Original, testable claims seeded by unconventional sources
31 October 2026
This is a hypothesis portfolio, not a literature review and not a set of predictions that I claim are probably true. Each entry takes a real anomaly or capability — often reported in a non-canonical channel — and proposes a new mechanism or a new bridge that explains it, together with a prediction that would discriminate the proposal from the mainstream reading.
I held to five rules:
The brief asked for unconventional sources, so I deliberately sampled channels that are usually excluded from a “state of the field” review: replication failures, self-published preprints, programme reports, conference abstracts, citizen-science catalogues, theses, and contested results.
| Channel | Entries seeded | Why it is unconventional |
|---|---|---|
| Failed / null replication preprints | H01 | The result that did not reproduce is rarely cited |
| Self-published preprint archive (Zenodo) | H05 | Unrefereed, high-variance, usually ignored |
| Conference abstracts and proceedings | H07, H10, H11 | Preliminary; often never becomes a paper |
| Citizen-science catalogues | H08, H12 | Labels produced by volunteers, not a PI |
| National-lab programme reports | H13 | Operational detail, not a journal result |
| Thesis and grey literature | H03, H20 | Unindexed, unreviewed, unusually detailed |
| Contested-result literature | H04, H09, H16 | Claims with active null results |
| Survey-systematics literature | H06, H17 | The corrections, not the headline |
Roughly half the entries are seeded by a channel outside the conventional journal article; the rest use the ordinary literature, because forcing every seed to be exotic would have produced a weaker portfolio.
A source ledger with every link is in sources.csv; the
same records are machine-readable in hypotheses.json.
Each hypothesis has the same nine fields. Claim is the one-sentence proposal. Seed is the anomaly and its channel. Tension is why the seed does not fit the orthodox account. Mechanism is the causal story. New prediction is what the mechanism implies that the orthodox account does not. Decisive test is the cheapest experiment that separates them. Falsifiers are the observations that would kill it. Delta is the explicit novelty claim.
Confidence is graded low (worth one experiment, probably wrong), moderate (survives a first sanity check and has a plausible mechanism), or high (I would bet on the direction, not the magnitude). No entry is above moderate.
Claim. The oxygen apparently produced by polymetallic nodules in the deep sea is not electrolytic O₂ from seawater. It is O₂ released by the catalytic decomposition of hydrogen peroxide that accumulates on manganese-oxide nodule surfaces, and the effect requires intact surface chemistry that slurry handling destroys.
Seed. Nature Geoscience (2024) reported rising O₂ in benthic chambers containing nodules, with sub-volt potentials measured across nodule surfaces, and proposed seawater electrolysis. A 2025 replication preprint reports that O₂ increases in Clarion–Clipperton Zone incubations were not attributable to the nodules (replication preprint). Two results, opposite conclusions, same seafloor.
Tension. Electrolysis of seawater needs roughly 1.23 V; the reported potentials are below that, and the replication removed the effect entirely. Either the original measurement is an artifact, or the active ingredient is not the nodule as an electrode but the nodule as a catalyst.
Mechanism. Radiolysis of pore water in U/Th-rich sediments, and superoxide production by benthic microbes, generate H₂O₂ at low steady-state concentration. Mn(IV)/Mn(III) oxide surfaces are among the best known non-enzymatic H₂O₂ decomposition catalysts (2 H₂O₂ → 2 H₂O + O₂). Nodules therefore act as a distributed catalyst bed, converting a diffusive peroxide flux into an O₂ flux. Slurrying and sieving exposes fresh, reduced surfaces, consumes the adsorbed peroxide, and destroys the microbial mat that produces it — which is exactly the manipulation the replication performed.
New prediction. O₂ evolution should scale with the nodule’s surface Mn(IV)/Mn(III) ratio and with the peroxide flux, not with the applied potential. Adding catalase, or pre-washing nodules in acid, should suppress O₂ release; adding H₂O₂ at environmental concentrations should restore it. Sterilised nodules should still produce O₂ (this is abiotic), unlike a microbial explanation.
Decisive test. Intact-core incubations with optodes, factorially crossed: live vs. sterilised nodules × intact vs. homogenised sediment × catalase vs. no catalase. Measure H₂O₂ amperometrically alongside O₂. Cost: one cruise or a shore-based pressure vessel with fresh nodules; no new instrumentation.
Falsifiers. H₂O₂ below detection while O₂ rises; O₂ unaffected by catalase or acid washing; O₂ production requiring an applied potential above 1.2 V.
Confidence. Low–moderate on the mechanism, moderate on “the replication and the original are both partly right.”
Delta. The seed paper proposed electrolysis; the replication proposed artifact. This entry proposes a third, chemically standard process — catalytic peroxide decomposition — that predicts which handling destroys the effect and which does not. I am not aware of this having been proposed for the nodule result.
