Skip to content
Science Journaling Club

Volume 2, Issue 1September 2025

Fall 2025

Two changes at planetary scale that almost nobody noticed, next to a sea slug that keeps stolen chloroplasts running for months, a 99-million-year-old wasp with a trap no living insect inherited, and the phase transition hiding in a plate of cacio e pepe.

This issueSJC 2.1
Published
September 2025
Articles
8
Research
3
Field notes
5
With code
5

Contents

8 articles, in the order the club meant them to be read.

  1. ResearchOriginal Research

    Why Sex Exists: Watching Mutations Pile Up in a Population That Cannot Recombine

    Mutation only adds. In a population with no recombination, once the last individual carrying the fewest mutations fails to leave descendants, nothing can rebuild that class, and the minimum mutation load in the population steps up by one forever. Muller called it a ratchet. We measured how fast it turns. The instrument was a Wright-Fisher simulation the club wrote and ran on a laptop, carrying the full distribution of mutation counts per individual; no organism was involved at any stage, because here the computation is the experiment. Across a 60-cell grid (N=100N = 100 to 20002000, U=0.1U = 0.1 to 0.80.8, s=0.02s = 0.02 to 0.10.1) plus 35 further cells, at 200 replicate populations each, we logged 11,220,000 replicate-generations in the main grid alone. The null passed exactly: with U=0U = 0 the ratchet clicked zero times in 1,200,000 replicate-generations and mean fitness stayed at 1.000000000000000. Haigh's deterministic equilibrium survives where it is defined, the mean load matching λ=U/s\lambda = U/s to +2.9% and the size of the unloaded class matching Ne−λN e^{-\lambda} to −4.7% ± 1.1%. The rate prediction built on it does not survive. Across 70 clicking cells the measured rate was a median 0.398× the classical e−n0e^{-n_0}, and in the worst cell it was 1.4×1055 times larger, sitting 11 standard errors above a prediction of 1.7×10−591.7\times10^{-59}. Three independently written implementations agree on our number to within 1.9 standard errors, so the fault is in the formula. The missing quantity is the selection coefficient: at fixed n0=5n_0 = 5 raising ss from 0.02 to 0.10 nearly doubled the click rate, a gap of 28.5 standard errors, and rescaling by s×n0s \times n_0 collapsed cells that differ five-fold in both onto one curve within 6.9%, against 78.4% disagreement when matched on n0n_0 alone. Mean fitness fell exactly as bookkeeping demands, the measured dln⁡Wˉ/dtd\ln \bar{W}/dt matching (click rate) ×ln⁡(1−s)\times \ln(1-s) to 0.9961 ± 0.0031 over six regimes. Adding recombination to an explicit 1024-site genome stalled the ratchet: 2.7% of offspring made sexually halved it, and 15.2%, roughly one offspring in seven, cut it to a hundredth of the asexual control. Seed 20250921.

    Fall 2025Read 34 minSections 13

  2. ResearchOriginal Research

    The Dosing Schedule Decides Whether Resistance Wins

    We built a stochastic two-strain bacterial population model under a one-compartment pharmacokinetic drug concentration and ran 178,002 simulated treatment courses on a laptop. There are no bacteria in this study and no patients: the computation is the experiment, and none of what follows is medical advice. Before measuring anything we validated the deterministic core. Bisection on the net growth rates recovered the mutant selection window as [1.000000000000, 16.000000000000] mg/L against the input MICs, agreeing to 2.9 × 10−15 mg/L; the lower edge of the wider selective window matched a closed form we solved by hand to 2.7 × 10−15 mg/L, landing at MIC/5.26, inside the MIC/4 to MIC/230 range Gullberg and colleagues measured experimentally [8]. With the fitness cost set to zero the resistant strain never declined without drug (minimum net growth rate exactly 0.000000/h); with mutation switched off, resistance appeared in 0 of 2000 replicates, and a schedule that cures only 8.90% of the time once mutation is switched back on cured every one of them. Across a grid of five dose sizes and five dosing intervals at perfect adherence, peak concentration relative to MIC was the best single predictor of cure (Spearman ρ = +0.9809) and total drug delivered was worse (+0.9333). At matched total exposure of 1246.5 mg/L·h, a two-day schedule of 48 mg/L every 8 h cleared 2000 of 2000 replicates while a fourteen-day schedule of 6.857 mg/L delivering exactly the same drug ended in resistance 91.10% ± 0.64% of the time, a gap of 143 standard errors. Holding dose fixed instead, one day of treatment cured 86.70% ± 0.76% against 100% for fourteen days, and the median time to clearance was 6.0 h in every row. Missing doses mattered, taking cure from 1.0000 to 0.7670 ± 0.0094 at 50% adherence, but a control that delivered the same reduced total drug with no gaps did considerably worse (0.4980 against 0.8595 at 80%). The short schedule won in all fourteen sensitivity variants, with the margin collapsing from +0.880 to +0.022 at a hundred-fold smaller inoculum. Seed 20250915.

