In one day the remaining agenda was computed, each item frozen in advance. Turned out law: the spacetime dimension (3,1) (a singleton under the observer gates) and complex quantum theory (from the composition axiom, with the Renou experiments as gate). Turned out address, with a price: the generation number (the cheapest valid observer address has one generation; our three costs exactly 10 bits) and the cosmological constant (3.5-9 bits, band over both literature windows). The dynamical cost term demonstrably cannot supply those bits (the naive reading is refuted by our own existence; renormalised it is flat), the arrow of time is formalised as a measure preference for compressible initial conditions, and the framework's total address ledger stands at ~18-25 bits. The whole site was made consistent in the same movement: the old "K gives three generations" claim is retracted everywhere and replaced by the stronger selection result.

Axiom A3 (continuity) is proven redundant (the codomain suffices, via Ostrowski): the system compresses from three axioms to two. The surprise ledger turns out to be Bayes' theorem with the universal prior. And the same Cauchy-Ostrowski engine fires a third time: the Born rule P = |ψ|² follows from composition plus branching consistency, after which the last assumption (no-drift) dissolved into timelessness, additivity and decoherence: α = 2 as the unique meeting point of measure additivity and orthogonal geometry. Open problem 11.2 (zero-point dynamics) is dissolved, conjecture 11.7 upgraded with a conditional theorem, and the observer gates are reduced to one self-validating core (asking the question is the proof), with the only back door (nuclear recorders) triple-locked: formation (Rees-Ostriker cooling), thermodynamics, and the measure.

The scale was applied to mathematics itself: the continuum hypothesis as the α of mathematics, the history of the axiom of choice as a matching Bayesian retrodiction, and the first measured axiom price list ever (mathlib, 439,764 theorems: choice 70.2%, genuine witness extraction at most 18.7%, constructive core 7.4%), plus a calibrated weighing of Ultimate-L against Martin's Maximum with a frozen flip trigger. And the prediction register was extended: the 0νββ decision tree is fixed before JUNO speaks, the anarchy reading stakes normal ordering (95-99.9%), a mass sum near 0.06 eV and m_ββ of 1-10 meV, and six new frozen stakes cover collapse-free quantum mechanics, quantum computers without a fundamental limit, constant dark energy, no primordial tensor modes, a large CP phase, and the stability of the price list. All with failure conditions attached. And the computational programme has been written up as a scientific paper, published as its own page on this site (/artikel, NL and EN): the site is the document, hence also the publication.

The two primal sentences of 2020 turn out to be two floors. "Nothing does not exist" is a-local: a statement about the existence layer, without place or moment. "Everything is based on effortless effect" is local: the weighing works per step, on every floor where steps exist (paths inside this universe, branches of the quantum state, members of the catalogue). The distinction is now explicit on the philosophy page (new card in the Meta-Law section) and carried through with short pointers on home, the mathematics-in-words, the physics page and the origin story.

Four visuals said something different from the text beside them and have been rebuilt. The zero-point-attractor animation (philosophy page and the standalone demo) now has one traveller taking a different path each round, with the smoothest path visibly carrying the most weight; the jump back has become the message (arriving is coming home: both points pulse together, because it is the same point). The alpha-running animation on the physics page now honestly shows the miss: the line ends at the predicted 152.2 with the measured 137.04 beside it as the missed target (F2 prime). In the Three Modes of Existence figure the actual point gained its indexical reading (pointed from within, not an objective lamp), and in the differentiation figure the non-selected branch now visibly carries less weight.