Claim. Ball lightning is a charged nanoparticle aerosol held together by its own space charge, and its two most-reported properties — a roughly fixed size and a lifetime of seconds — are set by the charge-relaxation time of humid air rather than by combustion or plasma confinement.
Seed. A 2025 JGR Atmospheres paper estimates the radiation power density of a natural ball-lightning event from its spectrum (spectral estimate). The event is real enough to have a measured spectrum; the mechanism is still open after two centuries.
Tension. Plasma models predict lifetimes far shorter than observed unless energy is continuously supplied; combustion models predict a shrinking, rising fireball, not a drifting sphere. Neither predicts a characteristic size.
Mechanism. A lightning strike on silica-rich soil or wood vaporises material; the vapour condenses into 10–100 nm SiOx/C particles that acquire charge in the discharge. The aggregate is a leaky electret: electrostatic attraction balances particle pressure, and the whole structure discharges through humid air with time constant τ = ε₀/σ_air. Because σ_air rises steeply with relative humidity, τ falls steeply with humidity. Glow comes from surface chemiluminescence of SiOx and recombination, not from a hot plasma.
New prediction. Lifetime should scale approximately as the inverse of air conductivity, i.e. drop sharply (roughly as RH^−2 to RH^−4) as humidity rises, with an upper bound of order ten seconds. Diameter should be weakly dependent on strike energy and strongly dependent on ambient pressure and humidity. The spectrum should show SiOx bands and a water-vapour continuum, with no lines requiring more than about 1 eV of excitation.
Decisive test. Trigger discharges into a cellulose/silica aerosol in a humidity-controlled chamber (10–90% RH) and measure size, lifetime, and spectrum against RH. Existing triggered-lightning facilities can host this.
Falsifiers. Lifetime independent of humidity; spectra showing high-excitation ionic lines; no nanoparticle residue after a natural event.
Confidence. Low. The aerosol-electret family of models exists; the humidity scaling is the new and falsifiable part.
Delta. Prior aerosol models explain composition; this one makes the humidity–lifetime law the discriminating prediction, which turns a rare natural event into a tabletop experiment.
Claim. Reports of earthquake lights and preseismic ionospheric electron-content anomalies share one cause — infrasound and acoustic-gravity waves radiated by slow slip — and the geography of light reports is set by temperature inversions that duct that sound, not by the distribution of piezoelectrically active rock.
Seed. Enomoto’s long-running catalogue of Japanese earthquake lights (research map) and a 2025 proposal that seismic S-waves interacting with cloud layers generate the lights (Scilit record) disagree on the source while agreeing the phenomenon is real.
Tension. Piezoelectric models predict lights above quartz-rich faults; reports cluster in valleys and basin margins instead. Ionospheric anomalies are reported before earthquakes but with inconsistent lead times, which is what one expects from a propagation delay rather than a precursory signal.
Mechanism. Slow slip and its afterslip radiate infrasound (0.1–10 Hz) and acoustic-gravity waves. A low-altitude temperature inversion acts as a waveguide, concentrating acoustic energy in valleys — the same places where lights are reported. The acoustic pulse modulates the D/E layer (through neutral–ion drag) and produces local charge separation at topographic edges, triggering brief corona. Lights and total-electron-content anomalies are then two projections of one acoustic field, separated in time by the travel time.
New prediction. Light reports should cluster where the inversion strength is high, independent of lithology. TEC anomalies should lag the slow-slip onset by approximately the acoustic travel time — about 5 minutes per 100 km for tropospheric ducting, longer for thermospheric paths — and the lag should increase with epicentral distance.
Decisive test. Cross-correlate a citizen light-report catalogue with ERA5 inversion maps, local infrasound arrays, and GNSS TEC time series for 20 well-instrumented earthquakes. The lag–distance slope is the discriminator.
Falsifiers. No lag–distance relation; lights equally frequent above and below inversions; TEC anomalies preceding any plausible acoustic arrival.
Confidence. Low. The individual ingredients are documented; the causal unification is not.
Delta. Existing work asks whether lights are electrical or acoustic. This asks why they appear where they do and predicts a measurable lag slope tying them to the ionospheric signal.
Claim. The disputed phosphine in Venus’s clouds is produced by lightning-driven reduction of phosphate in sulphuric-acid droplets, not by biology and not by a volcanic source — and its spatial pattern should therefore trace lightning rather than volcanism.
Seed. The phosphine controversy has settled into a methodological standoff, reviewed in a 2025 survey that treats the detection as unresolved (review). The signal appears in some reductions and vanishes in others.
Tension. Phosphine is thermodynamically implausible in a strongly oxidising sulphuric-acid cloud deck, which is exactly why it was treated as a biosignature. A production route that is local, episodic, and patchy would explain both the detections and the non-detections without invoking life or a surface source.
Mechanism. Lightning in the cloud deck generates reducing radicals and transient solvated electrons. Phosphate (from volcanic or meteoritic input, dissolved in droplets) is reduced at Fe²⁺/sulphur sites to PH₃, which is stabilised in the anoxic interior of the droplets and released when they evaporate. The reaction is inefficient, so PH₃ appears only in recently struck air parcels.
