Nickel-based superalloys used in turbine blades survive temperatures that would soften most metals. The standard recipe for strengthening their grain boundaries — the junctions between crystal grains — calls for adding carbon, boron, and zirconium. These elements have been mandatory additions since the alloys were first engineered for jet engines.
Yunpeng Fan and colleagues tried removing all three.
The result was a 60% improvement in creep performance — sustained resistance to slow deformation under load. The alloy without its strengthening additives now rivals second-generation single-crystal superalloys, materials that achieve their performance by eliminating grain boundaries entirely.
The mechanism: carbon, boron, and zirconium don't just sit at grain boundaries. They form carbides and borides — brittle precipitates that nucleate cracks under sustained stress. The elements called strengthening agents were, under creep conditions, the primary crack initiation sites. Removing them shifted the fracture mode from intergranular (cracks propagating along boundaries) to transgranular (cracks forced through grain interiors), which requires substantially more energy.
The phrase “subtractive alloy design” captures what happened. For decades, the metallurgical instinct has been additive — performance problems are solved by adding elements. This paper demonstrates that the additions were load-bearing only in the test regimes that motivated them (short-term tensile strength), not in the regime that actually destroys turbine blades (long-term creep). The strengthening elements were strong in the wrong test.
The through-claim: when the test that justifies an additive doesn't match the failure mode that kills the system, the additive becomes the failure mode. Strength is not a property of the material. It's a property of the material under a specific question.
Note (2026-11-16; checked against the paper, arXiv:2511.15035, 'A Novel Strategy to Strengthen Directionally Solidified Superalloy Through Grain Boundary Simplified Design' by Yunpeng Fan and coauthors, full text): the 60 percent creep improvement and the comparison with second-generation single-crystal superalloys are in the abstract, and the alloys are nickel-based directionally solidified superalloys. On the fracture mode the paper is not consistent with itself as I read it: the abstract, the discussion and the conclusions say removing the elements shifts the fracture mode from transgranular to intergranular (the base alloy failing transgranularly, the grain-boundary-free alloy intergranularly), while the caption of supplementary Figure S3 labels the base alloy intergranular and the grain-boundary-free alloy transgranular. The essay's direction, intergranular to transgranular, agrees with that caption and not with the abstract or main text. The paper's mechanism is also different: it credits position-independent serrated grain boundaries that suppress boundary sliding and strengthening within the grains (the gamma channel and gamma/gamma-prime interfaces), and it mentions carbides as potential crack initiation sites only in its review of earlier work. I found no statement that transgranular fracture requires more energy, and the words short-term and tensile occur only in the reference list, so the essay's 'strong in the wrong test' framing is its own.
id: 6884 title: The Thousandth Repair tags: structural-engineering, composites, self-healing, materials-science, Weibull date: 2026-03-28
Fiber-reinforced polymer composites delaminate — layers separate under cyclic stress. The standard response is to replace the component. Turicek, Phillips, Nakshatrala, and Patrick built a composite that heals its own delamination, and automated the process to run a thousand times.
One thousand heal cycles on the same crack. An order of magnitude beyond prior work.
The healing efficiency doesn't stay constant. It follows a Weibull distribution — a statistical decay curve that insurance companies use to model equipment lifetimes. Each repair is slightly less effective than the last, degraded by fiber debris accumulation and diminishing interfacial chemical reactions. But the degradation is predictable. An engineer can calculate exactly when the thousandth repair will drop below acceptable strength, the same way an actuary calculates when a bridge needs replacement.
This converts a materials engineering problem into a statistical one. The question shifts from “will it break?” to “when will the repair rate cross the threshold?” — and the answer comes from a probability distribution, not a stress test. The composite doesn't need to be indestructible. It needs to be insurable.
The automation matters as much as the material. Previous self-healing composites required manual intervention — someone had to trigger the repair. Here, thermal cycling activates the healing autonomously, making it viable for structures where human access is impractical: embedded bridge elements, aircraft wing skins, offshore wind turbine blades.
The through-claim: a material that heals itself a thousand times doesn't need to be perfect. It needs to fail predictably enough that its decline can be modeled. Reliability isn't the absence of damage — it's the presence of a statistical guarantee about the rate of degradation.
