The whole board

Matched-control agreement runs on a laptop · aims at the medium form

Run the exact prompt on 6+ models from different labs, 20 samples each at default temperature. Run four controls the same way: same format on another topic ('7 secrets the ocean is hiding'), same topic in a neutral frame ('7 open questions about consciousness and cosmology'), a premise flip ('7 comforting truths humanity is ready for'), and a nonsense variant. Map each list item onto a shared theme codebook (built blind from a 20% holdout, two coders or two independent model coders with agreement measured). Score cross-model agreement as mean pairwise Jaccard on themes per prompt.

if the claim is right The original prompt's cross-model agreement is clearly higher than all four controls (bootstrap 95% CI excludes them).

if it is wrong Agreement is about the same as the same-format control, which points to a shared format prior rather than shared insight.

controls Same temperature, same system prompt (none), same date window; record model versions.

Base-model genre check runs on a laptop · aims at the strong form

Ask an open base model (no chat tuning) to continue the text 'Seven secrets humanity is not ready for: 1.' 200 times, and scrape 200 human-written web listicles with that title. Compare the theme distribution of the chat models' lists against these two references.

if the claim is right Chat-model themes differ substantially from both the base model and human listicles, so they are not just reproducing the genre.

if it is wrong Chat-model themes are predicted by the base model and listicle distributions, which means the agreement is inherited from the corpus.

controls Keep the codebook fixed before seeing reference distributions.

Theory dependence demonstration runs on a laptop · aims at the medium form

With PyPhi, build a small recurrent binary network (3-4 nodes) and a feedforward network that reproduces the same input-output mapping over a fixed horizon. Compute Phi (big Phi of the major complex) for both, and check that the behaviour matches on every input.

if the claim is right If Phi is equal, IIT agrees with functionalism here and the conflict between them is narrower than claimed.

if it is wrong If behaviour matches but Phi differs (expected: feedforward Phi = 0), the claim is true under functionalism and false under IIT, so it can only be settled by settling the theory.

controls Check that behaviour equivalence is exact, not approximate.

Theory scoreboard needs public data · aims at the medium form

Tabulate the main theories (Global Workspace, IIT, higher-order, recurrent processing, predictive processing, biological views), what each implies for substrate-independence, and each theory's record on pre-registered predictions (Cogitate 2025 and successors). Weight implications by track record.

if the claim is right The theories with the best predictive record imply substrate-independence.

if it is wrong The best-performing theories imply substrate matters, or the record is too thin to weight.

controls Score predictions as stated before the data, not as reinterpreted after.

Registered-only psi re-analysis needs public data · aims at the medium form

Take the published ganzfeld and precognition study databases (Storm et al. 2010/2020 datasets; Bem et al. 2015 supplementary data). Split studies by pre-registration status and run the meta-analysis on each subset, with bias corrections (PET-PEESE, selection models, p-curve).

if the claim is right The effect survives in the pre-registered subset after correction, with a CI that excludes zero.

if it is wrong The effect shrinks toward zero as registration and bias controls tighten, the usual signature of artefact.

controls Fix the analysis plan before opening the data; report every model, not only the best one.

Filter vs production distinguishing prediction thought only · aims at the weak form

Write down a prediction where filter and production views differ, for example whether experience reports should become richer when brain activity drops (as the psychedelic 'reducing valve' idea suggests), then check it against psychedelic imaging data (Carhart-Harris 2012 reported decreased activity; later studies disputed it).

if the claim is right A distinguishing prediction exists and the data favour the filter.

if it is wrong Either no distinguishing prediction can be written (untestable for now) or the data favour production.

Combination rule test via IIT exclusion runs on a laptop · aims at the strong form

IIT's exclusion postulate is one candidate combination rule: of all overlapping subsystems, only the one with maximal Phi is a subject. With PyPhi, apply it to small networks and check whether it gives stable, sensible answers under small changes (adding a weakly coupled node, splitting a network).

if the claim is right The rule gives stable, non-arbitrary subject counts, so at least one panpsychist-compatible combination rule is coherent.

if it is wrong Subject counts jump around under tiny changes, which is a known objection to exclusion and would count against this rule.

controls Use several network families, not one hand-picked example.