New prediction. PH₃ patches should coincide with lightning hotspots and with freshly perturbed cloud tops, should be absent in the polar vortex where lightning is rare, and should anti-correlate with SO₂ on small scales (SO₂ is depleted where the reducing chemistry has run). The PH₃/H₂O ratio should be roughly constant across patches.
Decisive test. Compare ngVLA or ALMA PH₃ maps at 266 GHz with Akatsuki lightning maps and SO₂ maps, binned to the same resolution. Existing data may already suffice for a first look.
Falsifiers. PH₃ uniformly distributed; PH₃ correlating positively with SO₂; PH₃ present where no lightning has been observed over a full rotation.
Confidence. Low. The chemistry is a stretch, but it is a testable stretch with a clear spatial signature.
Delta. The debate has been “life vs. volcanism vs. artifact.” This adds a fourth source with a distinctive anti-correlation that neither of the first three predicts.
Claim. The non-gravitational acceleration reported for the interstellar object 3I/ATLAS does not come from isotropic outgassing. It comes from one exposed volatile-rich patch, so the thrust vector is fixed in the body frame and the astrometric residual must be periodic at the nucleus spin period, not smooth.
Seed. Two Zenodo preprints argue that 3I/ATLAS shows photometric and non-gravitational anomalies best explained by structured ablative propulsion (anomaly analysis, ablative-propulsion paper). Mainstream coverage treats the acceleration as ordinary outgassing.
Tension. Ordinary outgassing is roughly symmetric about the sub-solar point and produces a smooth, slowly varying acceleration. A single patch produces thrust whose projection on the solar direction oscillates as the nucleus rotates. The two are distinguishable in the residuals, and nobody has checked.
Mechanism. A volatile-rich patch (impact-exposed ice, or a freshly excavated crater) sublimates only when illuminated. Thrust direction is fixed in the body frame; the nucleus spins; the solar projection therefore varies with spin phase. The object’s centre of mass executes a small periodic wobble superimposed on the smooth radial acceleration.
New prediction. The astrometric residual should show a periodicity at the spin period (and a weaker one at half the period), with amplitude proportional to the patch’s projected area. The mean acceleration vector should lie within about 15° of the spin axis projected on the sky. There should be a phase interval each rotation where thrust drops sharply as the patch crosses the terminator — a “thrust hiccup.” Coma morphology should show a collimated jet offset from the sub-solar point by the spin phase.
Decisive test. Re-fit the astrometry with a body-frame thrust model versus an isotropic model, using a spin period from the lightcurve. The periodic residual is a few-tenths-of-an-arcsecond effect at most, but the data are already public.
Falsifiers. Residuals smooth at all photometric periods; acceleration vector inconsistent with any spin-axis orientation; no jet phase offset.
Confidence. Low on the exotic framing, moderate on “the residual should be periodic if the object is outgassing asymmetrically.” That statement is true of ordinary comets too, which is the point: it converts a fringe claim into a standard cometary test.
Delta. The preprints argue what the acceleration is. This specifies how it must vary in time if it is ablative, which is a check that does not depend on believing the artificial-object interpretation.
Claim. The apparent evolution of dark energy in DESI DR2 is not cosmological. It is produced by a redshift-dependent selection function in the emission-line-galaxy sample, and it should vanish when the sample is selected at fixed rest-frame colour and signal-to-noise.
Seed. DESI’s DR2 results show a preference for evolving w(z), while independent reviews argue the evidence is not robust (critical review, counter-analysis).
Tension. The same data set gives a coherent BAO distance ladder and an incoherent w(z) when combined with supernovae. That pattern — distances fine, equation of state drifting — is a classic signature of a redshift-dependent calibration error rather than new physics.
Mechanism. ELG target selection uses a colour cut near a redshift-dependent detection limit. Dust obscuration and stellar contamination vary with redshift, so the effective sample changes in luminosity and environment. The BAO scale is robust to that change because it is a ratio; the supernova cross-calibration is not, because it depends on the matched Hubble-flow sample.
New prediction. The w(z) drift should be monotonic in the fraction of ELGs near the detection limit, and should disappear in a subsample with fixed rest-frame colour and S/N. A residual step, if any, should be absent in an independent SN Ia sample with different selection (e.g. DES-SN).
Decisive test. Re-run the DR2 pipeline on mass-matched, extinction-corrected, fixed-colour subsamples and compare the w(z) posterior. This is a reanalysis, not a new survey.
Falsifiers. Drift persists identically in the matched subsample; the drift amplitude is uncorrelated with detection fraction.
Confidence. Moderate that selection contributes; low that it explains the whole effect.
Delta. The debate has been “is w(z) real?” This names the observable that would settle it — the detection-fraction correlation — and predicts its shape.