Three new structure stakes registered, each with a failure condition: no proton decay (Hyper-Kamiokande; one confirmed decay breaks the entire selection chain), no light sterile neutrinos, and no Lorentz violation. The stake "δ_CP is large" has been sharpened into numbers, frozen before DUNE and Hyper-K speak: P(|sin δ_CP| > 0.5) = 66.7%, median |J| = 0.024, and |sin δ_CP| < 0.25 will count as a miss. The verification page also gained an expectation calendar 2026-2030: per experiment what EEML says and what the other outcome means. The same night the root diagnosis of our four misses (level confusion: the a-local measure in a local chair) was turned into a meta-stake: no first-principles derivation of a continuous constant will ever succeed, by anyone; one accepted derivation of α falsifies the address reading. The meta-stake and the new structure stakes have also been folded into the paper itself (Secs. 9-11). Following on, the delegation stake was registered: local fixed-point consistency (asymptotic safety) bounds address coordinates rather than fixing them; for α the window shrinks from below (α⁻¹ ≳ 75-90, a gain of 0.15-0.27 bits), the strong form is explicitly closed, and delegation stake and meta-stake cover each other's flanks; both are also folded into the paper (Sec. 9) and the status page.

The registration has also been made externally verifiable. SHA-256 fingerprints of the frozen registers as they stand online today (independently checkable against the archived snapshots at the Internet Archive):

toetsing.content     a1d6ac5ca468b2371f54d8dbb8f559bf8a95ee4aeaa6f7ac450e9ccc14de1a20
toetsing-en.content  bd28fcbd6d443ca6946028a77b0bda5912c02b759a9e1522bde4f91c5cab2720
artikel.content      d034b8334bf387d394b74846bda9fbd77372d89b7417405409d7bdb01eb71296
artikel-en.content   6e0b73d597a50d25412661142674afb2f6e5bc2492e32d66bbcef2b2d8ac8fe9

The register has been extended with six new frozen stakes: f_NL consistent with zero (SPHEREx), no new physics at the HL-LHC, the Hubble tension resolving without new physics (the boldest), the pulsar-timing background being astrophysical (IPTA), the neutron EDM staying null (n2EDM), and the tool-independence of the axiom price list (Rocq/Coq cross-measurement). The axion stake now has numbers: the bit accounting yields a preferred band m_a ≈ 2-20 µeV (ADMX/MADMAX territory), with the post-inflationary scenario noted in advance as the other branch (~25-500 µeV). Three retrodictions are booked as such, outside the track record: the LHC desert, Higgs near-criticality and the existing f_NL limits. The expectation calendar gained six rows and the verification-page hashes have been refreshed.

The tool-independence stake was measured within hours of registration and honestly fell. Coq 8.20.1 standard library: 16,664 lemmas, 11.5% with a logical axiom (88.5% constructive); Mathematical Components 2.3.0: 11,442 lemmas, exactly 0% axiom use; mathlib: 70.2%. Axiom prices are thus to a large degree proof-style and tool properties, and mathlib's 70.2% is an upper bound within that style, not a property of mathematics. The miss is booked in the track record (the second, alongside Zipf). Labelled separately as a post-hoc hypothesis: in both ecosystems axiom use concentrates at the continuum (in Coq: LPO-like axioms and functional extensionality around the real numbers; genuine choice 0.03%), while discrete mathematics turns out axiom-free: the countable-cheap, continuum-expensive pattern again. Method and script archived; verification-page hashes refreshed.

Four deepenings in one round. One: the weight parameters of the decomposition lemma are forced: α_K = ln 2 (Levin maximality) and α_I = ½ (Gaussian integral): zero free weight parameters left, conditions named (mathematics page). Two: the joint anarchy Monte Carlo (2 million draws, frozen seed) cross-validates the separate neutrino stakes (P(NO) = 95.7%; mass-sum median 60.6 meV; |sin δ| reproduces Haar to 0.2 points) and delivers one honest revision before the data: m_ββ shifts from "1-10 meV" to a median of 15 meV with 68% [6, 30] meV, making a next-generation detection more likely. Three: new stakes: the top mass converges to the criticality window ~171.0-171.5 GeV (delegation consequence, current tension acknowledged), no ringdown echoes, no fifth force, the birefringence hint fades, c_gw = c; plus two retrodictions (g-2, GW170817) outside the track record. Four: an honesty note on the physics page: the promised Casimir-ratio table requires a mass-model specification that does not yet exist, so no table is claimed; and the formalisation project (machine-verifying the core theorem in Lean) is opened as a research goal.