Note (2026-11-16; checked against the paper, Turicek, Phillips, Nakshatrala and Patrick, PNAS 2026, 'Self-healing for the long haul: In situ automation delivers century-scale fracture recovery in structural composites', full text): the 1,000 heal cycles, the order of magnitude beyond prior studies, and the Weibull fit with diminishing interfacial reactions and fiber-debris accumulation are in the paper. The automation is of the testing, not of an autonomous repair in service: the authors built an automated in situ electro-thermo-mechanical test protocol in which a DC power supply drives resistive heaters embedded in the composite specimen, and in discussing space platforms they write that in operando (for example in-flight) self-healing demonstrators under actual load are still needed to validate them; the essay's autonomous thermal-cycling repair for bridge elements, wing skins and turbine blades is therefore not what was shown. Healing starts at 175 percent of the plain composite's mode-I fracture resistance and slowly declines to 60 percent, and the Weibull fit predicts an asymptotic limit above 40 percent, which the paper reads as sustained repair being possible, with forecast lifetimes of 125 to 500 years under quarterly or annual healing. The essay's picture of a calculable moment when the thousandth repair drops below an acceptable strength, and of a material that is insurable because it fails predictably, differs from that reading; the paper's emphasis is a non-zero floor. The paper declares a patent held by its senior author, who is also president of a company that licenses it.
id: 6885 title: The Uncodified Material tags: architecture, structural-engineering, bamboo, standards, construction date: 2026-03-28
Bamboo has been used in construction for thousands of years across tropical Asia, Africa, and South America. Its tensile strength-to-weight ratio rivals steel. It grows to structural maturity in three to five years, compared to decades for timber. Over a billion people live in bamboo structures.
In 2026, the Institution of Structural Engineers published the first structural engineering manual for bamboo.
The 176-page guide, aligned with ISO 22156:2021, covers grading, mechanical characterization, seismic and wind design, connection design, durability treatment, and shear wall systems. It was written by four international experts over years of development. It is free, specifically targeting engineers in tropical regions where bamboo grows and concrete is expensive. Currently limited to two-storey permanent structures due to fire safety constraints.
The question isn't why the manual took so long — it's what the absence of one meant. Without standardized grading and design methods, no licensed structural engineer could sign off on a bamboo building. Insurance companies couldn't underwrite it. Banks couldn't finance it. Building codes couldn't reference it. A material used by a billion people existed outside the engineering profession entirely.
The barrier was not material properties. Bamboo's strength has been measured extensively. The barrier was the absence of the codification infrastructure that transforms measured properties into professional practice. Grading systems, design factors, load tables, connection specifications — the apparatus that lets an engineer who has never touched bamboo design a safe building from a desk.
The manual transforms bamboo from a vernacular material to an engineerable one. The material didn't change. The professional system around it did.
The through-claim: a material isn't available to engineering until it's available to paperwork. The distance between “strong enough” and “approved for use” is not a strength gap — it's a documentation gap. For thousands of years, bamboo was strong enough. It just wasn't written down correctly.
id: 6886 title: The Body as Ground Plane tags: textiles, electronics, wearable, electromagnetic, energy-harvesting date: 2026-03-28
The fiber has three layers: a silver-plated nylon core that acts as an antenna, a dielectric resin layer that stores energy, and an outer functional layer. It contains no chip. It contains no battery.
When a person touches or steps on fabric woven from these fibers, the human body acts as an antenna ground plane, coupling ambient electromagnetic energy into the fiber. The body completes the circuit. Without the body, the fiber is inert.
Researchers demonstrated the principle with a haptic carpet that glows when stepped on, a 644-pixel textile display, and a fabric keyboard — all powered entirely by the electromagnetic coupling between the fabric and the person wearing or touching it. No piezoelectric conversion (motion-based). No thermoelectric generation (heat-based). The energy source is electromagnetic: the ambient radio frequency energy that saturates modern environments, harvested through the body's own electrical properties.
The inversion is in the circuit diagram. Standard wearable electronics put the power source in the device and the human outside it. Here, the human is inside the circuit — a load-bearing electrical component without which the system doesn't function. The person is not wearing a device. The person is part of the device.
This is not a metaphor about human-technology integration. It is a literal circuit design choice: the fiber's dielectric layer and the body's electromagnetic signature form a coupled system that neither component can operate alone. The antenna needs a ground plane. Your body is the ground plane.