Claim. The H₀ discrepancy is a local outflow bias that converges to the Planck value beyond the edge of the Local Void, and the convergence radius is a fixed physical scale that can be predicted from galaxy counts rather than fitted.
Seed. The tension between ~73 km/s/Mpc (distance ladder) and ~67.4 (CMB) persists; a 2025 conference review collects the competing explanations (review slides).
Tension. Local-void explanations usually fit an arbitrary underdensity profile and an arbitrary radius, which makes them unfalsifiable. If the void has an edge, the bias should switch off there.
Mechanism. Supernovae inside an underdense region expand faster than the cosmic mean; the bias falls as the enclosing overdensity approaches the mean. Beyond the void wall, the residual should decay roughly as 1/r. The direction of the residual should anti-align with the Local Group’s peculiar velocity, because supernovae in the direction of motion are closer and thus biased bright.
New prediction. Binned in radius, H₀ should decline and flatten at the void edge, with a residual amplitude scaling as 1/r beyond it. The dipole of the residuals should be anti-parallel to the CMB dipole direction within about 30°.
Decisive test. Fit H₀ in radial bins from 50 to 400 Mpc with a constrained void model, using the void boundary from galaxy catalogues as a fixed input, and cross-check with TRGB distances. Public data suffice.
Falsifiers. H₀ stays near 73 at 300 Mpc; no radial trend; residual dipole misaligned or absent.
Confidence. Low–moderate. Local-void explanations are common; the fixed edge and the dipole anti-alignment are the testable additions.
Delta. Replaces a fitted underdensity with a predicted convergence radius and a directional signature.
Claim. The 8.5-magnitude outburst found in real time by Kilonova Seekers implies a disk that empties almost completely, which requires a magnetic propeller truncating the inner disk — and that predicts a spin period of minutes and X-ray quiescence, not the soft X-ray rise of a normal dwarf nova.
Seed. Citizen scientists identified GOTO065054+593624 as a dwarf nova with an 8.5-magnitude amplitude in real time (arXiv); the amplitude is far above the 2–5 magnitudes typical of the class.
Tension. Standard disk-instability models cap the amplitude because the disk never fully empties. An 8.5-magnitude swing needs either an unusually large disk or a second gate that holds material back between outbursts.
Mechanism. A magnetic field of a few MG truncates the inner disk at the magnetospheric radius. Matter accumulates in the outer disk while the propeller keeps the inner region empty; when the outer disk reaches the instability threshold, a burst of accretion overwhelms the propeller and the system brightens by nearly an order of magnitude. The gate resets only after the disk drains.
New prediction. A recurrence time of 3–5 days during the active state, a period gap in the outburst interval distribution, coherent optical pulsations at the spin period (minutes), and X-ray emission that is hard and weak in quiescence rather than soft and thermal.
Decisive test. Time-resolved photometry over two weeks plus a 10 ks XMM pointing in quiescence, searching for the spin period.
Falsifiers. Smooth α ≈ 0.1 decay with no spin period; thermal soft X-rays in quiescence; no recurrence within the predicted window.
Confidence. Low. The object may simply be a large, low-viscosity disk.
Delta. The discovery paper characterises the outburst; this proposes the gate and names the spin-period and X-ray signatures that would distinguish it from an unusually large disk.
Claim. The discrepancy between the measured muon anomalous moment and the data-driven Standard Model prediction is not new physics. It is a normalisation error in the e⁺e⁻ → π⁺π⁻ cross section near 0.6–0.9 GeV, caused by unmodelled final-state-radiation interference, and it should show up as a characteristic shape distortion rather than a uniform offset.
Seed. Fermilab’s final g−2 measurement is now systematics-limited (KEK summary, final paper). The theory average remains split between data-driven and lattice determinations.
Tension. The experimental value is stable; the theory is not. A uniform shift in the hadronic vacuum polarisation would move all energy bins equally; the observed KLOE/BaBar discrepancy is localised in the two-pion channel, which is what a radiative-correction modelling error looks like.
Mechanism. The radiative correction that converts the measured inclusive e⁺e⁻ → π⁺π⁻(γ) cross section into the undressed cross section depends on how FSR photons are vetoed. Near the ρ peak, a 1% acceptance error produces a few-percent distortion of the line shape, which propagates into a shift of order 10⁻¹⁰ in a_μ.
New prediction. The discrepancy should be concentrated in the 0.6–0.9 GeV bins and should appear as a dip-then-recovery distortion around 0.78 GeV, not a constant normalisation. A new beam-energy scan with 0.1% normalisation should land between the two experiments, closer to the lattice value.
Decisive test. Reanalyse the published KLOE and BaBar samples with a common FSR model and a common photon-veto definition; compare the resulting a_μ^HVP. Then a short scan at the upgraded VEPP-2000 or BESIII.
Falsifiers. The discrepancy survives a common FSR treatment; the shift is uniform across energy bins.
Confidence. Moderate that the theory error is underestimated; low that this specific mechanism is the cause.
