Our mission, stated as an experiment

Mission: “an experiment we are working to prove”

Working hypothesis — registered, not a result.

“The correlations a two-state record carries constrain the process that induced them, and that process can be estimated from the record alone; nothing observes it directly.”

We test that reading on the Bloch sphere: on the Earth's magnetic record first, then on records from beyond the Earth. Every test is registered before its data is read, and every verdict on this page is read from its record, the misses first.

Of the thirteen predictions in our papers, two have run, both on the Earth's record: one held at its threshold — window-fragile, consistent with the textbook expectation and not as support — and one was refuted. Other tests on Earth records have failed and are listed below. The four tests beyond the Earth were registered on 7 October 2026 (frozen block sha256 34526205fe68…) and have not been run.

What we count, each from its record: 14 tests of the model's assumptions (4 failed, 6 held, 4 not run); 13 predictions in our papers, 2 run, both on the Earth's record; 41 failed tests across our status record, the table below and The unit of work, counted once each: 18 on this page, 16 more in that paper and 7 more in the record only.

Every object the model names, in one order: the figure Day drew from its record, its status word, its verdict read from the record that holds it, the limit it sits under and what would close it. The losses come first, and nothing on this page is a result about nature beyond what a record measured.

Every verdict is read from its record, never typed: held, held at its threshold, falsified, refuted, unattributed, not measurable, registered and not run. The losses come first.

The record, week by week: what was read, released and published.

Day AGI
What Day does for you, and what Day has been measured to do and not to do, failures first.
Roles
Who does what behind our numbers and our work, and what each role gives you.

Experience generation is Day computing something from a record we keep and saying, beside it, what kind of evidence it is. At this release, every figure on this page was drawn by Day from a record we keep, and its caption says what kind of evidence it is.

The axioms and the working hypothesis

What this is

The engine computes over structure: a two-state model read three ways, as the qubit at height m, the halo ring of radius √(1 − m²) at angle θ, and the inducing event carrying H₂(cos² α) bits per half. Its consequences are derived as theorems in the model and pinned, each re-derived by a named test to a stated tolerance. Its predictions are registered before the data is read, and each verdict we print is read from its record, the misses first.

The working hypothesis's falsifiers, losses first

What would count against the working hypothesis

A registered Earth, planted-body or cosmology row refuted at its threshold is a result against the working hypothesis as a whole. Said first: F-BH5 is measured: refuted (read from fixtures/earth/earth_readings_2026_10_04.json#s8_orbital_periods). Its registered consequence: the link above the Earth is a lineage, not a generation.

What you can check

What our own tests refuted

The tests in this table failed by their own registered lines before this round; each verdict is read from the record its row names.

What our own tests refuted, losses first
TestWhat it saidThe numberThe line it had to clearVerdictRecord
F12F12 · ΔE is the clock (the rate clause)r = −0.063; 0 of 20 null-rate beats vanish (smallest θ̇ 0.0726)r ≥ +0.5; θ̇ → 0 at in = outfalsifieddeploy/theta_clock.json (rows b2, b3)
F9 (S13)F9 / S13 · the emission condition and the alignment gate are one gatethe gates agree on 63 of 200 beats (0.315); aligned fires 198, Shannon 61agreement ≥ 0.9falsifieddeploy/theta_clock.json (row c)
F13 (S5.b)F13 / S5.b · the fixed protocol order is as fast as randomD_fixed/D_rand = 0.891 (the raster order is 10.9 % slower)within ±0.10 of 1falsifieddeploy/falsify_sim.json (S5.b)
F14 (S12)F14 / S12 · "the slider is m" (a broker's centrality on planted stars)22 of 120 stars agree (18.3 %)≥ 95 %falsifieddeploy/falsify_sim.json (S12)
F-A1F-A1 · the account has a direction (the trade desk's writer)12 of 24 series (0.5); every row reads +1≥ 0.9 of the seriesfalsifieddeploy/account_direction.json (F-A1.b)
F-M1 (i)F-M1 (i) · m strictly increasing in λ23 of 24 roster-temperatures strict; one float64 tie (frozen, T = 0.1: m = 1.0 at λ = 3 and 4)every roster-temperature strictfalsifiedfixtures/m_measured/rosters.json
F-M2F-M2 · the frozen roster reads |m| > 0.9 at T = 1south −0.761594, −0.753066, −0.739783 (north held)every |m| > 0.9falsifiedfixtures/m_measured/rosters.json
S8ERT-F7 / F-BH5 / S8 · an external clock sets the Earth's phase2 of 3 periods significant (41 kyr ratio 1.885, 100 kyr 1.040; 23 kyr 0.809 not)the registered expectation (refuted; "not the standard null either")refutedfixtures/earth/earth_readings_2026_10_04.json (s8_orbital_periods)
LENS R6LENS R6 · the Earth stand-in card's θ stays in the ballthe M-step ran θ to the box corner, max 2.0≤ 1falsifieddeploy/lens_cards.json (earth_standin R6)
LENS R4, R5LENS R4 / R5 · the trading cardR4 0.125; R5 the null sits inside the noisethe registered R4 bound; R5 a null outside the noisefalsifieddeploy/lens_cards.json (trading R4, R5)
SEISMO (iii)SEISMO (iii) · the 1-minute control reads NULLcoda S = 1.53 nT against the bound 0.17 nT: EXCEEDSS ≤ the boundfalsifiedfixtures/seismo/seismo_lens_results_2026_10_04.json (call iii)

