}<\alpha<2^{1/(k-1)}$ (where we write $2^{1/0} = +\infty$), with probability tending to 1 as $n\to\infty$, the eventual winner is the unique leader after round $k$. These results show that reinforced voting processes reach consensus remarkably quickly even for large electorates.

Figures

Figures reproduced from arXiv: 2607.22324 by the authors.

**Figure 1.** Simulation results on the conclave model. 1.3. Open problems. Our work reiterates a number of future research directions as we discuss below. The critical case α = λk. In comparison to Theorem 1.3, Theorem 1.6 does not include the result when α = λk. Our proof shows that, in this case, P α n (W ∈ Lk+1) = 1 − o(1), P α n (W ∈ Lk−1) = o(1). 5 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 2 linked inside Pith

[1]

Journal of Applied Probability , volume=

On the genealogy of large populations , author=. Journal of Applied Probability , volume=. 1982 , publisher=

1982
[2]

Electronic Journal of Combinatorics , volume=

Compositions of random functions on a finite set , author=. Electronic Journal of Combinatorics , volume=
[3]

Statistics & Probability Letters , volume=

Central limit theorem for the size of the range of a renewal process , author=. Statistics & Probability Letters , volume=. 2005 , publisher=

2005
[4]

Advances in Applied Probability , volume=

Expected coalescence time for a nonuniform allocation process , author=. Advances in Applied Probability , volume=. 2008 , publisher=

2008
[5]

Discrete Mathematics and Applications , volume=

Images of a finite set under iterations of two random dependent mappings , author=. Discrete Mathematics and Applications , volume=. 2016 , publisher=

2016
[6]

arXiv preprint math/0207276 , year=

Iterating random functions on a finite set , author=. arXiv preprint math/0207276 , year=

arXiv
[7]

and Ahn, H

Adler, I. and Ahn, H. and Karp, R. and Ross, S. , journal=. Coalescing times for. 2003 , publisher=

2003
[8]

and Steger, A

Raab, M. and Steger, A. , journal=. ``. 1998 , publisher=

1998

Show all 12 references

[9]

and Lugosi, G

Boucheron, S. and Lugosi, G. and Massart, P. , title =
[10]

Probability Surveys , volume =

A survey of random processes with reinforcement , author=. Probability Surveys , volume =
[11]

arXiv preprint arXiv:2310.00159 , year=

P\'olya urns on hypergraphs , author=. arXiv preprint arXiv:2310.00159 , year=

arXiv
[12]

2024 , publisher=

Random graphs and complex networks , author=. 2024 , publisher=

2024

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

Journal of Applied Probability , volume=

On the genealogy of large populations , author=. Journal of Applied Probability , volume=. 1982 , publisher=

1982

[2] [2]

Electronic Journal of Combinatorics , volume=

Compositions of random functions on a finite set , author=. Electronic Journal of Combinatorics , volume=

[3] [3]

Statistics & Probability Letters , volume=

Central limit theorem for the size of the range of a renewal process , author=. Statistics & Probability Letters , volume=. 2005 , publisher=

2005

[4] [4]

Advances in Applied Probability , volume=

Expected coalescence time for a nonuniform allocation process , author=. Advances in Applied Probability , volume=. 2008 , publisher=

2008

[5] [5]

Discrete Mathematics and Applications , volume=

Images of a finite set under iterations of two random dependent mappings , author=. Discrete Mathematics and Applications , volume=. 2016 , publisher=

2016

[6] [6]

arXiv preprint math/0207276 , year=

Iterating random functions on a finite set , author=. arXiv preprint math/0207276 , year=

arXiv

[7] [7]

and Ahn, H

Adler, I. and Ahn, H. and Karp, R. and Ross, S. , journal=. Coalescing times for. 2003 , publisher=

2003

[8] [8]

and Steger, A

Raab, M. and Steger, A. , journal=. ``. 1998 , publisher=

1998