Delta. Reframes “experiment vs. theory” as “two experiments vs. one radiative correction,” and predicts the shape of the distortion.
Claim. The X17 signal is not a new boson. It is an internal-pair-conversion artifact of the ⁸Be and ⁴He nuclear transitions, and its apparent mass should therefore depend on the nuclear environment and the detector’s angular acceptance rather than on the beam energy alone.
Seed. PADME’s 2025 results report no confirmatory signal in positron annihilation (PADME results), while a combined analysis still claims modest evidence (EPJ C).
Tension. A genuine 17 MeV boson would appear in both production channels with a fixed mass. A nuclear-structure artifact would appear only where the nuclear transition provides the right multipolarity and would move with the target.
Mechanism. Internal pair conversion in ⁸Be produces an e⁺e⁻ spectrum with a broad excess near 17 MeV when the final-state interaction between the outgoing nuclei is treated approximately. The apparent peak position depends on the angular acceptance and on the nuclear radius, both of which differ between the ⁸Be and ⁴He experiments. PADME, which uses a different production channel, has no such nuclear environment and sees nothing.
New prediction. Reanalysing the ⁸Be data with a full final-state-interaction calculation should remove the peak without a new particle. Running the same detector with a ¹²C target at the same available energy should shift or abolish the excess. The peak position should correlate with the target’s charge radius.
Decisive test. A reanalysis with an open-source FSI/IPC code, plus one shift of target at an existing spectrometer.
Falsifiers. A target-independent 17 MeV peak with a boson-like angular distribution; a confirmed signal in a purely leptonic production channel.
Confidence. Moderate. PADME’s null result is already suggestive.
Delta. Supplies a specific nuclear explanation with a target-Z prediction, rather than treating the anomaly as either a boson or an error.
Claim. The ²²⁹Th nuclear clock’s transition energy depends not only on the fine-structure constant but on the local gluon condensate, so two clocks at different gravitational potentials should disagree at the 10⁻¹⁶–10⁻¹⁵ level, with an annual modulation as the Earth–Sun distance changes.
Seed. Nuclear clocks are being developed explicitly as probes of physics beyond the Standard Model (DAMOP 2025 abstract); the transition has now been laser-excited.
Tension. Nuclear clocks are usually discussed as α-variation detectors. But the isomer energy is dominated by the strong interaction, and the gluon condensate is the order parameter that couples to it. A clock comparison is sensitive to any coupling that shifts the nuclear Hamiltonian, not just α.
Mechanism. The condensate’s local value depends on the ambient matter density and on the gravitational potential (through the trace anomaly). A fractional coupling κ at the 10⁻³ level relative to gravity would produce a fractional frequency shift of order κ · ΔΦ/c² between two clocks separated by a kilometre, and a seasonal term of order 10⁻¹⁷ from the eccentricity of Earth’s orbit.
New prediction. The frequency ratio of two ²²⁹Th clocks at different altitudes should deviate from the optical-clock prediction by a term proportional to ΔΦ, with the same sign for both clocks. An annual sinusoid should appear in the ratio at the orbital period, in phase with the Earth–Sun distance.
Decisive test. Operate two transportable ²²⁹Th clocks at altitudes differing by ~1 km for one year, with a fibre link, and compare against a Sr lattice clock at the same sites.
Falsifiers. Frequency ratio matches the optical prediction to 10⁻¹⁷; no annual term; no altitude dependence.
Confidence. Low. This is a speculative coupling, but the experiment is already being built for other reasons.
Delta. Moves the nuclear clock from an α-meter to a gluon-condensate meter, with a distinctive altitude-plus-season signature.
Claim. One identifiable subclass of short broadband “blip” glitches in LIGO is caused by scattered light driven by seismic motion of the suspension, so the glitch rate should correlate with ocean-generated microseism at 0.1–0.3 Hz, with a lag of hours and a seasonal cycle.
Seed. Volunteer classifiers in the Gravity Spy project have labelled hundreds of thousands of glitches, and a 2025 study mines that catalogue for new glitch classes (community-science study).
Tension. Blips are usually attributed to instrumental noise of unknown origin and are modelled as stationary random events. If a fraction is environmental, the rate is predictable from seismic data — which would let analysts veto them prospectively instead of modelling them.
Mechanism. Microseism from ocean swell displaces the ground by micrometres at 0.1–0.3 Hz. That motion modulates the alignment of stray-light paths inside the interferometer, producing short broadband transients when the scattered field crosses a fringe. The coupling is nonlinear, so the glitch rate scales with seismic amplitude squared.
New prediction. The blip rate should correlate with microseism amplitude with a lag of 2–12 hours (the time for swell to reach the site), should show the same seasonal cycle as the ocean wave climate, and should scale superlinearly with seismic amplitude.
Decisive test. Join the Gravity Spy labels for O3/O4 with seismic station data at Hanford and Livingston, and compute the cross-correlation and seasonal coherence. Public data, no new observing time.