The objects, in one order

Fifteen objects in the order both sites share: the three referentials, the clock, the wake rule, the mirror, the lock gate, the generating body, the Earth, the Sun and the microwave sky, the elements ladder, hardware, the seismic experiment, the Hilbert library and the working hypothesis. Status words, verdicts and records are read from deploy/mission_objects.json.

  1. 1. The qubit: the first referential

    theorem-in-the-model

    One half of a two-qubit pair in Schmidt form, cos a|00⟩ + e^{iθ} sin a|11⟩: each half's Bloch vector is (0, 0, m) with m = cos 2a. Standard two-qubit quantum mechanics; the product's moving parts are classical two-way switches, and no qubit is in it.

    Day’s figure: One pair in Schmidt form, read three ways, drawn on embark.day’s Mission page.

    Verdict: F1 registered, not run

    Limits (LIM-15, LIM-02): There is no qubit in the product: the moving parts are two-way switches on classical hardware, and the halo paper describes the textbook picture they can be read onto. No hardware pair has been read: every lens reading comes from an exact simulator.

    What would close it: One pair prepared and read on quantum hardware, with the readout calibrated first.

    Falsifiers: F1.

    Record: deploy/axioms.json, deploy/papers_graph.json.

  2. 2. The halo ring: the second referential

    theorem-in-the-model · measured-in-model

    The partner's reduced state at the same m: the ring at latitude m with radius √(1 − m²) and angle θ. With a non-degenerate ground state the halo needs a transverse term (γ ≠ 0).

    Day’s figure: The elliptical halo of one planted instance, γ = 0 and γ = 0.7, drawn on embark.day’s Mission page.

    Verdict: S2.a held; S2.b held; S2.c held; F2 registered, not run

    Limits (LIM-01, LIM-03): No phase has been measured on any space we hold: where a pair sits round its ring is the model's clock, not a reading. The composition law is untested: several variables tied by one event, and the halo they share, are future work, so the papers treat pairs only.

    What would close it: A pair read on hardware, or a recorded run whose phase is read and compared with the clock.

    Falsifiers: S2.a, S2.b, S2.c, F2.

    Record: deploy/axioms.json, deploy/atlas/halo_ellipse.json, deploy/falsify_sim.json, deploy/papers_graph.json.

  3. 3. The inducing event: the third referential

    hypothesis

    The process that induced the pair, read as a third referential and inferred from the correlations alone; the entropy of the parts is the measured limit of what the record can say about it.

    Day’s figure: One pair in Schmidt form, read three ways, drawn on embark.day’s Mission page.

    Verdict: none registered

    Limits (LIM-01): No phase has been measured on any space we hold: where a pair sits round its ring is the model's clock, not a reading.

    What would close it: A registered prediction from it tested on a record (F1 purity and F-BH1..5, F-BH8, F-BH-H are the nearest)

    Falsifiers: F1, F-BH1, F-BH2, F-BH3, F-BH4, F-BH5, F-BH8, F-BH-H.