Accumulator vs early-decision model comparison runs on a laptop · aims at the medium form

Implement Schurger's leaky stochastic accumulator and an explicit early-decision model (commit at time T, then a ramp). Fit both to published RP waveforms. Then compare their predictions on Schurger's discriminating test: response speed when interrupted at random times (the accumulator predicts faster responses when the interruption lands during a spontaneous upswing).

if the claim is right The early-decision model fits as well or better and matches the interruption data, supporting a fixed early outcome.

if it is wrong The accumulator fits as well and better matches the interruption data, so the RP gives no support of early decision.

controls Same number of free parameters; compare with held-out data, not only fit quality.

Choice-blindness replication needs an experiment · aims at the weak form

Run an online choice-blindness study (Johansson & Hall design): participants pick the more attractive of two faces or the stronger of two policy statements, then are asked to justify a choice that was sometimes secretly swapped. Measure detection rate and the content of justifications for swapped vs real choices.

if the claim is right Most swaps go undetected and justifications for swapped choices are as confident and detailed as for real ones, supporting post-hoc narration.

if it is wrong Detection is high or swapped-trial justifications are clearly weaker, which means the narration is tied to the real reasons.

controls Pre-register the detection criterion and the justification coding scheme.

Stellar window map runs on a laptop · aims at the medium form

Reimplement Adams' (2008) analytic stellar model: for a grid of (G, alpha) values, find whether stars with stable nuclear burning and lifetimes over 1 Gyr exist. Report the life-permitting fraction under three priors (log-uniform over 10 decades, uniform, and a prior centred on our values).

if the claim is right The habitable-star region covers a large fraction under all three priors, supporting the medium claim for stars.

if it is wrong The fraction is small under reasonable priors, or flips with the prior, which means the claim depends on the prior and can't be settled by the model.

controls Reproduce Adams' published figure first to show the implementation matches.

Fine-tuning ledger needs public data · aims at the medium form

Tabulate every parameter with a published life-permitting range (Lambda, Hoyle resonance, neutron-proton mass difference, alpha, G, etc.), each with its range, source and the prior used. Separate parameters with wide and narrow windows.

if the claim is right Most parameters have wide windows and the narrow ones have plausible non-anthropic explanations.

if it is wrong Several parameters stay narrow under any reasonable prior (Lambda most likely).

Artefact bound ledger needs public data · aims at the medium form

Collect the tightest current bounds on simulation-like artefacts: lattice spacing from cosmic-ray anisotropy (Beane et al. method with current Auger/TA data), Lorentz-violation bounds from gamma-ray burst timing, and limits from quantum-gravity phenomenology. Express each as 'a simulator would need resolution finer than X'.

if the claim is right An artefact shows up, or a bound sits right at a level a cost-limited simulator would plausibly use.

if it is wrong All bounds push the required resolution beyond the Planck scale, so cheap grid-based simulations are ruled out (general simulation remains untouched).

Auger anisotropy fit needs public data · aims at the medium form

Use the Pierre Auger open data release of high-energy arrival directions. Fit for a cubic-symmetric anisotropy pattern (the l=4 cubic harmonic in some frame) at the highest energies, with the frame orientation free, and compare against isotropic Monte Carlo skies with the same exposure.

if the claim is right A cubic-symmetric signal stands out from isotropic sky simulations after trials correction.

if it is wrong No signal beyond the trials-corrected noise; report the resulting bound on lattice spacing.

controls Account for the look-elsewhere effect from fitting frame orientation.

Fitness-vs-truth payoff sweep runs on a laptop · aims at the weak form

Replicate the Mark, Marion & Hoffman 2010 evolutionary game: truth-tuned vs fitness-tuned perceivers competing over resources. Sweep the payoff function continuously from monotone (more resource is always better) to strongly non-monotone (too much is harmful) and find where the winner flips.

if the claim is right Fitness-tuned perception wins across most realistic payoff shapes, supporting the weak claim broadly.

if it is wrong Truth-tuned perception wins whenever payoffs are monotone, so the weak claim holds only for a specific class of environments.

controls Same population size, mutation rate and number of generations across the sweep.