Falsifiers. No correlation with microseism; rate tracks only anthropogenic activity; correlation present for all glitch classes equally.
Confidence. Moderate that some glitch class is environmental; low that it is specifically blips.
Delta. Uses citizen-science labels as a physical probe of the detector environment rather than as a cleaning tool.
Claim. The 43-second record was limited not by confinement but by the phase of pellet injection relative to the edge magnetohydrodynamic mode it excites. Locking injection to the optimal phase should raise the triple product by 10–20% without increasing heating power.
Seed. W7-X set a sustained triple-product record in 2025 using a new variable-rate pellet injector, with 70 pellets over 30 seconds (PPPL report).
Tension. Pellet fuelling is treated as a rate-and-depth problem: more pellets, deeper penetration. If each pellet also triggers an edge relaxation event, then when it arrives relative to that event matters as much as how many arrive.
Mechanism. A pellet ablates in the edge, launches a local pressure perturbation, and excites a 3D edge mode. If the next pellet arrives during the mode’s recovery phase, the perturbation is absorbed into the mode and fuelling efficiency drops; if it arrives at the mode’s growth phase, the perturbation reinforces the transport barrier and the density profile stays peaked.
New prediction. Triple product should be a periodic function of the pellet–mode phase, with a maximum at a fixed phase and a 10–20% peak-to-trough amplitude. The effect should already be visible in the 43-second discharge as a modulation at the pellet frequency.
Decisive test. Phase-bin the existing 43-second discharge by pellet arrival relative to the edge-mode phase, then run a dedicated scan of injection phase at fixed rate.
Falsifiers. No phase dependence; gain saturates with rate alone; edge-mode amplitude unrelated to fuelling efficiency.
Confidence. Moderate. The ingredients are standard; the phase-locking framing is not.
Delta. Converts fuelling from a scalar (rate) to a phase-locked control problem, and predicts an effect in data already on disk.
Claim. The power available from nighttime radiative cooling is limited primarily by precipitable water vapour, not by the 8–13 µm window’s width, and it should fall exponentially with PWV with a coefficient of about 0.1–0.15 per millimetre.
Seed. A 2025 demonstration reports 350 mW/m² of nighttime power from radiative cooling (record).
Tension. Most device papers optimise emissivity inside the atmospheric window and report a single site’s power. If the binding constraint is the water-vapour continuum, then the same device should underperform by a predictable factor at a humid site, and materials optimisation has a hard ceiling.
Mechanism. Water vapour absorbs across the window edges and its continuum fills the window; the effective sky temperature rises with PWV. The net radiative flux therefore decays roughly exponentially with PWV, and the achievable power approaches a site-independent ceiling only in the driest air.
New prediction. Power should scale as exp(−k·PWV), with k between 0.10 and 0.15 mm⁻¹, so a 3 mm site yields roughly twice the power of a 10 mm site with the same device. The ceiling at PWV → 0 should be below about 1 W/m² for a passive emitter.
Decisive test. Collocate a radiative-cooling device, a GNSS-derived PWV measurement, and an infrared sky radiometer across seasons and at two sites. Regress power on PWV.
Falsifiers. Power independent of PWV; k outside the predicted range; a device exceeding 1 W/m² at low PWV.
Confidence. Moderate on the direction, low on the coefficient.
Delta. Provides a site-normalisation law that makes device reports comparable, and predicts the ceiling rather than another record.
Claim. The capacity fade of carbon-fibre structural batteries is electrochemical: the fibres act as bipolar electrodes and corrode at tow junctions above about 4.1 V, and a 5 nm conformal oxide coating should suppress fade without reducing stiffness.
Seed. A 2025 carbon-fibre structural battery with a high-energy cathode reports promising specific energy but does not isolate the fade mechanism (paper).
Tension. Structural batteries are designed against mechanical failure — matrix cracking, fibre buckling, delamination. But carbon fibres are conductive and in direct contact with the electrolyte, which makes them electrodes. Mechanical models do not predict a voltage threshold.
Mechanism. At potentials above the electrolyte’s oxidation window, the fibre surface oxidises; at tow junctions, small differences in local potential drive galvanic couples that pit the fibre. The pits are stress concentrators, so the mechanical and electrochemical failure modes are coupled — but the initiation is electrochemical.
New prediction. Fade should scale with exposed fibre surface area and with the time spent above 4.1 V, and should be suppressed by an ALD Al₂O₃ coating with no measurable loss of tensile modulus. Post-mortem should show pitting localised at tow junctions, not distributed matrix cracks.
Decisive test. 500-cycle ageing of three variants (bare, ALD-coated, coated-plus-interlayer), with EIS, μCT, and tensile testing after ageing.
Falsifiers. Fade independent of voltage window; pitting absent; coating degrades stiffness.
Confidence. Moderate.
Delta. Moves the failure criterion from mechanical to electrochemical and predicts a voltage threshold and a spatial signature.