    Record: deploy/experiments_status.json.

  4. 4. The clock θ = ΔE·t

    theorem-in-the-model

    θ = ΔE·t (`induction.theta_from_gap`): the phase turns at the gap; the pair's and the mirror's lock is |cos ΔE·t|. The rate clause F12, that the account's ΔE is the clock, was registered and falsified.

    Day’s figure: The ring turning at θ = ΔE·t, and the lock |cos ΔE·t|, drawn on embark.day’s Mission page.

    Verdict: F12 falsified; b2 falsified; b3 falsified; a1 held; a2 held; a3.i held; a3.ii held; a3.iii held; b1 held; d held

    Limits (LIM-01, LIM-31): No phase has been measured on any space we hold: where a pair sits round its ring is the model's clock, not a reading. The phase clock is the model's own: no running record reads a pair's turning, so the current side of the dual-field reading is unmeasured.

    What would close it: A clocked record that keeps the temperature and the phase as two named clocks and shows a turning halo on a driven pair.

    Falsifiers: F12, F2.

    Record: deploy/theta_clock.json.

  5. 5. The wake rule

    hypothesis

    The model's own step for m and θ between readings: θ advances with the gap and m relaxes toward its pole by the confidence c. Assumed (postulate P2); it never moves m away from its pole (pinned).

    Day’s figure: The wake rule's paths on the sphere, and the forty recorded wakes, drawn on embark.day’s Mission page.

    Verdict: F-B2 held; F-BH8 registered, not run

    Limits (LIM-04, LIM-30): The rules by which confidence and phase change over time are the model's own: no data has tested them. Nothing writes a belief back from what was observed: the step the papers call confidence is defined, and the engine does not yet book it on each settle.

    What would close it: A recorded sequence of decisions, read at each step and compared with the rules.

    Falsifiers: F-B2, F-BH8.

    Record: deploy/experiments_status.json, deploy/day_blooms.json, fixtures/dualfield/fbh8_halo_widening.json.

  6. 6. The mirror: the symmetric two-state chain

    theorem-in-the-model

    The symmetric chain P(S|R) = P(R|S) = τ: one step sends m to m′ = (1 − 2τ) m (eigenvalues 1 and 1 − 2τ); τ = ½ is the independent point and τ = 1 the perfect mirror.

    Day’s figure: The mirror's contraction m′ = (1 − 2τ) m, drawn on embark.day’s Mission page.

    Verdict: none registered

    Record: deploy/axioms.json.

  7. 7. The lock gate: two gates, one wording

    theorem-in-the-model · measured-in-model

    The emission condition is stated in information theory: a system may emit, putting a product, a message or a trade into its environment, once its own uncertainty, measured as Shannon entropy in bits, has reached the environment's. The alignment gate asks something else: how closely the system's units point the same way, a ratio from 0 to 1 against the environment's reading. It is not the emission condition, which measures no alignment. With every unit in one quarter of the plane the ratio cannot fall below 1/√2 (about 0.707), a theorem in the model. The prediction that the two agree nine times in ten is falsified: on 200 simulated beats from planted test data they agreed 63 times.

    Day’s figure: The alignment gate's lock and the primer's 2 S_A, on the same axes, drawn on embark.day’s Mission page.

    Also drawn by Day: The lock's floor 1/√2 under the 240 committed readings.

    Verdict: F9 falsified

    Falsifiers: F9.

    Record: deploy/experiments_status.json, deploy/theta_clock.json.

  8. 8. The generating body and its wire

    named-only

    A generating body, in the engine a decided pole with no halo and no clock: no wire carries its state into the field the engine reads, so F-BH1 to F-BH4 are registered and wait for the wire's planted-body fixture.

    Verdict: F-BH1 registered, not run; F-BH2 registered, not run; F-BH3 registered, not run; F-BH4 registered, not run; F-BH-H registered, not run

    Limits (LIM-33): No wire carries a generating body's state into the field the engine reads: in the engine that body is a decided pole with no halo and no clock.

    What would close it: The wire built and measured on planted bodies, with its predictions registered before the run.