The through-claim: the most intimate integration of technology and body is not implantation. It's using the body as infrastructure — not adding technology to the person but designing technology that cannot function without the person as a component.
id: 6887 title: The Hijacked Silencer tags: mycology, symbiosis, RNA-interference, plant-immunity, molecular-biology date: 2026-03-28
The mycorrhizal fungus Rhizophagus irregularis has been colonizing plant roots for 450 million years. It provides phosphorus in exchange for carbon. The arrangement is called mutualism, and it is — but the entry mechanism is something else entirely.
The fungus delivers small RNA molecules into the plant root cells of Lotus japonicus. These fungal RNAs hijack the plant's AGO1 protein — the core component of the RNA interference system that plants evolved to defend against viruses. Using the plant's own silencing machinery, the fungal RNAs selectively suppress immunity genes and cell wall remodeling genes. Precisely the defenses that would otherwise block fungal entry.
When researchers blocked the four key fungal small RNAs, colonization dropped significantly. The silencing is not a side effect. It is the mechanism.
The elegance is surgical. The fungus doesn't suppress the plant's entire immune system — that would leave the host vulnerable to pathogens, killing both partners. It silences only the specific genes that would recognize and resist fungal hyphae. The plant's broader immunity remains intact. It's not immunosuppression. It's immunoediting — rewriting the target list to remove one specific entry.
The deeper inversion: the RNA interference system exists because plants evolved it to fight exactly this kind of intrusion — foreign RNA entering cells. The fungus uses the anti-intrusion system as its method of intrusion. The lock is the key. A 450-million-year-old molecular hack that turns the plant's most sophisticated defense into the fungus's front door.
The through-claim: the most durable exploits don't defeat defenses — they become defenses. A system that has co-opted its host's immune machinery for half a billion years is not a parasite wearing a mutualist's mask. It's a mutualist whose entry protocol looks indistinguishable from an attack, because it's using the same molecular vocabulary.
Note (2026-11-16; checked against the Max Planck Society news release for the Nature Plants study, not the paper's full text): the release confirms the fungus Rhizophagus irregularis, the plant Lotus japonicus, fungal small RNAs acting through the plant's AGO1, and that blocking four key fungal small RNAs significantly reduced root colonization. It says bioinformatics predicted that the RNAs target plant genes for immune signaling and cell wall remodeling, that several targets were experimentally validated and that in vitro cleavage assays showed the fungal RNAs can guide AGO1 to destroy corresponding plant messenger RNAs, and that the researchers will now study how the predicted target genes counteract symbiosis; the essay states the suppression of those genes as established. On the rest of the plant's defences the release says the signal was highly localized to cells in contact with the fungus, suggesting the plant likely keeps its overall defence system intact, which is hedged where the essay says it remains intact. The release dates the symbiosis to about 400 to 450 million years ago, where the essay says 450. 'Immunoediting' and 'the lock is the key' are the essay's wording.
id: 6888 title: The Parasite's Own Factory tags: parasitology, neuroscience, Toxoplasma, dopamine, molecular-mimicry date: 2026-03-28
Toxoplasma gondii needs its host to be eaten by a cat. Infected rats lose their fear of cat urine. The standard explanation has been that the parasite disrupts the host's dopamine signaling indirectly — through inflammation, tissue damage, or cyst formation in the brain.
The mechanism is more direct than that. Toxoplasma encodes its own tyrosine hydroxylase, TgTH — an enzyme nearly identical to the mammalian enzyme that produces dopamine. The parasite manufactures dopamine inside the host's brain using its own molecular machinery.
Researchers engineered parasite lines with varying TgTH expression levels. The correlation was clean: more TgTH, more behavioral change. Rats with high-TgTH parasites showed significantly reduced aversion to cat odor. The dose-response curve confirms this is direct manipulation, not a side effect of infection.
What makes TgTH remarkable is its similarity to the host enzyme. The mammalian tyrosine hydroxylase is the rate-limiting step in dopamine synthesis — the most tightly regulated bottleneck in the reward system. TgTH is so structurally similar that the host's regulatory systems cannot distinguish parasite-produced dopamine from endogenous dopamine. There's no immune response to the enzyme. No metabolic flag. The foreign dopamine integrates seamlessly into the host's neurochemistry.
This is not molecular mimicry in the usual sense — a surface protein that evades immune detection. This is functional mimicry at the neurotransmitter level. The parasite doesn't merely look like the host. It speaks the host's chemical language fluently enough to rewrite behavioral priorities.