    Fall 2025Read 35 minSections 12

  3. ResearchReplication

    Running Axelrod's Tournament Again, With Noise This Time

    Axelrod's 1980 round robin found that tit for tat beat every program sent to it. We asked whether that survives when moves are occasionally misread. No person and no organism was involved at any point: the computation is the experiment. We implemented fifteen classical strategies, ran a full round robin of 400 repetitions at each of ten noise levels from 0 to 0.20, and then ran replicator dynamics for 4,000 generations on each measured payoff matrix. The main sweep is 480,000 matches and 96 million rounds. Fifty analytic pairwise payoffs at zero noise were reproduced exactly, the payoff structure satisfies T>R>P>ST > R > P > S and 2R>T+S2R > T+S, and the replicator update conserved total frequency to 5.55×10−165.55 \times 10^{-16} over every generation of every run. At zero noise tit for tat scored 2.62890 and finished third: tied with contrite tit for tat (paired t=0.47t = 0.47) and 3.04 standard errors behind generous tit for tat at 2.63999. A leave-one-out sweep shows the entire margin comes from one opponent, Joss. Under noise the board turns over twice. Generous tit for tat leads to 2% noise, contrite tit for tat from 5% to 15%, and at 20% noise always defect wins with 2.25965. At 5% noise tit for tat trails the leader by 0.07785 per round, a paired tt of 24.1. The evolutionary run is cleaner: contrite tit for tat takes the entire population at every noise level from 0.005 upward, while the mean payoff of that all-cooperative population falls from 3.0000 to 2.6421, a loss of 11.9% caused by errors alone. An exact Markov chain for two tit-for-tats agrees with the simulation to within 1.84 standard errors at every noise level. Seed 20250913.

    Fall 2025Read 31 minSections 12

  4. Field noteExplainer

    Snowball Earth Was Flickering: A Planet Stuck in a 56 Million Year Loop

    The Sturtian glaciation opened around 717 million years ago and closed around 661 million years ago, and the 56 million years of ice lying between those two dates have embarrassed every account of the Cryogenian written since anyone first learned to measure the interval at all [5, 6]. A Harvard group published a different reading of it on 27 April 2026 [1]. In their coupled model of Neoproterozoic climate and carbon, the rapid weathering of the newly erupted Franklin basalt province destroys every stable climate the planet has available to it, so that Earth never rests anywhere but circles instead, frozen and thawed and frozen again, turn after turn, across the whole 56 million years. We rebuilt the mechanism in a deliberately small model of our own, two equations rather than a research code, because we wanted to watch the thing work. Our version puts the freeze threshold at 257 ppmv of carbon dioxide and the thaw threshold at 30,328 ppmv, a factor of 118 apart, and finds that any weatherability above 3.59 times the modern value leaves the planet nowhere stable to sit. Out of that comes a cycle with a period of 1.31 million years and a peak-to-trough swing of 68.8 K, spending 94% of each turn frozen: roughly 43 freeze-and-thaw cycles inside the Sturtian. The digits belong to us rather than to the paper, and the mechanism matters far more than the digits do.

    PaleoclimateRead 33 minSections 12

  5. Field noteData Story

    Twenty One Percent of the Ocean Got Darker and Almost Nobody Noticed

    Two researchers at Plymouth took twenty years of satellite measurements of how fast light dies in seawater, one number per pixel per year on a 9 km grid, and asked a question nobody had asked globally: is the ocean getting dimmer? Over 2003 to 2022 the attenuation coefficient rose across 75,341,181 km², which is 21 percent of the ocean, while it fell across only 10 percent. The lit layer shoaled by more than 50 metres over 32,449,129 km² and by more than 10 metres over 19 percent of the sea [1]. We rebuild the underlying optics from the Beer-Lambert equation, work out why clear water loses hundreds of times more depth per unit of added murk than muddy water does (a ratio of about 2,010 to 1 between the clearest gyre and a river plume), build a Secchi disk lookup table anyone with a rope can use, and then add up the lit water that disappeared: 3.75 million km³, or about one Mediterranean Sea, roughly 0.28 percent of all the water in the ocean. Our volume figure is a bookkeeping exercise built on the paper's reported areas, not a measurement, and the section on objections explains why twenty years may still be too few.