The axiom measurements (mathlib, Coq standard library, Mathematical Components) are now online as a standalone page, with method, numbers and interpretation: intended as a contribution to the proof-assistant community, usable independently of EEML. Reachable via /axiom-audit and from the mathematics page.

The retracted LISA frequency gets a worthy successor that is not a frequency. At a generic moduli point nothing forces a ratio (the point is address), but the cheap-address weighting favours symmetric points, and at the frozen Fermat point the 1536-element symmetry group forces degeneracy multiplets: axions [12, 6, 2, 1, 1], Kähler moduli [12, 6, 1, 1], together at most 9 distinct lines instead of 42. Potential-independent, protocol and seed frozen before computing, genericity cross-check passed. The old Casimir line on the physics page has been rewritten (it was posed wrongly), the calendar gained a LISA row, and the failure condition is included: dozens of separate lines without multiplets makes the address weighting lose. Following on, the full four-branch decision tree has been frozen, after the example of the guillotine tree: full win (the address becomes measurable), point miss (compute the group back), flat-measure fallback (with mandatory recomputations of the generation bits and axion window), and undecided.

Four interventions for findability. One: the navigation now has a "More" menu making every page reachable (Mathematics in Words, Chain, Origin, The Paper, Axiom audit, About, Contact, Changelog were partly absent from the menu); the mobile menu now shows everything. Two: every long page automatically gets a table of contents at the top that jumps to its sections. Three: the home page opens with "Pick your route": three entrances for three kinds of readers (no formulas, the derivations, the evidence). Four: the verification page starts with an "at a glance" summary of the register, the track record and the next referee.

The card "HERE vs the Multiverse" (philosophy) and FAQ Q4 (verification) have been sharpened with three points from the local/a-local line: the bookcase-versus-table-of-contents image (the bubble multiverse is local, the catalogue is a-local, so "where are those universes" is a category error), the reversed burden of proof (the other members are not the extra assumption, the lamp is, and it is refuted), and chance as the local face of the a-local weight (Born over branches, typicality over members).

The repaired successor of the missed tool-independence stake was frozen and measured the same day: mathcomp-analysis (plus classical and reals, 5,887 lemmas, full count) uses a logical axiom in 77.2% of its lemmas, against 0.0% for the discrete work of the same community. A hit on the frozen threshold (> 50%); the directional expectation of > 90% was not reached and that is stated. The mix was exactly as predicted: extensionality plus choice (69.2%), no LPO. Continuum concentration is thus no longer a post-hoc observation but a hit stake; the axiom-audit page has the new row.

The framework's main theorem (composition plus codomain force P = e⁻λˢ) has been formalised and accepted in Lean 4 against mathlib: zero errors. The formalisation nails down two things that slip easily on paper: monotonicity need not be assumed (it follows from the codomain), and the proof uses exactly Lean's standard axioms (propext, choice, quotients), the same trio the price list measures. The Ostrowski part (additive plus non-negative implies linear) is formalised from scratch via rational scaling and a density sandwich. Proof and log archived; a note added to Theorem 3.1 on the mathematics page; the research goal on the status page is ticked off.

The last major formal gap on the status page has been closed. The orthogonality of the decomposition lemma is now a theorem in three parts (flow ⊥ gauge everywhere; critical orbits stay critical; the Hessian vanishes on gauge directions), each formalised and accepted in Lean 4, in a single compile round, on the standard axiom trio. This proves that A, K and I count no directions twice. Honestly marked what stays open: the identification of the three blocks with S_dyn, K and I remains motivated, not a theorem. New card at the lemma (Theorem 4.2), badge updated, status page ticked off, proof archived.

The last open research goal has been tackled along the only route the meta-stake allows: not deriving the values, but measuring the price. Under the frozen anarchy conventions the observed quark hierarchy costs ~95 bits as bare address, against 39-49 bits for a Froggatt-Nielsen-like mechanism: a 46-56 bit margin, three times the axion margin, with the random-matrix tail exponent (2.0) as internal cross-check and the correlation caveat honestly stated. Reading: quarks structural, leptons anarchic, now in one currency. The K3 texture stays honestly open (bundle choice unspecified, no texture claimed). Status page updated; verification-page hashes refreshed.