    Falsifiers: F-BH1, F-BH2, F-BH3, F-BH4, F-BH-H.

    Record: deploy/experiments_status.json, deploy/papers_graph.json.

  9. 9. The Earth: its dipole as a two-state recorder

    measured · registered-not-run

    The axial dipole moment read onto the sphere from PADM2M and IGRF-14: an isomorphism of dynamics, not an identity. P1 (registered before the record was read): at ≥ 7 of 9 reversal boundaries the smallest moment within 20 kyr is at or below the record's tenth percentile; met 7. The null block: a random-window null meets the rule in 19 of 1,000 draws (r = 0.019), so the rule separates the boundaries from random windows (that is the null's power, against random windows only); it does not separate the model from the textbook account, which expects the same weakening at every reversal. The hold is window-fragile (two of four variants fall to 6 of 9). The statuses record reads them together: held at its threshold; window-fragile; consistent with the textbook expectation and not as support.

    The Earth's dipole at its nine reversal boundaries: each window's weakest moment against the tenth percentile 0 1 2 3 4 5 6 7 axial dipole moment (10²² A m²) tenth percentile 3.25 boundary (Ma): younger | older 0.773 Brunhes | Matuyama 0.990 Matuyama | Jaramillo 1.070 Jaramillo | Matuyama 1.180 Matuyama | Cobb Mountain 1.215 Cobb Mountain | Matuyama 1.775 Matuyama | Olduvai 1.934 Olduvai | Matuyama 2.116 Matuyama | Réunion no data: past the record's end 2.140 Réunion | Matuyama no data: past the record's end 7 of 9 met, 7 needed Data: PADM2M (Ziegler et al. 2011, EarthRef.org); GTS2020 (Ogg 2020, via PmagPy) embark.day
    The pre-registered prediction on the Earth's dipole, drawn. At each of the 9 polarity boundaries it names, 0.773 to 2.14 million years ago, the dot is the smallest axial dipole moment within ±20 thousand years of it and the ring is the moment at the boundary itself; the dashed line is the record's tenth percentile, 3.25 × 10²² A m². 7 of 9 windows reach the line and 7 were needed, so the prediction holds at its threshold, with no margin. The 2 oldest boundaries lie past the record's end at 2,000 thousand years and count as misses. It is a textbook expectation (the dipole weakens at a reversal), and no null baseline was registered beside it. Status: measured. Verdict: held at its threshold; window-fragile; consistent with the textbook expectation and not as support. Record: deploy/atlas/earth_boundaries.json, drawn from fixtures/earth/earth_readings_2026_10_04.json. Credit: EarthRef PADM2M (Ziegler, Constable, Johnson and Tauxe 2011), used for research under the source's published terms; written confirmation pending; GTS2020 polarity time scale, Ogg 2020, via PmagPy (BSD-3-Clause).

    Also drawn by Day: S8: each band's largest power against the red-noise line it had to clear; The dipole axis's tip on the sphere, 1900 to 2025, with the 1955 peak; The Earth's 40 gravest modes: observed minus PREM.

    Verdict: S8 refuted; P1 held at its threshold; window-fragile; consistent with the textbook expectation and not as support; F-BH8 registered, not run

    Limits (LIM-20, LIM-21, LIM-23): The Earth's halo has not been measured: reading the magnetosphere as the dipole's halo is the model's reading, it fails a first check (the magnetosphere shrinks as the dipole weakens, the halo grows), and no record here reads it. The retention hypothesis is contradicted at today's escape rates and untested for earlier times: that a body keeps what it prints because a field holds it has no test of its own, and Venus is its standing counterexample. The EarthRef and INTERMAGNET licence confirmations are still to be received: the dipole stack is used for research with its authors credited, and no product sells it.

    What would close it: A recorded reading of the magnetosphere's size or shape set against the dipole's margin over the same years, with the method written down first.

    Falsifiers: EARTH-P1, F-BH5, F-BH8.

    Data: EarthRef PADM2M (Ziegler, Constable, Johnson and Tauxe 2011), used for research under the source's published terms; written confirmation pending; GTS2020 polarity time scale, Ogg 2020, via PmagPy (BSD-3-Clause); IGRF-14, IAGA/NOAA NCEI, Alken et al. 2021; Masters & Widmer 1995, AGU, cited.