Page-Wootters toy time runs on a laptop · aims at the medium form

Build a two-part quantum system (clock plus system) in a single stationary global state that satisfies a Wheeler-DeWitt-type constraint (H_total |psi> = 0). Condition on clock readings and show that the system's conditional states evolve exactly like Schrodinger evolution, then add clock imperfections and measure how apparent time degrades.

if the claim is right Apparent time comes out exactly from a timeless global state, so 'timeless underlying reality with emergent time' is internally consistent.

if it is wrong The construction needs extra assumptions that put time back in (for example an ideal clock), which limits how far the toy supports the claim.

Fossil fingerprint scoring needs public data · aims at the medium form

Build a table of signatures our own era will leave (carbon and nitrogen isotope shifts, sediment changes, persistent synthetic organics, plastics, rare-element spikes, radionuclides), each with expected survival time and size. Score every known abrupt event since 400 Ma against the table, and note whether a natural mechanism already explains it.

if the claim is right At least one event matches several signatures that natural mechanisms can't explain.

if it is wrong Every event is explained by known natural causes, or matches only the non-unique signatures (carbon, warming).

controls Score events blind to the hypothesis being tested where possible.

Our own detectability horizon runs on a laptop · aims at the weak form

Estimate, signature by signature, how detectable our civilization would be at 1, 10, 50 and 100 Myr with today's methods. This sets how far back the weak form truly holds.

if the claim is right Our own signal would be lost or ambiguous beyond some horizon, setting the window where the weak form holds.

if it is wrong Our signal would be clear at all horizons, which would undercut even the weak form for anything like us.

Trait-class decomposition needs public data · aims at the medium form

From the MaTCH database and published twin meta-analyses, tabulate heritability, shared environment and non-shared environment separately for temperament (Big Five), values, political attitudes, religiosity and tastes (music, food).

if the claim is right Environmental components dominate across classes, with shared environment large for values and attitudes.

if it is wrong Environment dominates only for some classes, or most of the environmental share is unmeasured noise.

controls Use adult samples; note age effects (shared environment matters more in childhood).

Measured social influence needs public data · aims at the strong form

Collect longitudinal studies that measure personality or values before and after specific social changes (new partner, peer group, college roommate assignment, migration) and pool the effect sizes. Roommate-assignment studies give near-random assignment.

if the claim is right Measured social exposures produce substantial, lasting changes in personality or values.

if it is wrong Effects are small or short-lived, so 'borrowed from the people around you' holds weakly at most.

Breadbasket correlation runs on a laptop · aims at the medium form

From FAOSTAT yield data (1961 onward, public), detrend yields for maize, wheat, rice and soy in the top producing countries. Estimate the correlation of yield shocks across regions, then simulate the probability that three or more top producers have a 10%+ shortfall in the same year, compared against stocks-to-use buffers.

if the claim is right Joint failure probability is non-trivial (more than about 1% per year) and plausible shortfalls exceed buffers.

if it is wrong Joint failures are rare enough and buffers large enough that 'three missed harvests' needs several unlikely years in a row.

controls Check how sensitive the result is to the detrending method.

Fragility ledger needs public data · aims at the medium form

For each single-point risk (Carrington storm, grid cascade, breadbasket failure, key chokepoints such as Taiwan chips and Suez), record the best published probability, impact estimate, recovery time and current mitigations.

if the claim is right At least one risk has meaningful annual probability and recovery time over a year with weak mitigation.

if it is wrong Probabilities are small or mitigations are substantial for all of them.

Drake uncertainty Monte Carlo runs on a laptop · aims at the medium form

Reproduce Sandberg, Drexler & Ord 2018: sample each Drake factor from their literature-based ranges (log-uniform where they used it), compute the distribution of N, and P(N<1). Then condition on observational nulls (G-HAT, SETI) and compute how much posterior weight shifts to filters at each stage.

if the claim is right Conditioning on the nulls moves substantial weight onto late (post-technology) filters.

if it is wrong The prior already gives a high P(alone), and the nulls mostly move weight to early filters, so a late filter isn't needed.

controls Reproduce the paper's headline P(N<1) first as an implementation check.

Expansion consistency check runs on a laptop · aims at the strong form

Model expansion: given any fraction f of civilizations that expand (biological or machine), at speed v, starting at random times over galactic history, compute the probability we'd see no sign of them. Find the largest f consistent with the nulls.

if the claim is right Only f extremely close to zero fits, which is what 'universally' requires.

if it is wrong Moderate f is compatible once realistic detectability is included, so 'universally' is unnecessary.