Three completions. One: all research work and the site are now under version control with a dated offsite snapshot (83 MB, with check hash). Two: the full Froggatt-Nielsen search space has been traversed: the CKM angles pin qQ = (3,2,0) uniquely and the cheapest texture is the classic textbook texture, found rather than chosen, with parameter-free order relations as by-catch (addendum to the Yukawa card). Three: the own 1-loop computation of Higgs criticality confirms the decidability of the top-mass stake (slope ~0.006 per GeV) and honestly books that the window itself is loop-sensitive: the stake stays on the 2/3-loop literature band.

The paper is now also available as a print-ready PDF preprint (7 A4 pages), generated from the canonical paper page, with an appendix on the living register, the two Lean verifications and the timestamp machinery including the register hash. Found via the paper page or directly at /eeml-preprint.pdf. The site remains the publication of record; the PDF is the distributable snapshot.

The preprint has been deposited at Zenodo (CERN) under DOI 10.5281/zenodo.21908485, CC BY 4.0, with references to the canonical paper page, the register and the axiom audit. The prediction programme thereby has a timestamp fully independent of this site: whoever later wants to check that the stakes predated the measurements need not take our word for it.

The honeycomb question ("does stability force the point α?") has been computed with a criterion frozen in advance. Own edges: hydrogen stability sharpens the lower edge from 59 to α⁻¹ ≳ 77 (lattice band 65-86), the carbon candidate drops out, the cooling upper edge stays soft around 685. The window remains a hall (width factor 8.9; 137 at log position 0.26, nowhere edge-pinned): the needle reading loses, the address reading stands, gain 0.16 bits. Most striking by-catch: the anthropic lower edge and the independent fixed-point lower edge (75-90) land in the same place. Verification-page hashes refreshed.

Habitability within the α window has been mapped as far as published sensitivities reach. The carbon island (Hoyle, ±4%) spans α⁻¹ ∈ [131.8, 142.7]; the claim "136 burns, 138 freezes" is thereby refuted (both sit well inside the island). The island address costs 4.8 bits, independently consistent with the ~5 bits of carbon conditioning already booked, and 137.036 sits at log position 0.49: exactly mid-island, no boundary. Whether other islands exist elsewhere in the hall remains an open ab-initio computational question, with no claim.

The open ab-initio question has been tackled at cluster-model level, with a validation gate frozen in advance that passed (d = 6.9 fm reproduces the published ±4% sensitivity). Outcome: next to the Hoyle island, a continent opens at α⁻¹ ≳ 146: bound Be-8 removes the mass-8 bottleneck and 74% of the hall becomes habitable. The ~5-bit figure now appears three times independently (conditioning, island-in-hall, island-in-habitable): the interpretation shifts from habitability price to address choice. Caveats frozen (stellar behaviour with bound Be-8, O/C balance); real lattice EFT remains the referee.

The law-address triage has been run over physics' famous open riddles, with frozen priors. Three new outcomes: the hierarchy problem costs only 5 bits under the log prior (the 10³⁴ drama is a linear-prior artifact, and the HL-LHC desert stake becomes retroactively logical), the Koide relation is honestly deflated to undecided (14 bits of coincidence against 10-20 bits of mechanism, with pole-mass sensitivity as a demerit), and the why-now coincidence turns out to be 1.7 bits: a shrug. Plus the Gatto receipt (0.2% agreement, exactly the found texture). The full five-column table (law, address, mechanism, undecided, grave) is in the register.

The consolidated address ledger now sits on the home page as a postcard: the five address items with their bits, totalling ~25-32 bits: the entire remaining "why" of the observable universe in less information than one line of text, with "from: no one; there is no outside". In the same move the outdated final-position paragraph on the home page was updated (it still promised an α computation retracted in v0.19).

The site has gained a light/dark switch (the ◐ button next to the language toggle, also in the mobile menu). The choice is remembered, loads without flashing, and the full palette plus all components (cards, code, badges, navigation, table of contents, the postcard) have been retuned for the light theme. Dark remains the default.