    Record: fixtures/earth/earth_readings_2026_10_04.json, deploy/atlas/earth_track.json, deploy/earth_p1_sensitivity.json, fixtures/dualfield/fbh8_halo_widening.json.

  10. 10. The Sun and the microwave sky: read and cited

    measured

    BiSON's low-degree p modes and Planck 2018 TT, read by fixed rules: a reading and a lens calibration that test nothing in the model. The four cosmology and telescope checks are registered and not run; registered 2026-10-07, not run; the data are fetched after the registration and the result is published either way.

    The microwave sky's temperature spectrum from Planck: 83 binned values and three acoustic peaks ℓ 223.4 ℓ 535.2 ℓ 819 0 1,000 2,000 3,000 4,000 5,000 6,000 0 500 1,000 1,500 2,000 2,500 multipole ℓ (an angle on the sky) D_ℓ (µK²) Data: Planck 2018 TT binned R3.01 (ESA; Planck Collaboration 2020, A&A 641, A6). Based on observations obtained with Planck (http://www.esa.int/Planck), an ESA science mission with instruments and contributions directly funded by ESA Member States, NASA, and Canada. embark.day
    The microwave sky's temperature spectrum as Planck measured it: 83 binned values of D_ℓ = ℓ(ℓ+1)C_ℓ/2π in µK² with their ±1σ bars, ℓ from 47.7 to 2,499. The three rings are the acoustic peaks a fixed rule found (a bin above both neighbours, the lowest three in ℓ, each placed at the vertex of the parabola through it and its neighbours): ℓ = 223.357, 535.214 and 818.974. This is a reading of public data by a fixed rule, and it tests nothing in the model. The lens card on the same bins held 6 of 6 of its rows (deploy/lens_cards.json); that calibrates the lens against the published best fit, and it is not a test of the model either. Status: measured. Verdict: none: a fixed-rule reading with no prediction. Record: deploy/atlas/planck_tt.json, drawn from fixtures/spectra/records/universe_planck2018_tt_binned.json (the peaks: fixtures/spectra/universe_2026_10_04.json). Credit: Based on observations obtained with Planck (http://www.esa.int/Planck), an ESA science mission with instruments and contributions directly funded by ESA Member States, NASA, and Canada. Planck Collaboration 2020, 'Planck 2018 results. VI. Cosmological parameters', A&A 641, A6.

    Also drawn by Day: The Sun's sound modes in an échelle diagram, Δν = 135.317 µHz.

    Verdict: COSMO-CHECK-1 (1) registered, not run; COSMO-CHECK-1 (2) registered, not run; F-BH6 registered, not run; F-BH7 registered, not run

    What would close it: Nothing: a reading with no registered prediction

    Falsifiers: COSMO-CHECK-1 (1), COSMO-CHECK-1 (2), F-BH6, F-BH7.

    Data: Based on observations obtained with Planck (http://www.esa.int/Planck), an ESA science mission with instruments and contributions directly funded by ESA Member States, NASA, and Canada. Planck Collaboration 2020, 'Planck 2018 results. VI. Cosmological parameters', A&A 641, A6; BiSON, University of Birmingham, CC BY 4.0.

    Record: fixtures/spectra/sun_2026_10_04.json, fixtures/spectra/universe_2026_10_04.json, deploy/cosmo_register.json.

  11. 11. The elements as a ground-state ladder

    measured-in-model

    Subshell filling as a ground-state ladder, fitted on Z = 1–36 and checked against NIST ASD ground configurations: 82 of 108 match; held out (Z = 37–108), 46 of 72. The registered prediction F4 (unpaired electrons and susceptibility against the halo of the margin) waits for its table.

    Verdict: F4 registered, not run

    Limits (LIM-08, LIM-09, LIM-14): The periodic-table model is fitted on the first 36 elements and misses some of the rest; its misses are reported, not tuned away. Ten answers, meitnerium to oganesson, have no reference: NIST lists no neutral ground level above hassium. The periodic-table model fills subshells and does not calculate atoms: no electron moves between orbitals, and there are no molecules, no absolute energies and no excited states; the filling order is an input.

