Essay — April 2026

The Economy of Scholarship and Recognition

Abstract: A review request that pissed me off, why journals love prestige more than the science they champion, and how mathematics necessitates double-blind.

I recently received a request to referee a resubmission for a major physics journal. This was not unusual. I have refereed for this journal a number of times. I know the work, I know the field, and I am, apparently, qualified to evaluate what appears in their pages.

What I am not qualified to do, it seems, is publish there. My own submissions to this journal have been consistently desk-rejected — returned without review, without comment, without the basic courtesy of being read by a peer.

I want to sit with that asymmetry for a moment, because I think it deserves more attention than it usually gets.

Peer review is labor. It is skilled, time-consuming, and entirely unpaid. When a journal asks me to referee a paper, it is asking me to donate several hours of expert attention to its operation. I do this willingly, because I believe in the system. I believe that careful evaluation makes science better, that it protects readers from error, and that my participation in the process is part of what I owe to the community I belong to. Moreover, I actually learn from the process — refereeing keeps me honest, forces me to engage with work outside my immediate focus, and occasionally changes how I think about my own problems.

But the system is not symmetric. The labor flows in one direction; the recognition flows in another.

A journal that desk-rejects submissions from researchers at smaller institutions, in less prestigious countries, while simultaneously relying on those same researchers to keep its review pipeline running, is not operating a meritocracy. It is running a protection racket. Your expertise is good enough to guard the gate. Your science is not good enough to pass through it. That is not a filtering mechanism. That is extraction.

The second gate

Open access publishing adds a second gate. The article processing charge — typically ranging from two to five thousand dollars — is levied after acceptance. It is presented as a fee for dissemination, for making science freely available to the world, when in fact it does the exact opposite. A researcher at a well-funded institution files a form and the grant absorbs it. A researcher at a small college, with marginal institutional support and no large grant, faces a personal financial decision about whether their work can exist in public. The open access revolution, which was supposed to democratize science, has in many cases simply relocated the paywall — from the reader to the author — and paradoxically made science a functional theocracy, with its clerics, priests, altar boys¹, and indulgences.

The feedback loop

The deeper problem is structural, and even more sinister. The desk editor is not, in practice, making an independent judgment about scientific quality. They are not reading your work. They are evaluating your status: These are supposed to correlate with quality, and perhaps once they did. But the proxy is generated by the same system that consumes it.

Prestige accrues to researchers at prestigious institutions, who publish in prestigious journals, which reinforces the prestige of those institutions. The editor who filters by proxy is not standing outside this circuit — they are part of it. Over time, the system does not optimize for good science. It optimizes for the appearance of good science, as defined by its own prior outputs. The correlation between the proxy and the thing it was meant to measure quietly decays, and nobody is in a position to notice, because the mechanism for noticing has been replaced by the mechanism for reproducing.

This mechanism has a formal analog. In recent work on Bayesian polarization, I show that an agent exposed to an asymmetric evidence stream cannot escape convergence toward that stream's implied position through skepticism alone — because the agent's credibility assessment of the source is evaluated against a world-model the source itself shaped. The loop is self-sealing. The publishing system runs the same circuit. The desk editor cannot verify scientific quality directly — that coordinate is unverifiable without peer review, which is precisely what the desk filter withholds. So the editor falls back on the only observable: institutional prestige. But prestige is itself the output of prior publication decisions made by editors using the same heuristic. The proxy is generated by the system that consumes it. There is no ground truth anchor. The credibility basin deepens with every iteration.

Don Quixote before the waterfall and the noisy hovels — Gustave Doré. A tiny armoured figure on horseback faces an enormous cliff face. The institution is just rock. It does not notice him. — *The Waterfall and the Noisy Hovels* (I, 20) — Gustave Doré & Héliodore Pisan, 1863 — From *Don Quixote* — Wood engraving — Public domain — The hovels are noisy. The cliff does not care.

The dynamics can be modelled precisely. The natural phase space has two coordinates, both in $[0,1]$: prestige $p$ and publication probability $P_{\mathrm{pub}}(p)$. These are genuinely coupled. The publication probability is determined by three sequential gates:

$$ \begin{equation} P_{\mathrm{pub}}(p) \;=\; \underbrace{\sigma\!\bigl(w_1 p + (1-w_1)\bar{q} - \tfrac{1}{2}\bigr)}_{\text{Gate 1: prestige filter}} \;\cdot\; \underbrace{\sigma\!\bigl(\tfrac{p - \theta_2}{w_2}\bigr)}_{\text{Gate 2: APC threshold}} \;\cdot\; \underbrace{\bigl(1 - \mathbf{1}_{[p<0.45]}\cdot w_3\bigr)}_{\text{Gate 3: chilling effect}} \end{equation} $$

Gate 1 is the prestige filter at the desk: a logistic function in $p$ with weight $w_1$. Gate 2 is the APC threshold: a logistic in $p$ with center $\theta_2$ and width $w_2$. Higher prestige increases APC-clearing probability through better funding access. Gate 3 is the chilling effect: low-prestige researchers self-select out at rate $w_3$. The product of the three is the effective publication probability.