The light theme is now the default (dark stays one click away via the ◐ button, and an earlier choice is remembered). In addition, single-line formulas are now shown as book typesetting: centred, in a serif mathematics face, without a frame: while the underlying text remains exactly the same Unicode and thus fully selectable and copyable. Multi-line chain and code blocks deliberately keep the monospace style. Same-day follow-up: the last hard-coded dark surfaces in the content (the postcard background, the freedom-determinism box and the HERE diagram colours on the philosophy page) have been converted to theme variables, so the light theme carries through everywhere.

The improved spectral-action matching has been carried out: 2-loop running from the unification boundary condition 5g₁² = 3g₂² = 3g₃² gives α⁻¹(m_e) = 152.2 against a measured 137.04. That is an 11% deviation, outside the framework's own criterion of ~10%. F2′ is thereby triggered and the promise "α from EEML" is retracted. Not weakened, not reformulated: retracted, and the derivation joins the earlier α formulas in this graveyard.

Two further independent routes to α were subsequently tested under pre-registered criteria and failed as well: the selection route (the cheapest consistent finite geometry does not yield the observed forces, under all three frozen complexity measures) and the weight-function route (EEML's own e⁻ˢ as cutoff function predicts the unification strength a factor 9 off). Three falsifications, each with criteria frozen before computing. The pattern: countable structure is derivable, continuous strengths are address. See the testing page, section 2.

What the computational programme delivered after the falsifications: with observer conditions frozen in advance (an electron must be able to exist, charged nuclei, the observed forces reachable) and the measurements as referee (neutrino mass, the bottom-tau ratio), exactly one candidate remains within the explored universe of finite geometries: the standard model with right-handed neutrinos. The hypercharges come out of the consistency conditions themselves; a cheaper Pati-Salam challenger is eliminated by the measured bottom and tau masses via a theorem from the exact structural conditions.

The chain carries one guillotine prediction: neutrinos are Majorana particles; definitive exclusion of neutrinoless double beta decay would bring everything down. The full working documents and ten scripts (phases 8-17 of the programme) are available. Updated: testing page (new section 2, scores, FAQ Q13, track record) and physics page (F2′ status).

Four follow-up tests, each frozen in advance (phases 18-21). Generations: the cheapest valid observer address is one generation plus two ν_R (leptogenesis); our three-generation address is exactly 10 bits more expensive: the number 3 is address with a price tag, not a derivation. S_dyn: the dynamical cost term cannot supply those 10 bits and does not change the ranking (the naive vacuum reading is refuted by our own existence); the complexity ranking is the complete selection. Robustness: the uniqueness of SM+ν_R holds in the n=5 universe (SU(5) flank: 136 algebras, all eliminated via the forced bottom-tau equality plus proton decay) and against four building blocks (analytic cost bound). Typicality: α sits at log position 0.34 in the habitability window from the literature: typical. Total address price ~15 bits. New on the testing page: the card "The address ledger" in section 2. Also updated: the mathematics page, where the old §12 line "K-minimum + I-penalty → N_gen = 3" has been retracted (too strong; refuted by the framework's own computations) and replaced by the stronger, actually derived selection result (SM + ν_R unique, Theorem S′) plus the S_dyn status in §5. Finally a site-wide consistency scan: home (cards 2 and 3), physics (TL;DR, chapter 5 status revision, §6), status (α card, green list, scores), chain (links 11 and 12), philosophy (selection-mechanism card) and mathematics-in-words (new section 12 "Law or address") have all been brought up to the August 2026 state; the old "N = 3 strongly derived" and "number open" phrasings are replaced or struck through with a date everywhere.

Six new phases (22-27), each frozen in advance. Law: dimension (3,1) singleton under observer gates; complex quantum theory from the composition axiom (Renou et al. 2021 as experimental gate). Predictions: no WIMPs; and under the ratified combined ledger (K-bits and typicality bits as one currency, after an honestly documented conflict between two same-day registrations) a θ-solving axion-like sector as dark-matter candidate. Address: cosmological constant 3.5-9 bits (band, edge case). Motivated: the arrow of time as measure preference. The testing page (address-ledger card) has been extended; the full balance is in the working documents.