    OceanographyRead 27 minSections 12

  6. Field notePaper Analysis

    The Slug That Eats Its Own Solar Panels

    Some sea slugs eat algae, digest everything except the chloroplasts, and keep those chloroplasts running inside their own cells — photosynthesizing, on borrowed hardware, for nearly a year, a state biologists call kleptoplasty. Nobody knew how an animal cell kept a plant organelle alive that long without the plant's own nucleus to keep supplying it with parts. In June 2025 a team led by Corey Allard and Nicholas Bellono answered it: the slug Elysia crispata builds a dedicated compartment, wraps one stolen chloroplast in it at a time, and marks the membrane with its own late-endosome machinery — proteins named Rab7, VHA, P2X4 and NPC2, all host-encoded, none algal [1]. They named the compartment the kleptosome. An ATP-gated channel in its membrane, P2X4, tunes the interior to keep photosynthesis running; block that one channel with the antagonist 5-BDBD and net oxygen output falls 62% (p = 0.0250) [1]. Then comes the part that reframes the whole system: starve the slug, and it digests the chloroplasts stored inside its own kleptosomes for calories. Elysia crispata survives close to four months without food this way, next to three to four weeks for a non-photosynthetic relative, the sea hare Aplysia californica, run through the identical experiment [1]. A structure built to keep a solar panel switched on turns out to double as the battery it drains when the lights would otherwise go out. The same authors found a convergent version of the trick in coral and anemone tissue, which arrest and digest their own symbiotic algae by a similar route [1]. This field note is our own read of a paper we did not write and an organism we have never touched; we did not run a lab, and every number below traces back to Allard et al. or to the sacoglossan literature they cite.

    Cell BiologyRead 18 minSections 13

  7. Field notePaper Analysis

    The Trap That Evolution Tried Exactly Once

    In March 2025 a team led by Qiong Wu and Lars Vilhelmsen described a wasp trapped in 99-million-year-old Burmese amber that does something no living wasp does, and, as far as anyone can tell, no fossil wasp ever did again [1]. Sixteen females of †Sirenobethylus charybdis carry, at the very tip of the abdomen, a hinged three-flap apparatus — one plate each from a modified sternum 6, tergum 7 and sternum 7 — fringed with long trigger setae and caught, across the sixteen specimens, in a spread of open and closed positions. The authors call it, cautiously, reminiscent of a Venus flytrap, and write that nothing similar is known from any other insect, living or fossil [1]. This note is not about the wasp finding a host. It is about how you prove an abdomen used to grab things when the abdomen has been dead for 99 million years and cannot be watched doing anything. We walk through the comparative anatomy the authors leaned on, the muscle-attachment reasoning built from two unusually large sclerites with a straight line across them, and the process of elimination that ruled out the one alternative hypothesis anyone could think of. We also do a little arithmetic of our own, none of it the paper's: the amber is dated to 98.79 ± 0.62 million years [17], which places this wasp roughly 33 million years before the asteroid that ended the Cretaceous [20] — meaning the entire 66-million-year run of the Cenozoic happened after this structure and nothing living inherited it. A trait does not need to fail to disappear. That is the part of the story worth sitting with.

    PaleoentomologyRead 23 minSections 13

  8. Field notePaper Analysis

    The Phase Transition Hiding in Your Cacio e Pepe

    Cacio e pepe is three ingredients — pecorino, black pepper, pasta water — and it fails constantly, seizing into a grainy, stringy mess that any home cook has produced at least once. In April 2025, eight physicists gave that failure a name and a boundary. Writing in Physics of Fluids, Bartolucci and colleagues showed that the clumped state is a real thermodynamic phase, which they call the Mozzarella Phase, separated from a smooth one by a starch concentration of almost exactly 1% relative to the cheese's mass [1]. Below that line the sauce clumps once it passes roughly 65 °C; above it, starch holds the cheese proteins apart all the way to 95 °C. The transition is a lower critical solution temperature, sitting slightly above 60 °C at a protein mass fraction of φ ≈ 0.134, and the physics runs backward from kitchen folklore: it is heating the mixture too far, not letting it cool, that breaks it. This note works through how eight scientists turned a ruined dinner into 50–95 °C temperature ramps over backlit Petri dishes, and does the club's own arithmetic on the recipe they end up recommending: 5 g of starch against 200 g of cheese is 2.5% by mass, about 1⅞ teaspoons by a standard kitchen conversion, and it puts the actual dish at a protein fraction of φ ≈ 0.154 — close enough to the paper's most clump-prone concentration that the 1% line is not a formality.

    Soft Matter PhysicsRead 22 minSections 13