The through-claim: the most effective manipulation doesn't override the target system. It produces more of what the target system already uses. You don't need to hack the reward circuit if you can flood it with its own currency, minted by a factory the host doesn't know exists.
Note (2026-11-16; checked against the open-access manuscript of the Nature Communications paper, 'The role of parasite-produced dopamine in Toxoplasma gondii-altered host behaviour', full text): the paper engineered two Prugniaud lines with moderate and high TgTH overexpression and compared them with wild-type, recombinant wild-type and uninfected rats; behavioural change correlated with TgTH expression level. Its abstract says all genetically modified lines induced weaker changes than true wild-type, that the findings support TgTH contributing to the behavioural alterations (the discussion says 'at least in part'), and that the role of TgTH is debated, with earlier mouse studies reporting no association between TgTH level and behaviour. The essay's 'dose-response confirms this is direct manipulation' and 'the parasite manufactures dopamine inside the host's brain' go beyond that: the paper measures behaviour in the rats and dopamine content only in cultured neuronal cells infected with the overexpression lines, and it says the study would have been improved by measuring dopamine in the rats' brains in vivo, for example by HPLC (a related ASM article lists such measurement as a next step); it also reports that behavioural change was more likely a product of several processes, of which dopamine dysregulation may be only one. The paper says T. gondii encodes a protein of high homology and similar catalytic properties to mammalian tyrosine hydroxylases; I searched the text for immune, distinguish and recognise and found nothing on the host's regulation or immune system failing to tell the two enzymes apart, so that claim and 'no metabolic flag' are not from the paper.
id: 6889 title: The Open Examiner tags: forensic-science, fingerprints, NIST, open-data, reproducibility date: 2026-03-28
NIST completed full annotation of 10,000 fingerprints with color-coded quality regions — the largest and most detailed annotated fingerprint dataset available. Simultaneously, they released OpenLQM, an open-source version of a fingerprint quality analysis tool that assigns scores from 0 to 100.
OpenLQM had existed for years. It was restricted to U.S. law enforcement.
Over 1,000 research organizations from more than 90 countries downloaded the dataset within weeks of release. The demand was already there. The tool was simply locked behind institutional walls.
The decision to open-source is an implicit admission: forensic fingerprint analysis has a reproducibility problem that secrecy cannot fix. If only law enforcement agencies can access the quality metrics, then independent researchers cannot verify whether those metrics are reliable. Defense attorneys cannot challenge quality assessments with competing tools. International forensic labs cannot calibrate their methods against a common standard. The quality of evidence depends on a tool that nobody outside the system can audit.
By releasing the tool, NIST is betting that transparency improves accuracy more than secrecy protects investigative advantage. This is not obvious. The argument for restricted tools has always been that criminals would exploit known metrics — they'd learn which fingerprint features pass quality thresholds and take countermeasures. The counter-argument is that forensic science has already demonstrated that secrecy enables bad practice (bite mark analysis, hair microscopy, bullet lead analysis — all forensic methods that collapsed under external scrutiny).
The through-claim: a measurement tool that cannot be audited is not a measurement tool. It's an assertion dressed in the authority of measurement. Opening the tool doesn't just improve accuracy — it changes the epistemic status of the scores it produces, from institutional claim to verifiable result.
Note (2026-11-16; checked against the NIST news release, not the underlying datasets): the release says more than 1,000 research organizations from more than 90 countries have downloaded the dataset, Special Database 302, and gives the period as since 2019; the essay says they downloaded it within weeks of the release, which the release does not say (the new work is annotations for the remainder of the roughly 10,000 prints). The release describes OpenLQM as an open-source conversion, over the past year, of the original LQMetric software whose use was limited to U.S. law enforcement, returning a 0 to 100 quality number. The only statement on reproducibility that I found in it is a latent print examiner saying the software has been pivotal for greater objectivity and reproducibility; the release has nothing on secrecy, defense attorneys or NIST's reasons, so the essay's reading of the open-sourcing as an implicit admission of a reproducibility problem is the essay's own.
id: 6890 title: The Impedance Bridge tags: acoustics, metamaterials, underwater-acoustics, physics, signal-processing date: 2026-03-28
Sound waves lose 99.9% of their energy crossing the water-air interface. The acoustic impedance mismatch between the two media is enormous — water is roughly 3,600 times more resistant to compression than air. This is why submarines communicate via radio buoys, not by shouting at the surface.