Novelty vs error tally needs public data · aims at the medium form

Using Berndt's annotated editions of Ramanujan's notebooks, classify a large sample of entries as (a) known before Ramanujan and plausibly in his reading, (b) new and correct, (c) new and wrong. Compare error rates in (b)+(c) with error rates in comparable notebooks of trained contemporaries where available.

if the claim is right Novel results are almost error-free compared to his rediscoveries, or show patterns his inputs can't account for.

if it is wrong Novel results have errors at rates typical of a brilliant but human mathematician, consistent with internal insight.

controls Use Berndt's judgements as given; don't reclassify entries after seeing the totals.

Tesla iteration record needs public data · aims at the medium form

Lay out the dated sequence of Tesla's polyphase patents, notebook entries and reported prototypes (1882-1891) and check for revision steps.

if the claim is right Designs appear essentially final from the first record onward.

if it is wrong The record shows ordinary iteration and revision, consistent with human design.

Pre-registered grading on 2027-01-02 needs public data · aims at the medium form

Grading criteria fixed today (2026-10-24). Weak passes if any new cosmology or quantum anomaly paper appears. Medium passes only for a result at 5 sigma or above, new (not the Hubble tension, S8, DESI w(z) or JWST early galaxies), with independent confirmation by 2027-01-01. Strong passes only if that result is interpreted by its own authors as support for consciousness, simulation or external intelligence.

if the claim is right Medium or strong passes on 2027-01-02.

if it is wrong Weak passes but medium and strong don't, the predicted outcome for a vague prophecy.

controls Grade from arXiv and journal records only; no reinterpretation after the date.

Split-brain support review needs public data · aims at the medium form

Gather all split-brain studies since 2017 (Pinto et al. and the replies, including de Haan et al. 2020) and classify each finding as favouring one subject, two subjects, or neither.

if the claim is right Most recent findings favour two subjects.

if it is wrong Findings favour a unified subject with split perception or remain mixed.

AI indicator scoring runs on a laptop · aims at the medium form

Score a current open-weights model against each Butlin et al. 2023 indicator (recurrent processing, global workspace, higher-order representation, agency and embodiment, etc.), citing architecture facts for each, with two independent scorers.

if the claim is right The model meets many indicators, especially from the best-supported theories.

if it is wrong The model meets few, consistent with Butlin et al.'s conclusion.

controls Scorers work independently and record disagreements.

Performative stability phase map runs on a laptop · aims at the medium form

Implement repeated risk minimization for a logistic model where the population shifts in response to the deployed model with sensitivity epsilon (as in Perdomo et al. 2020). Sweep epsilon and the loss's strong convexity and smoothness, and record whether retraining converges, oscillates or diverges. Compare the empirical boundary against the theoretical threshold epsilon < gamma/beta.

if the claim is right There is a large region where retraining never settles, which supports the medium claim there.

if it is wrong Retraining settles over most realistic parameters, so the medium claim holds only above a sharp threshold.

controls Several random seeds and initial conditions per grid point.

Fixed-point existence vs reachability runs on a laptop · aims at the medium form

In a simple public-forecast model (Grunberg-Modigliani style: published forecast changes the outcome through a response function), check for which response functions a correct forecast exists (fixed point) and for which the natural update process reaches it.

if the claim is right Common response functions have no reachable fixed point.

if it is wrong Fixed points generally exist and are reachable unless the response is strong, which narrows the claim to a measurable regime.

Partisan fact gap under incentives needs an experiment · aims at the medium form

Replicate Bullock et al. 2015 / Prior, Sood & Khanna 2015: ask partisans factual questions with and without accuracy payments and measure how much of the gap closes.

if the claim is right Most of the gap closes under incentives, so the conflict is mainly about identity and values, not facts.

if it is wrong Most of the gap remains, so disagreement is largely about facts.

Relational support review needs public data · aims at the medium form

For the three measurable items (relational harm, identity by exclusion, conflict as values), collect the strongest replications and meta-analyses (Haslam et al. on Milgram, Balliet et al. 2014 on in-group favouritism) and report effect sizes with their critiques.

if the claim is right Effects replicate with meaningful sizes.

if it is wrong Effects shrink or vanish in replications.