The 0νββ decision tree has been frozen before JUNO measures the mass ordering (detection = confirmed; emptied inverted-ordering band = refuted; funnel = honestly undecidable, declared in advance including the cosmological risk). And via the address reading (= the anarchy scenario, with the 2012 θ₁₃ success as retrodiction) the framework predicts its own verdict: normal ordering (95-99.9%), mass sum ~0.06 eV, m_ββ small but nonzero (~1-10 meV). Two near-term tests are thereby fixed before the data arrives. New card on the testing page, section 2.

Six items on the core. Two new theorems on the mathematics page: axiom A3 (continuity) turns out redundant (the codomain [0,1] suffices via Ostrowski; the system compresses from three axioms to two: the law itself became simpler) and the surprise ledger turns out to be Bayes' theorem with the universal prior (K = prior, typicality = likelihood), giving the axion reasoning a theorem beneath it. Open problem 11.2 (zero-point dynamics) is dissolved into a category error plus the arrow of time; conjecture 11.7 is upgraded with a conditional theorem (self-representation is available in every observer member; only its use remains open). Furthermore: L2 sharpened, and a two-line Boltzmann-brain defence noted (vacuum metastability of the selected F plus compression suppression), as defence, not theorem. Working document: kernwerk-12aug. Encore: the ur-quote on the origin page received three formal mirrors (tree root, 0/0 origin, weight-information duality) as a new Anno-nu annotation. Fifth addendum: the scale has been applied one floor up, to mathematics itself: independent statements as the address facts of the set-theoretic multiverse (CH as the α of mathematics), the history of the axiom of choice as a matching Bayesian retrodiction, and the first measured axiom price list ever (mathlib, 439,764 theorems: choice 70.2%, genuine witness extraction at most 18.7%, constructive core 7.4%), plus a calibrated, forward-registered weighing of Ultimate-L against Martin's Maximum. New card on the mathematics page, section 12. Sixth addendum: the Born rule from the same weighing engine: composition (A2) plus the codomain forces, via Ostrowski, a power law in the amplitude, and weight conservation under branching forces the power 2: P = |ψ|² as the third output of the same two axioms (after e⁻ˢ and the complex numbers), with the conditions (decoherence for phase independence and preferred basis; no-drift assumption) honestly named. New card on the mathematics page, section 10. And the same evening the last remaining assumption (no-drift) was dissolved into timelessness + additivity + the already-named decoherence, whereby α = 2 turns out to be the unique point where the additivity of the measure and the geometry of orthogonal branches coincide. Seventh addendum: the observer gates grounded: three gates reduced to one abstract gate (stable rewritable self-representation), from which electron and nuclei follow as the minimal implementation; the residual assumption is performatively self-validating (asking the question is the proof); one loophole honestly named (gravitationally bound nuclear recorders), and locked the same evening with three independent locks: formation (without a light charged particle gas cools ~10⁶× more slowly, Rees-Ostriker: the loophole world cannot build its substrate), thermodynamics (isolated shrinking energy budget: the memory freezes) and the measure (chemistry is orders more abundant: typicality picks the abundant class). New card on the testing page, section 2; L4 sharpened. Eighth addendum: the prediction register extended with six frozen stakes: no objective collapse, no fundamental limit to quantum computers (against the superdeterminism camp), dark energy constant (the DESI w anomaly fades), no primordial tensor modes within reach (the boldest: the smooth start as measure preference makes an inflaton unnecessary), CP phase generically large (DUNE), and the stability of the axiom price list. New card, testing section 2.