    What would close it: A fit on the first 36 that also reproduces the others, or a stated reason each miss is out of reach.

    Falsifiers: F4.

    Record: fixtures/spectra/elements_2026_10_04.json, deploy/papers_graph.json.

  12. 12. Hardware: the bridge to a quantum processor

    measured-in-model · registered-not-run

    The vendor-neutral bridge: F-Q1, F-Q1b and F-Q2 ran on the exact simulator; the hardware rows F-Q3 are registered and not run, because the procurement record is unsigned (6 of 9 items open).

    Day’s figure: Twenty simulated pairs: T_xy against T_xx, inside the band the model requires, drawn on embark.day’s Mission page.

    Also drawn by Day: The lens on an exact simulator: witness intervals and the phase fit.

    Verdict: F-Q1 held; F-Q1b held; F-Q2 held; F-Q3 registered, not run

    Limits (LIM-02): No hardware pair has been read: every lens reading comes from an exact simulator.

    What would close it: One pair prepared and read on quantum hardware, with the readout calibrated first.

    Falsifiers: F-Q1, F-Q1b, F-Q2, F-Q3.

    Record: fixtures/qpu_bridge/fq_simulator_results.json.

  13. 13. SEISMO-LENS-1: Day's mission experiment

    measured

    Nine registered calls on four committed events, the failed control first: the one-minute control at FRD exceeded its bound (call iii), so the calibration that every other call leans on failed; the two 1-second nulls held (Indios at SJG, Sand Point at SHU); at Tohoku (HON, 1 minute) F moved in the declared window and the move is unattributed.

    SEISMO-LENS-1: four earthquake windows, the failed one-minute control on top 0.1 1 10 F witness (nT, log scale): dot; bound: tick (iii) Mineral M5.8 FRD 59 km, 1 min definitive (INTERMAGNET) the control: coda window, 600 s floor fails (v) not measurable (i) Indios M6.4 SJG 71.9 km, 1 s a null: P-S window, 240 s floor (declared) holds (i-c) fails (ii-a) Sand Point M7.6 SHU 98.6 km, 1 s a null: P-S window, 240 s floor (declared) holds (ii-b) not measurable; (ii-c) holds (iv) Tohoku M9.1 HON 1 min, 12:00 to 16:00 UTC (INTERMAGNET) the one positive call: 24 of 24 ten-minute windows exceed unattributed verdict Data: USGS Geomagnetism Program (public domain); IRIS/EarthScope. INTERMAGNET 1-minute data (non-commercial, not tracked): no value of it is drawn; those rows show our statistic and its bound only. embark.day
    SEISMO-LENS-1, Day's mission experiment, on its four committed events, the failed control in the top row. Each row is one event's registered call on the scalar field F: the dot is the witness (the change of F's median in the window against the median before the waves arrive) and the tick is the bound it had to exceed, on a log scale in nT. The one-minute control at Mineral (iii) fails: 1.535 nT against a bound of 0.165 nT, where the call said nothing would move; because the control fails, none of these windows carries a field claim without a remote reference. The nulls at Indios (i) and Sand Point (ii-a) hold, read on the declared 240 s floor because no one-second series holds the 600 s the calls asked for. Tohoku (iv): F moved in the declared window (24 of 24 ten-minute windows exceed), unattributed: no sea level or remote reference is in the tree. The other calls: (ii-b) not measurable; (ii-c) holds; (i-c) fails; (v) not measurable; (vi) holds. No value of INTERMAGNET's one-minute series is drawn; the Mineral and Tohoku rows show our statistic and its bound only. Status: measured. Verdict: the one-minute control fails; two nulls hold; Tohoku unattributed. Record: deploy/atlas/seismo_windows.json, drawn from fixtures/seismo/seismo_lens_results_2026_10_04.json. Credit: USGS Geomagnetism Program (public domain); IRIS/EarthScope; INTERMAGNET 1-minute data, used for research under the source's published terms (non-commercial); no value of it is shown.

    Verdict: (iii) falsified; (i) held; (ii-a) held; (iv) unattributed; (i-c) falsified; (ii-b) not measurable; (ii-c) held; (v) not measurable; (vi) held

    What would close it: Closed: falsified as registered; only a new registration, committed before its data, re-opens it

    Falsifiers: (iii), (i), (ii-a), (iv), (i-c), (ii-b), (ii-c), (v), (vi).