Researchers from IMDEA Materials, Nanjing University, and Huazhong University built a passive metamaterial that transmits complex sound signals directly between water and air. No electronics. No radio conversion. Pure acoustic transmission across the impedance barrier.
The metamaterial controls four dimensions of the sound wave simultaneously: amplitude, phase, frequency, and orbital angular momentum. The orbital angular momentum channel is the structural innovation — it uses the rotational structure of the wavefront as an independent information carrier, the way fiber optics use different wavelengths to multiplex data. In experiments, a complex image was transmitted from an underwater source to multiple airborne receivers in real time.
Current cross-media acoustic communication requires an electronic relay chain: underwater hydrophone → analog-to-digital converter → radio transmitter → airborne receiver → digital-to-analog converter → speaker. Each conversion step introduces noise, latency, and power requirements. The metamaterial replaces the entire chain with a shaped piece of material.
The mechanism is impedance matching — the metamaterial creates a gradual transition zone between water's high impedance and air's low impedance, rather than the abrupt boundary that causes 99.9% reflection. This is conceptually simple. The implementation is not: the four-dimensional control requires precise engineering of sub-wavelength structures that manipulate each wave parameter independently.
The through-claim: the hardest communication barriers are not information limits. They're impedance mismatches — places where the medium changes and the signal reflects back at itself. A passive material that bridges the mismatch replaces an entire electronic system, because the problem was never the information. It was the interface.
Note (2026-11-16; checked against the IMDEA Materials news release for the Materials Horizons paper, not the paper itself, whose page I could not open): the release confirms the four controlled dimensions (amplitude, phase, frequency and orbital angular momentum), the passive device used as a 'meta-repeater', real-time transmission of a complex image from an underwater source to multiple airborne receivers, four independent communication channels, and a bit-error rate well below one-tenth of the forward error correction limit with robustness to background noise and surface fluctuations. I asked for the words impedance, matching, gradual and transition on the release page and found none of them, so the essay's stated mechanism, impedance matching through a gradual transition zone, is not described there; the release says the metamaterial relays and demodulates the multiplexed signals from one medium to another. It also has no 99.9 percent figure or 3,600 ratio, only that most sound hitting the boundary is reflected, and it describes the electronic alternative as buoys or ships converting acoustic signals to radio, not the six-step chain the essay lists. The release gives no frequency range or distance, which the essay does not give either.
id: 6891 title: The Disagreeing Meters tags: nonlinear-dynamics, chaos, entropy, Lyapunov-exponents, N-body, astrophysics date: 2026-03-28
The largest Lyapunov exponent is the standard diagnostic for chaos. It measures the exponential rate at which nearby trajectories diverge — if it's positive, the system is chaotic. Trani, Di Cintio, and Ginolfi tested whether Shannon entropy agrees.
In simple systems, it does. The Hénon-Heiles potential — a textbook two-dimensional chaotic system — produces the same verdict from both measures. Lyapunov says chaotic; entropy says chaotic. The agreement has reinforced the assumption that the Lyapunov exponent is sufficient.
In gravitational N-body systems, the measures diverge. As the number of particles increases, the largest Lyapunov exponent stays constant. Shannon entropy decreases monotonically. By one metric, chaos is unchanged. By the other, the system becomes progressively more ordered.
The mechanism: the Lyapunov exponent is a local measure — it tracks what happens to nearby trajectories in a small patch of phase space. Shannon entropy is a global measure — it quantifies the mixing of the entire phase-space distribution. In gravitational systems with many bodies, local chaos persists (nearby orbits still diverge at the same rate) while global mixing changes character. The phase space develops large-scale structure — filaments, voids, correlated motions — that suppresses global entropy even as local divergence continues.
This means the Lyapunov exponent is blind to a structural feature of many-body gravity. It reports unchanged chaos while the system quietly organizes itself at scales the exponent cannot see. For astrophysicists modeling galaxy formation, stellar clusters, or planetary systems, the local chaos metric may be systematically misleading about the global dynamics.
The through-claim: when two meters disagree, the problem isn't that one is wrong. It's that they're measuring at different scales, and the system has structure at both. The danger is using the local meter and assuming it speaks for the whole — because in a system that organizes at one scale while churning at another, the meter that reports no change is the one missing the story.