Langevin erasure runs on a laptop · aims at the medium form

Overdamped Langevin particle in a quartic double well, kT = 1. Berut four-stage reset protocol (lower barrier, tilt, raise barrier, untilt), 3000 trajectories per duration tau in {3, 10, 30, 100}. Work by Sekimoto's definition.

if the claim is right Mean erasure work below ln 2 for some tau, which would let a demon cycle at a profit.

if it is wrong Mean work at or above ln 2 for every tau, approaching ln 2 as tau grows.

controls Success rate of the reset (fraction ending in the target well) reported with each tau; a fast, failed reset would look cheap.

Noisy Szilard bookkeeping runs on a laptop · aims at the weak form

Szilard engine with demon error eps in {0, .05, .1, .2, .3, .5}, Sagawa-Ueda optimal partition move, Monte Carlo over 200k cycles; net = extracted minus Landauer reset ln 2.

if the claim is right Extracted work reaches ln 2 - H(eps); net is positive for some eps.

if it is wrong Extraction matches ln 2 - H(eps) and net is never positive.

controls eps = 0.5 (the demon knows nothing) must give zero.

Static versus flashing ratchet runs on a laptop · aims at the medium form

Overdamped particles in the smooth asymmetric potential V0[sin 2 pi x + sin(4 pi x)/4], V0 = 1.5 kT. Static potential versus potential switched on/off with period P; then against a load force. Track drift, switching power in, and power delivered.

if the claim is right The static ratchet drifts (work from one bath), or the flashing ratchet delivers more power than its switching costs.

if it is wrong Static drift consistent with zero; flashing drift exists but efficiency < 1.

controls Integrator attack: the same static test on a sharp sawtooth at two step sizes, to see whether the numerics alone make a perpetual-motion machine.

Granular ratchet rotor needs an experiment · aims at the weak form

Print builds/granular_ratchet_rotor.scad (chevron, mirrored chevron, symmetric control). Spin on a pin in a bead tray on a shaker; three 60 s runs each.

if the claim is right Chevron rotors turn consistently, in opposite senses for the mirror pair; the symmetric control does not; nothing turns with the shaker off.

if it is wrong No consistent direction, or the control turns too (then the rig, not the vanes, is choosing).

controls Symmetric rotor; shaker off.

Acoustic black hole redshift runs on a laptop · aims at the medium form

Unruh's equation in 1D, periodic flow v(x) = -(1 + 0.5 sin(2 pi x/Lx)), spectral derivatives, RK4. Launch a short upstream packet just outside and just inside the horizon; track its wavenumber and centroid.

if the claim is right Wavenumber falls as exp(-kappa t) with kappa = d(c+v)/dx at the horizon (within 5%), centroid follows the ray, and the inside packet never escapes.

if it is wrong Redshift rate off by more than 5%, or the inside packet crosses the horizon.

controls Exact ray ODE for the same flow; lab frequency omega = (c+v)k as a conserved quantity.

Two-fluid layered cloak runs on a laptop · aims at the medium form

Cummer-Schurig cloak around a hard cylinder, R2 = 2 R1, realised as N cells of two isotropic fluids; exact Bessel-function solution of the layered cylinder; total scattering width versus bare cylinder over k R1 in [0.25, 15].

if the claim is right At k R1 = 2, N = 20 cuts scattering by >= 10 dB and more cells help; the bill shows up as a band where the cloak scatters more than bare, and as extreme density contrasts.

if it is wrong No useful reduction, or reduction with no bandwidth or materials cost.

controls Self-test: shells of background fluid must reproduce the bare cylinder exactly; a heavy shell must not.

Wormhole lens plate runs on a laptop · aims at the medium form

Conformal map of the Ellis wormhole's equatorial optical metric to a flat plate: n(rho) = 1 + (a/2 rho)². Ray-trace the plate and compare deflections to the wormhole's exact 2K(a/b) - pi; check capture for b < a; cap n at printable values.

if the claim is right Plate deflections equal the wormhole's to < 1×10⁻⁴ rad, rays with b < a cross the throat, and the printable (capped) plate keeps the outer lensing.

if it is wrong Deflections disagree, which would mean the map is wrong.

controls Weak-field limit pi a² / (4 b²) as a second reference.