    Data: USGS Geomagnetism Program (public domain); IRIS/EarthScope; INTERMAGNET 1-minute data, used for research under the source's published terms (non-commercial); no value of it is shown.

    Record: fixtures/seismo/seismo_lens_results_2026_10_04.json.

  14. 14. The Hilbert library

    measured-in-model

    99 spaces × 2^10 = 1,024 states each, exhaustively enumerable; the dense NO (one space for the library aligns in sample only: held-out agreement collapses for text and vision); nearest-space-first recall as the paper's one Hilbert sentence.

    Verdict: HILB-2-CONDENSE-KEPT falsified; HILB-1-DENSE-TEXT held; HILB-1-DENSE-VISION held; LIB-CENSUS-SPACES held; LIB-CENSUS-WIDTH held; NEAREST-FIRST held

    Limits (LIM-06): Timings and recall figures are one fixture on one processor, and no graphics-card figure has been read.

    What would close it: The same fixtures read on a graphics card and on a second machine, with the load recorded.

    Falsifiers: HILB-2-CONDENSE-KEPT, HILB-1-DENSE-TEXT, HILB-1-DENSE-VISION.

    Record: deploy/model_spaces/*.json, docs/HILBERT_SPECTRUM_TRANSFER_2026_09_17.md, docs/HILBERT_LIBRARY_2026_09_16.md, docs/HOPE_RECALL_RECONCILIATION_2026_09_15.md.

  15. 15. The working hypothesis

    hypothesis

    The working hypothesis, quoted in the legend above and set out below with its falsifiers, losses first; registered, never a result.

    Verdict: F-BH5 refuted; EARTH-P1 held at its threshold; window-fragile; consistent with the textbook expectation and not as support; F-BH8 registered, not run; F-BH1 registered, not run; F-BH2 registered, not run; F-BH3 registered, not run; F-BH4 registered, not run; COSMO-CHECK-1 (1) registered, not run; COSMO-CHECK-1 (2) registered, not run; F-BH6 registered, not run; F-BH7 registered, not run; LITHIUM-7 no verdict

    What would close it: The working hypothesis has no registered falsifier of its own; until one is registered, the falsifiers listed with it are the only places it can fail.

    Falsifiers: F-BH5, EARTH-P1, F-BH8, F-BH1, F-BH2, F-BH3, F-BH4, COSMO-CHECK-1 (1), COSMO-CHECK-1 (2), F-BH6, F-BH7, LITHIUM-7.

    Record: deploy/experiments_status.json, fixtures/earth/earth_readings_2026_10_04.json, fixtures/dualfield/fbh8_halo_widening.json, fixtures/em_drive/planted_holes.json, fixtures/dualfield/fbh2_step_response.json, fixtures/dualfield/fbh3_clock_rate.json, deploy/cosmo_register.json, docs/EVIDENCE_THE_EXTENDED_PERIODIC_TABLE_2026_10_04.md.

Our working hypothesis and its falsifiers

Every test of it is registered before its data is read, and every verdict on those tests is published below, the misses included.

What it is not

Nothing observes the inducing process directly: it is estimated from the correlations a record carries, with the entropy of the parts as the measured limit. The mathematics of the pair is standard two-qubit quantum mechanics, nothing invented here; what is ours is the three-referential reading and its use as an engine model.

What would count against it

A registered Earth, planted-body or cosmology row refuted at its threshold is a result against the working hypothesis as a whole. Said first: F-BH5 is measured: refuted (read from fixtures/earth/earth_readings_2026_10_04.json#s8_orbital_periods). Its registered consequence: the link above the Earth is a lineage, not a generation.

The falsifiers it has, losses first

The papers and their registered tests

Eight papers; thirteen predictions registered before their data. Two have run (one held at its threshold, one refuted) and eleven wait for their records.

The Operational Unit: an actor–agent model of information economies, and what our engine runs of it today

It puts forward. One description that would fit a program that learns, a person and a physical system alike. The paper calls it a hypothesis and not a finding.