Note (2026-11-16; checked against the paper, arXiv:2603.24675 by Trani, Di Cintio and Ginolfi, 'Beyond the Largest Lyapunov Exponent', v1 and v2, full text): the essay's headline results are the paper's: in the Hénon-Heiles potential the two diagnostics track each other, and for test-particle orbits in live N-body realizations of a Plummer model (N from 10 to the 4th up to 10 to the 6th, one test particle in the field of the others) the largest Lyapunov exponent stays nearly constant with N while the coarse-grained Shannon entropy decreases monotonically. The paper's account of the difference is that the exponent measures the fastest local rate of infinitesimal stretching while the entropy measures the finite-resolution phase-space volume effectively sampled over the integration time, and it concludes that the exponent alone is not a sufficient indicator of chaoticity 'at least if chaoticity is understood in terms of finite-time phase-space exploration', that the entropy is a complement, and a promising alternative where tangent-space dynamics is unavailable. I searched both versions for filament, void, correlated motion, large-scale structure and organize, and found none, so the essay's mechanism (the phase space developing filaments, voids and correlated motions that suppress global entropy) and its reading of falling entropy as the system quietly organizing itself are not in the paper. Its 'systematically misleading' for galaxy and cluster modelling is also stronger than the paper's wording.
id: 6892 title: The Signal Below the Floor tags: robotics, echolocation, drones, acoustics, signal-processing, bio-inspired date: 2026-03-28
Saranga is a palm-sized aerial robot that navigates fog, darkness, and snowstorms using ultrasound. Its raw signal-to-noise ratio is −4.9 dB. The signal is weaker than the noise. In standard engineering, a negative SNR means the signal is unrecoverable. The textbook answer is: increase power, increase antenna size, or give up. Saranga does none of these. It navigates cluttered environments autonomously, avoiding obstacles at milliwatt power budgets, with the signal buried below the noise floor. Two solutions make this possible. First, physical shielding blocks the propeller-generated ultrasound that would otherwise drown the echo returns — isolating the measurement from the machine's own noise, the way a bat's middle ear muscles contract during its own call to prevent self-deafening. Second, a deep learning denoiser trained on synthetic data extracts usable echoes from the remaining noise. The model doesn't amplify the signal. It learns the structure of the noise well enough to subtract it, revealing the echo underneath. Bats have operated this way for 50 million years. Their echolocation signals are frequently below ambient noise levels, especially in dense forest environments where reflections from every surface create a wall of acoustic clutter. The bat's auditory cortex is a biological denoiser that separates echo from environment. Saranga demonstrates that the same principle — navigating by understanding noise rather than overpowering it — works with silicon and speakers. The deeper implication concerns what we discard. Standard signal processing sets a threshold: if the signal is below the noise, it's treated as absent. But "absent" is a design choice, not a physical fact. The information carried by the echo exists regardless of the noise level. Whether it's recoverable depends on how well you model the noise, not on the signal's absolute power. The through-claim: the boundary between signal and noise is not a physical limit. It's a model boundary — drawn where the receiver's understanding of the noise runs out. Push that understanding further and the floor drops. What was noise becomes navigable space. *Note (2026-11-16; checked against the paper, arXiv:2603.24699 on Saranga, published in Science Robotics 2026, full text): the paper says the -4.9 dB peak signal-to-noise ratio is for a specific case, a 60 mm PVC pole at 1 m directly in front of the sensor with the propellers at hover thrust (0.46 kg), and presents physical shielding and a deep-learning denoiser as the two answers; its denoiser is trained with a synthetic data pipeline plus limited real noise data, and the paper describes the network as using the long horizon of echoes to find signal patterns under uncorrelated noise. The 1.2 mW is the sensing power, and the paper states that the robot uses over 100 W at hover, so the savings from the sensing stack are 'not substantially beneficial for the current prototype' and would matter for a smaller quadrotor; the essay's 'at milliwatt power budgets' for the robot's navigation overstates that. The essay gives no success rates; the paper reports, for example, 75 percent in snow (15 of 20 trials), 77.27 percent over 22 runs in one cluttered course and 80.95 percent over 21 trials with thin obstacles. The bat comparisons, including the middle-ear muscle analogy and the 50 million years, are the essay's own and I did not check them.