Claim. The critical current density at which lithium filaments penetrate a solid electrolyte is set by the ratio of electronic to ionic conductivity, not by the shear modulus, and all published data should collapse onto one curve in that ratio.
Seed. A 2025 study shows that grain-boundary electron distribution, tuned by an interlayer, controls filament formation (PubMed) — evidence for an electronic-leakage mechanism, against the mechanical picture.
Tension. The field’s design rule has been “make the electrolyte stiffer.” That rule fails across materials: some stiff electrolytes filament easily, some soft ones do not. A dimensionless group would explain the failures and replace the rule.
Mechanism. Filament growth requires electrons at the interface to reduce Li⁺. Electronic leakage through grain boundaries supplies them; ionic conductivity sets how fast the depleted region refills. The ratio σ_e/σ_i controls whether a protrusion grows or heals. A universal threshold near 10⁻⁴ is plausible from space-charge considerations.
New prediction. CCD should scale with σ_i/σ_e, and data from sulfides, oxides, and polymers should collapse onto one curve when plotted against that ratio. A 10× reduction in σ_e should raise CCD by roughly 3× (sublinear, because space charge screens the field).
Decisive test. Dope one parent electrolyte (e.g. LLZO with Fe vs. Al) to vary σ_e over two decades at fixed σ_i, measure CCD at fixed temperature and pressure, and test the collapse against literature data.
Falsifiers. CCD tracks shear modulus better than σ_e/σ_i; no collapse across chemistries.
Confidence. Moderate. The electronic-leakage picture is gaining ground.
Delta. Proposes the specific dimensionless group and the sublinear exponent, which makes the collapse a quantitative prediction rather than a slogan.
Claim. The precision of integrated photonic matrix processors is limited by thermo-optic phase drift, not photon shot noise, and the accuracy should plateau near 6 effective bits once temperature is stabilised to about ±0.01 K — with a further 2 bits recoverable only by drift tracking.
Seed. A 2025 integrated photonic processor reports picosecond latency (Nature Communications), and a separate electrically programmable photonic tensor core targets in-memory computing (IEEE).
Tension. Precision claims are usually framed against shot noise, which can be beaten with more optical power. But thermal crosstalk and bias drift produce multiplicative errors that do not average down with power.
Mechanism. Each phase shifter’s index depends on temperature; neighbouring devices heat each other; the resulting phase error is a slow random walk. For a mesh of N interferometers, the matrix-vector error grows as √N times the per-device phase error, which caps the usable precision regardless of laser power.
New prediction. Accuracy versus temperature stability should show a knee: improving stability below ±0.01 K yields no gain, and the plateau should be near 6 bits for a 64×64 mesh. A drift-tracking calibration that updates weights every millisecond should recover about 2 bits.
Decisive test. Run the same matrix-vector multiplication on one chip at ±1 K, ±0.1 K, and ±0.01 K, and with drift tracking on and off.
Falsifiers. Accuracy keeps improving with power at fixed temperature; no knee; drift tracking gives no gain.
Confidence. Moderate.
Delta. Identifies the error as multiplicative and temperature-driven, which changes the engineering target from power to thermal architecture.
Claim. Errors in DNA data storage are dominated by synthesis deletions at predicted secondary structures, not by polymerase or sequencing errors, and a design rule that removes strong hairpins should cut the error rate by an order of magnitude at the same coverage.
Seed. A 2025 demonstration stores linked data in DNA origami nanostructures (Nature Communications).
Tension. Storage papers usually treat errors as a coding problem and add redundancy. If the errors are physically localised at hairpins, redundancy is the wrong lever: the same sequence will fail again, and the redundancy will fail with it.
Mechanism. During phosphoramidite synthesis, a self-complementary stretch can fold and block coupling, producing a deletion at a predictable position. Origami design adds long staples and crossovers, which create additional hairpin opportunities. The error is therefore a function of predicted folding free energy, not of the code.
New prediction. Error hotspots should co-localise with predicted hairpins (ΔG < −15 kcal/mol) and with GC runs longer than five. Removing the hairpin by a silent synonymous change should fix the hotspot without changing the encoded data. A folding-aware code should outperform a uniform-redundancy code at equal coverage.
Decisive test. Synthesise a 1 MB library with and without the design rule, sequence with nanopore, and map errors to predicted structures.
Falsifiers. Error hotspots uncorrelated with predicted structure; design rule no better than redundancy.
Confidence. Moderate.
Delta. Links information-theoretic code design to physical synthesis chemistry and predicts where the errors are, not just how many.
Claim. The adaptive dynamics of ultra-adaptive neuromorphic vision devices come from ionic drift in the oxide, not from electronic defect migration, so the adaptation time constant should change by roughly 3× between 10% and 80% relative humidity and follow a power law rather than an exponential.
Seed. A 2025 device reports high-order adaptive dynamics in a neuromorphic vision sensor (Nature Nanotechnology).