The flow equation for prestige is:

$$ \begin{equation} \dot{p} \;=\; f(p) \;=\; P_{\mathrm{pub}}(p)\cdot\alpha^{+} \;-\; \bigl(1 - P_{\mathrm{pub}}(p)\bigr)\cdot\alpha^{-} \end{equation} $$

where $\alpha^{+}$ is the prestige boost per publication and $\alpha^{-}$ is the erosion per rejection. The system has two attractors: the celebrated pole at $p\to 1$ and the expelled pole at $p\to 0$. Between them sits an unstable fixed point $p^*$ — the separatrix — determined by $\dot{p}=0$:

$$ \begin{equation} P_{\mathrm{pub}}(p^*) \;=\; \frac{\alpha^{-}}{\alpha^{+}+\alpha^{-}} \end{equation} $$

Quality does not appear in this equation. Geography does not appear in this equation. The content of your work does not appear. Only the gate parameters matter — and those are set by editors at institutions you have never visited.

― ✦ ―

Fig. 1 — Interactive phase portrait: prestige × publication probability

Basin → celebrated ($p\to 1$) Basin → expelled ($p\to 0$) System curve $P_{\mathrm{pub}}(p)$ Separatrix $P^*$ Orbit → celebrated Orbit → expelled

GATE PARAMETERS

Prestige weight $w_1$0.85

APC threshold $\theta_2$0.45

APC gate width $w_2$0.15

Chilling weight $w_3$0.60

Publication boost $\alpha^+$0.07

Rejection erosion $\alpha^-$0.012

Mean quality $\bar{q}$0.65

PRESETS

Each trajectory starts at a grid point $(p_0, P_0)$ and evolves under the coupled system. The solid curve is $P_{\mathrm{pub}}(p)$ — the attractor manifold. The dashed red line is the separatrix $P^* = \alpha^-/(\alpha^++\alpha^-)$: above it prestige grows, below it prestige erodes. Drag the sliders to watch the basin boundary shift.

― ✦ ―

Storming the Gate (Sturm) by Käthe Kollwitz, 1893-97. People hammer against an iron gate that does not open. — *Storming the Gate* (Sturm) — Käthe Kollwitz, 1893–97 — Sheet 5 of *Ein Weberaufstand* — Etching and sandpaper — Dallas Museum of Art — Public domain — Kaiser Wilhelm II vetoed the gold medal this work had been awarded. He disapproved of socialist sympathies.

The instanton: crossing the separatrix

The real publishing process is not deterministic. There is noise: a paper lands with an unusually sympathetic editor, a referee happens to share the author's framework, a lucky citation cascade. In the stochastic version of the system, the most probable noise-driven escape route between the two basins is the instanton — the time-reversed trajectory that climbs uphill against the gradient and crosses the separatrix at $p^*$. I worked out the Onsager–Machlup path-integral for this system and computed the Kramers escape rates in both directions. The result is what the phase portrait already suggests, but sharper: the potential barrier is not symmetric. The sigmoid shape of $P_{\mathrm{pub}}(p)$ makes the barrier much steeper on the left side of $p^*$ than on the right.

The upshot is that crossing the separatrix upward — out of the expelled basin toward the celebrated one — is exponentially harder than the reverse. The system is ratcheted: it concentrates prestige upward and is structurally reluctant to reverse. A low-prestige researcher reaching the celebrated basin by luck alone is an exponentially rare event; a high-prestige researcher falling back is far less so. The full derivation is in the addendum below for those who want the details.

― ✦ ―

Why double-blind review is not a nicety

Double-blind — and ideally triple-blind — peer review protocols are not procedural niceties. They are the only structural intervention that severs the feedback loop at the point where it does its damage: the desk decision. Remove the author's name and institution from the manuscript, and the prestige proxy vanishes. The editor is left with the one thing the system claims to care about but systematically avoids measuring: the science itself.

In the language of the model: double-blind review drives $w_1 \to 0$. When the prestige weight drops, the sigmoid in Gate 1 flattens, $p^*$ shifts leftward, the celebrated basin expands, and the barrier asymmetry that makes upward crossings so rare begins to dissolve. Low-prestige researchers gain a fighting chance not because the noise increases, but because the potential landscape becomes less hostile.

Who has the leverage

The leverage here does not sit with researchers at small colleges or unfashionable addresses. We can decline to referee, we can write letters, we can name the asymmetry in public — and we should. But the people who can actually change the architecture are the ones the architecture has rewarded.

If you are at a prestigious institution, if your name opens doors, if journals court you as a referee and a member of their editorial boards — you have something the rest of us do not: the ability to make demands that will be heard. Refuse to referee for journals that do not operate double-blind review. Raise it in editorial meetings. Make it a condition of your participation. The system runs on your labor and your legitimacy. Withdrawing either, with a stated reason, costs the journal something real.