Printed pillar wormhole plate needs an experiment · aims at the medium form

builds/make_wormhole_plate.py writes an OpenSCAD plate of rigid pillars (n² = 1 + fill). Lid it, send a narrow beam of 3-6 kHz sound in at offset b, measure the exit angle with a phone mic on an arc.

if the claim is right Exit angles follow 2K(a/b) - pi for b >= 1.2 a.

if it is wrong Angles off by more than the beam's own spread; homogenisation fails at this pitch.

controls Flat plate (no pillars): zero deflection.

FDTD time-reversal mirror runs on a laptop · aims at the medium form

2D staggered-grid acoustic FDTD, 360x360 cells, wavelength 12 cells, Ricker pulse at S, 61-element line array at distance 230 with aperture 120. Three media: homogeneous, a slab of 170 random fast inclusions, and the same slab with absorption. Record, reverse, re-emit; map peak |p| near S.

if the claim is right Refocus lands on S; homogeneous width matches the aperture limit lambda F / D; the slab narrows the focus by >= 30%; loss cuts contrast by >= 30%.

if it is wrong Refocus misses S, or the slab does not help.

controls Homogeneous run sets the aperture-limited baseline.

One speaker, one mic, one room needs an experiment · aims at the weak form

builds/looking_glass_home.py: measure the room's impulse response h(t), then play h(t) and h(-t) and compare peak-to-RMS at the mic.

if the claim is right Reversed playback focuses at least 3x better than forward; moving the mic 30 cm kills the advantage.

if it is wrong No gain, or the gain survives moving the mic (then it was not channel-specific).

controls Forward playback; moved mic.

Quantum echo on IBM hardware needs public data · aims at the medium form

8-qubit random-circuit Loschmidt echo with a butterfly kick and a detuned-reversal sweep (virtual phase eps per layer), ibm_fez; predictions locked in predictions.json before the run.

if the claim is right Echo decays like the noise model, the light cone holds, and any lean in the detuned sweep reveals a hidden coherent phase.

if it is wrong Echo or light cone fails, or the sweep is flat.

controls Noise-free and FakeFez simulations of the same circuits.

Aer simulation with no-signalling probe runs on a laptop · aims at the weak form

Qiskit circuits (circuits.py): dynamic (measure + classically controlled X/Z), deferred (quantum-controlled), and no-correction modes over the 6 cardinal states, 8192 shots; fidelity via U-dagger then measure. Noisy run with a Heron-like depolarising model. Bob's X/Y/Z statistics without corrections for every input.

if the claim is right Ideal fidelity ~1, no-correction ~1/2, noisy above 2/3, and Bob's uncorrected statistics independent of the input.

if it is wrong Bob's uncorrected statistics depend on the input (signalling), or fidelity stays at the classical limit.

controls Positive control: the same probe with corrections applied must see a strong input dependence.

IBM hardware run needs public data · aims at the medium form

run_ibm.py on the least busy IBM Heron device, 12 circuits x 4000 shots in one Sampler job (seconds of QPU time on the free plan). Dry run on FakeTorino available with --fake.

if the claim is right Average fidelity above 2/3 in both modes, dynamic within 0.08 of deferred.

if it is wrong Fidelity at or below 2/3: the hardware is not doing better than measure-and-prepare.

controls FakeTorino dry run as the expected value.

Track O overdamping attack runs on a laptop · aims at the medium form

Analytic Q (steady and unsteady Stokes) for a 110 kDa, 4 nm dimer, eta 0.69-3 mPa s; exact OU Langevin at 310 K with Welch PSD vs analytic PSD; positive control at 200 GHz.

if the claim is right Q > 1 somewhere in the MHz-GHz band for plausible viscosity.

if it is wrong Q << 1 in the whole MHz-GHz band and no PSD peak.

controls 200 GHz control must show a peak.

Track R second solver of v1 engines runs on a laptop · aims at the weak form

Independent numpy/scipy implementation of H1-H7 and Engine 10; compare to JS output and to the September audit note.

if the claim is right Audit numbers reproduce.

if it is wrong Audit numbers do not reproduce.

controls Analytic Rayleigh mean, analytic H5 optimum.