The test it registered. The six properties it says such a system must keep, measured on a person and on a physical system, with the method written down first. A system that fits the description and breaks one of them counts against it.

Its registered predictions: F7 registered, not run.

The periodic table as a ground-state ladder

It puts forward. Each element read as the lowest-energy filling of its electrons, a model with a fitted table, checked element by element against the published reference.

The test it registered. An element's count of unpaired electrons and its measured magnetic susceptibility should rise with the halo of its margin on the ladder. If they do not, the reading of the margin is wrong.

Its registered predictions: F4 registered, not run.

Closing the gaps: what a scheduling engine can and cannot do on four sustainability problems

It puts forward. That an engine which schedules can keep what already exists from being run badly, and cannot close a technology gap on its own.

The test it registered. Each example's answer is checked against every possible answer wherever the problem is small enough. The open test is a customer's own tables in place of the illustrative ones three of the four examples use.

Its registered predictions: none registered yet in the papers' graph.

Sense, remember, act: what the engine keeps

It puts forward. What the engine senses, what it remembers and what it learns, with the memory's failures listed first.

The test it registered. Two checks on the engine's own records: that each region's entropy, taken alone, equals the coin entropy of its pole, and that a forming decision moves the lock of its halo with the sign the model fixes.

Its registered predictions: F-BH1 registered, not run; F-BH4 registered, not run; F-BH-H registered, not run; F1 registered, not run; F3 registered, not run.

The halo of a decision: a two-way choice, its partner, and the event read as setting them

It puts forward. A two-way choice, the partner it is tied to and the event that tied them, in textbook terms.

The test it registered. A pair driven on a clock has to show a correlation around its equator somewhere on its clocked record. If no clocked pair ever does, the turning side of the reading is empty.

Its registered predictions: F2 registered, not run; F-BH2 registered, not run; F-BH3 registered, not run.

The Earth as a two-state recorder: its dipole on the Bloch sphere, the rock that keeps the record, and the clocks inside

It puts forward. The Earth's dipole read onto the same sphere from public records: its sign and strength become one number between minus one and one, a pole when the field is strong and the middle when it is weak. The paper calls the mapping an isomorphism of dynamics, not an identity: the Earth is not a qubit.

The test it registered. A prediction written before the record was read: at no fewer than seven of the nine reversal boundaries of the last 2.2 million years, the smallest moment within twenty thousand years of the boundary is at or below the record's tenth percentile. One more miss than that refutes it.

Its registered predictions: P1 held at its threshold; window-fragile; consistent with the textbook expectation and not as support; S8 refuted; fbh8 registered, not run.

HOPE: recall in a library of Hilbert spaces, what it finds and where it stops

It puts forward. How the engine remembers: a library of small spaces, one for each model it has read, where recall is settling to the nearest stored state. The paper says the library is many small spaces and never one large one.

The test it registered. None registered yet, and the paper says so. What would count against what it measured: a reading's own space named first no more often than chance on the same test, or one fit that carries the spaces' agreement to readings it has not seen.

Its registered predictions: none registered yet in the papers' graph.

The unit of work: a receipt for a settled problem, and what it is not

It puts forward. What one unit of computed work is: a receipt for one settled problem, booked on a hash-chained ledger, never traded and not a currency, an investment or a commodity. The losses come first: the ledger reads one moving number where six economies were expected.

The test it registered. Sixteen tests registered in the paper's own registry before any script that runs them, none run yet. What would count against the receipt: a retired receipt accepted a second time, or a reduction that does not rise with the energy on the instances in the tree.

Its registered predictions: none registered yet in the papers' graph.

What is still open, across the papers

Every paper ends with what it does not say, and one ledger holds each of those limits once. Each object above carries its own; the whole ledger, with what would close each limit, is on the research hub, embark.day/research.

Every paper carries the same credit line: Jonathan Klinshaw, Agile Integrations — author, concept and rulings · Day — the figures, the computed records and the measurements · Claude (Anthropic) — engineering consultant of record: design and build.

Where to read more

The papers are on our notes and papers shelf. The research hub on embark.day, embark.day/research, sets out each paper’s numbers by kind and what is to be measured next; the record, week by week, lists what was read and registered.

Read the papers What we sell