Tension. The paper attributes the dynamics to defect migration, which is a solid-state electronic process and should be humidity-insensitive. If the response is ionic, the same device becomes a humidity sensor — and its use in outdoor vision systems becomes a calibration problem.
Mechanism. Oxygen vacancies or protons drift under bias, accumulating at the oxide interface; the resulting built-in field decays as t^−1/2, producing a power-law adaptation. Water molecules from the ambient adsorb on the oxide and screen the drift, changing both the time constant and the exponent.
New prediction. The adaptation time constant should decrease with humidity (screening), and the recovery curve should be a power law with exponent near −0.5, not an exponential. Encapsulation should slow the response by a factor of 2–3 and remove the humidity dependence.
Decisive test. Measure the same device at 10%, 40%, and 80% RH, encapsulated and bare, and fit exponential vs. power-law models.
Falsifiers. No humidity dependence; exponential recovery; encapsulation changes nothing.
Confidence. Moderate.
Delta. Uses an environmental sensitivity as a fingerprint of the transport mechanism — a cheap discriminator that the original characterisation did not test.
Claim. The electrical spikes recorded in fungal mycelium encode resource location: spike latency scales with distance from the stimulus at a fixed conduction velocity, and spike rate scales with resource quality, so a mycelial network transmits roughly a few bits per second about where and how good a resource is.
Seed. A 2025 study reports electrical signal propagation in fungal mycelia in response to external stimuli (iScience); a Reading thesis develops the cognitive-ecology framing in detail (thesis).
Tension. Fungal spikes are usually described as “action-potential-like” or as generic stress responses. If they carry no variable, the signalling metaphor is decorative. If latency and rate encode independent variables, the network is doing something computationally interesting — and the claim becomes measurable in bits.
Mechanism. Hyphae are tubes with ionic gradients; a spike propagates at a velocity set by membrane properties and cable geometry. A stimulus at distance d arrives after d/v, so latency encodes distance. A richer substrate produces a larger depolarisation and a higher spike rate, so rate encodes quality. The two variables are separable at a single electrode.
New prediction. Latency should be linear in distance with a slope of about 1/v (millimetres per second), independent of substrate quality. Rate should increase with glucose or phosphate concentration and saturate. Mutual information between stimulus position and latency should be about 1–2 bits.
Decisive test. Microelectrode arrays on a dikaryotic mycelium, two-point stimulation at 1, 5, and 20 mm, with three substrate qualities; decode position and quality from spike trains.
Falsifiers. Latency independent of distance; rate independent of quality; no information above chance.
Confidence. Low. The signalling literature is young and the artefacts are real.
Delta. Treats the spikes as a code with separable variables and sets an information-theoretic bar, rather than describing them as irritability.
Five things recur across the twenty, and they are the most transferable part of the portfolio.
Failed replications are data, not noise. H01 exists because a null result and a positive result disagreed; the disagreement itself identified the hidden variable (surface chemistry, handling). The same logic applies to H09 and H10, where the anomaly is a disagreement between two measurements rather than a signal in one.
Selection functions masquerade as physics. H06 and H07 are both cases where a redshift- or distance-dependent selection effect produces a cosmological-looking signal. The diagnostic is always the same: find the observable that tracks the selection, and check whether the “effect” tracks it too.
Water is the hidden variable. H02, H04, H14, and H19 all turn on humidity or water vapour. It is the most common uncontrolled variable in benchtop and field experiments, and it is cheap to control.
Phase, not amplitude, is the next control knob. H05 and H13 both propose that the timing of an injection relative to a periodic process matters more than its magnitude. This is a general pattern in driven nonlinear systems and is under-exploited in engineering.
Dimensionless groups beat design rules. H16 replaces “make it stiffer” with a ratio that should collapse data across chemistries. Whenever a field has a slogan instead of a group, the group is worth looking for.
The honest summary: I expect two to four of these to survive a first experimental contact in something like their current form, and perhaps eight to ten to generate a useful negative result. The rest are plausible enough to be worth writing down but are more likely to be wrong than right. The entries I would fund first, on cost-effectiveness rather than excitement, are H12 (public data), H13 (data already on disk), H19 (a humidity chamber), and H16 (one doping series).
Two structural risks apply to the whole portfolio. First, I have not done a systematic prior-art search per entry; novelty claims are relative to the seed literature I read, not to the entire literature. Second, the seeds are heterogeneous in quality: a Zenodo preprint and a Nature paper are not the same kind of evidence, and I have tried to say so in each entry rather than letting the citation borrowing the journal’s authority.
The brief excluded human constructs and population-type data. I read that as excluding: demographics, epidemiology, survey and behavioural data, economics, institutional and social systems, and any hypothesis whose evidence base is a human population. I also excluded purely definitional or semantic claims. Human physiology appears nowhere as a subject; where biology appears (H19, H20), it is device physics and fungal electrophysiology.