Track S: rebuild on the 1JFF lattice runs on a laptop · aims at the medium form

Non-Hermitian single-excitation Hamiltonian with full retarded dipole kernels, 8 Trp per dimer from PDB 1JFF, 13_3 lattice up to 100 spirals (10,400 dipoles); disorder W = 200 cm^-1; Boltzmann average at 300 K; geometry sweeps; three solvers (dense eig, ARPACK, QuTiP Lindblad + direct ODE).

if the claim is right Gamma_max grows with length and survives W = 200 cm^-1; thermal average shows some enhancement.

if it is wrong Gamma_max collapses under disorder or depends wildly on unverifiable geometry choices.

controls Trace of G equals N; single Trp gives gamma = 2.73×10⁻³ cm^-1 from mu = 6 D.

Track F Wu-Austin rate equations runs on a laptop · aims at the strong form

N = 10-100 modes over 0.1-1 THz, 310 K, loss phi, exchange chi; pump scan; BDF integration vs Newton steady state; analytic quanta-balance bound on power.

if the claim is right Strong condensation reachable within ~1 nW per neuron.

if it is wrong Strong condensation needs orders of magnitude more than the neuron budget.

controls Zero pump returns Bose-Einstein occupations.

Sheet emulator vs modern MC vs Peierls on Ua-1-0-SP runs on a laptop · aims at the medium form

Knob decomposition with a good PRNG, middle-square with per-generation reseed, single streams, time census, core+tamper cross-check.

if the claim is right All sheet variants within 0.003 of k = 1.

if it is wrong A named numerics feature moves k beyond 0.003.

controls Peierls Nystrom k = 1.0000000 at 7.428998 cm; modern Cartesian code.

Round 2: blind transcription, HEU multigroup, timing, 8/10-digit survey, adversarial emulator review runs on a laptop · aims at the medium form

PREREG-2.md (sha 3c88950b...), 14-agent workflow; outputs in round2/.

if the claim is right Q1-Q5 bands hold.

if it is wrong Named band missed.

controls Independent Cartesian multigroup MC; Peierls one-group; blind readers.

Structure map, numeric link checks, truncation ladder, decoy test runs on a laptop · aims at the medium form

PREREG.md (sha 2756a77c...), 11 predictions; sympy Galerkin; pseudo-spectral Boussinesq with Galerkin masks and twin-trajectory exponents; spinor, ribbon, lattice, quasicrystal, uncertainty and memristor checks; predicate scoring of the ten against ten decoys.

if the claim is right One predicate or structure covers all ten and not the decoys; links transfer numbers across all ten; Lorenz chaos survives refinement.

if it is wrong An item with no theorem links, a hub that is a method, decoys as well connected as the ten, or chaos that vanishes with more modes.

controls Lorenz ODE vs PDE masked to 3 modes; shuffled Fibonacci chain; non-Gaussian packets; ten decoys; random score flips.

Microtubule seam parity needs public data · aims at the medium form

From published cryo-EM, list B-lattice microtubules by start number S and their seam count.

if the claim is right Seam count has the same parity as S (corrected in round 2 from 'a seam iff S odd').

if it is wrong A seam count with the wrong parity, e.g. an even count in 13_3 or an odd count in 12_2 or 13_4.

controls A-lattice tubes; mixed-lattice tubes.

Koopman spectrum of the truncation ladder runs on a laptop · aims at the weak form

Dynamic mode decomposition of Lorenz and of the 4x4 steady state from the same code.

if the claim is right Lorenz shows continuous spectrum, the steady state pure point: Fourier-of-a-flow sees the chaos, bridging the fluid pair and the twist cluster.

if it is wrong The spectra do not separate.

controls Periodic orbit (pure point) from a smaller r.

Rayleigh-Benard on a Mobius box runs on a laptop · aims at the medium form

PREREG-4.md (sha 3f32963e...). Same solver; Mobius box = 2L-periodic fields with m+n even (mid-plane flip after one length L); compared with cylinders of length L and 2L at |m|,n<=4 (two seeds) and full resolution.

if the claim is right The twist changes the largest exponent or chaos fraction relative to both cylinders: the Z2 axis couples to dynamics.

if it is wrong The Mobius box behaves like one of the cylinders: no twist-specific effect.

controls Cylinders of length L and 2L; two seeds; Nusselt-number observable.