Good science is not correlated with the postcode of the institution that produced it.

The loop that makes it appear so is a man-made artifact, and it can be dismantled — but only by the people with enough prestige to make dismantling it credible.

The mathematics is unambiguous. The basin boundary is where it is because of how the gates are set. Change the gates, move the boundary. The celebrated pole is not reserved by nature. It was reserved by policy, and policy can change.

Usually abused in comparable ways.

Addendum: the stochastic calculation

For completeness, here is the path-integral argument that underlies the asymmetry claim in the main text.

The stochastic version of the system is governed by the Onsager–Machlup path integral, whose Lagrangian is:

$$ \begin{equation} \mathcal{L}_{\mathrm{OM}}(p,\dot{p}) \;=\; \frac{1}{2\sigma^2}\bigl(\dot{p} - f(p)\bigr)^2 \end{equation} $$

The action $S[p] = \int \mathcal{L}_{\mathrm{OM}}\,\mathrm{d}t$ is minimized by the deterministic trajectory $\dot{p} = f(p)$, which costs zero action. The Euler–Lagrange equation also admits a second classical solution:

$$ \begin{equation} \dot{p} \;=\; -f(p) \qquad \text{[the instanton]} \end{equation} $$

This is the time-reversed trajectory — the path that climbs uphill against the gradient, crossing the separatrix at $p^*$ where $f(p^*)=0$. It is the most probable noise-driven escape route between the two basins.

The Noether current

The Lagrangian $\mathcal{L}_{\mathrm{OM}}$ has no explicit time dependence. By Noether’s theorem, time-translation invariance yields a conserved quantity — the Hamiltonian of the system:

$$ \begin{equation} \mathcal{H} \;=\; \frac{1}{2\sigma^2}\bigl(\dot{p}^2 - f(p)^2\bigr) \end{equation} $$

This is the conserved Noether current. On the zero-energy manifold $\mathcal{H}=0$, we recover $\dot{p} = \pm f(p)$ — the forward flow and the instanton. Both are zero-energy trajectories. The separatrix $p^*$ is the turning point of the instanton: since $f(p^*)=0$, the trajectory momentarily halts there before committing to the celebrated basin. There is also a deeper symmetry: $\mathcal{L}_{\mathrm{OM}}$ is invariant under $t\to -t,\; f\to -f$ simultaneously — the detailed balance symmetry of the underlying stochastic process, whose Noether current encodes the entropy production rate.

The escape rates

Substituting the instanton $\dot{p}=-f(p)$ into $\mathcal{L}_{\mathrm{OM}}$ and changing variable $\mathrm{d}t = \mathrm{d}p/\dot{p}$, the instanton action from starting position $p_0$ to the saddle $p^*$ is:

$$ \begin{equation} S_{\mathrm{inst}}(p_0 \to p^*) \;=\; \frac{2}{\sigma^2}\int_{p_0}^{p^*}\!|f(p)|\,\mathrm{d}p \;=\; \frac{2\,\Delta V}{\sigma^2} \end{equation} $$

where $\Delta V = V(p^*)-V(p_0)$ is the potential barrier height and $V(p) = -\int_0^p f(p')\,\mathrm{d}p'$ is the effective potential. The Kramers escape rates for the two crossing directions are:

$$ \begin{align} \Gamma_{-\to+} &\;\sim\; \exp\!\left(-\frac{2\,\Delta V_-}{\sigma^2}\right) \qquad \text{[expelled $\to$ celebrated]} \\[4pt] \Gamma_{+\to-} &\;\sim\; \exp\!\left(-\frac{2\,\Delta V_+}{\sigma^2}\right) \qquad \text{[celebrated $\to$ expelled]} \end{align} $$

These two rates are not equal. The sigmoid shape of $P_{\mathrm{pub}}(p)$ means $f(p)$ is steep on the right side of $p^*$ and shallow on the left. Therefore $\Delta V_- \gg \Delta V_+$: the barrier is much higher climbing out of the expelled basin than falling out of the celebrated one. The asymmetry ratio:

$$ \begin{equation} \frac{\Gamma_{-\to+}}{\Gamma_{+\to-}} \;=\; \exp\!\left(-\frac{2(\Delta V_- - \Delta V_+)}{\sigma^2}\right) \;\ll\; 1 \end{equation} $$

is exponentially small. A low-prestige researcher crossing $p^*$ by noise alone is an exponentially rare event. A high-prestige researcher falling back across $p^*$ is exponentially less rare. The system is ratcheted: it concentrates prestige upward and is structurally reluctant to reverse.

The only way to make $\Gamma_{-\to+}/\Gamma_{+\to-} \to 1$ is to make $f(p)$ antisymmetric around $p^*$ — to make the potential barrier equal on both sides. That requires the gates to treat both basins equally. Which is what double-blind review does: it removes the prestige signal from Gate 1, flattens the sigmoid, lowers $\Delta V_-$, and makes the landscape more symmetric.