The heat-kernel coefficient a₆ on the Fermat K3 has been determined for the first time — not via the contaminable spectral fit, but via local curvature integrals: the exact Gilkey formula reduces on closed Ricci-flat manifolds (with a self-derived Bianchi identity for ∫|∇Riem|²) to pointwise cubic invariants. The curvature engine is validated with exact topological calibrations (Gauss-Bonnet/c₂ = 24.00 on four different metrics; Ricci embedding anchor; the fact that ∫|∇Riem|² must be positive caught a global sign error along the way). Result, stable over three independent seeds and three resolution levels:

a₆(Fermat K3, scalar, Vol = 1) = −0.71 ± 0.02

Small and negative: the definitive confirmation that the old α window (~10³–10⁴) never stood a chance. With the true a₆, ζ'(0) is refined to +1.0 ± 0.1 (the earlier spectral a₆ fit of ~+12 proved to be undercounting contamination; the bare plateau was nearly correct). By-catch: first direct Ricci-flatness measurement of the Donaldson metrics (∫|Ric|²: 109 → 36 → 17.5 over k = 3/4/5).

And the discipline also worked internally: a working-document hypothesis two days old (volume stabilisation of the internal space via the a₀/a₆ balance in the spectral action) required a₆ > 0 and was felled by this measurement before it reached the site. As it should be: one's own wrecking ball first.

The last spectral caveat at F2′ (the spectral action uses Dirac, not the scalar Laplacian) turns out to be a theorem on K3: via the trivial canonical bundle, the parallel antiholomorphic 2-form (hyperkähler), Hodge decomposition with b₁ = 0, and the Kähler identity D² = Δ_d, the full Dirac spectrum is the scalar spectrum with multiplicity 4 (plus 2 zero modes; index check 2 = −σ/8, Atiyah-Singer ✓). Exact: ζ'_D²(0) = 4·ζ'(0) = +4.0 ± 0.4 and ζ_D²(0) = −44/15. The spectral sector is completely closed; F2′ now rests on C_gauge with no spectral caveat whatsoever.

The a₆ programme (open problem 6.1 / criterion F2) has been actually carried out for the first time: Donaldson-balanced metrics on the Fermat quartic K3 plus the BBDO spectral method delivered the first ~200 eigenvalues of the scalar Laplacian, triply validated (volume exactly at 8π² within 0.1%; zero mode ~0; symmetry multiplets 12/3/3/3/18 with k-convergence over three independent runs). Lowest non-trivial eigenvalue: λ₁ ≈ 43.8 (12-fold degenerate, Vol = 1). As far as known, the first computed eigenvalues on this surface.

The zeta result: ζ'_K3(0) = +1.0 (interval −0.3 to +2.4) at the frozen convention Vol = 1, with exact internal calibration ζ(0) = −11/15. That is order 1, not the order 10³ to 10⁴ that criterion F2 posed as its window: three orders off, robust against all tested systematics and not repairable by rescaling. The burden of α⁻¹ ≈ 137 thereby shifts entirely to the matching constant C_gauge.

Confirmed and revised. The computation was pushed to k=6 (basis 5476, ~340 modes): ζ'(0) firms up to a tight plateau around +0.9, confirmed across four resolution levels, with λ₁ monotonically 44.19 → 43.52. The convergence does not move toward the window — order 1 becomes more certain, not less.

Decision: F2 reformulated, not retracted. The old window (ζ' ~ 10³–10⁴) is dropped — refuted by the framework's own computation. But the α structure (spectral action on K3) stands; now that the spectral term is demonstrably O(1), the entire burden of "137" shifts to the matching constant C_gauge. The new, narrower criterion F2′: a first-principles computation of C_gauge that does not yield the required value refutes the derivation. Follow-up computation begun: the Dirac operator (last caveat; via Lichnerowicz on Ricci-flat the (0,0) sector is already the computed scalar spectrum). Updated at 6.1, F2, the self-assessment (constant derivations 6→5/10) and the Testing page. Exactly what falsification criteria are written down in advance for: the window fell, the theory got sharper.

The ur-document of ±2020 (cups, lighters, bananas, "everything, yes everything") is now on the site in full and unedited, as an artifact with annotations from today: what survived (the on-ramp ladder, the zero-point embryo "nothing does not exist, therefore it must be a point of everything", the tone), what fell and why (objects-as-evidence became the lens; the flat everything-claim became the layer discipline), and the timeline from intuition to measurement. The changelog ethos applied to the history of the idea itself. Reachable via /ontstaan and via the About page.

Measured. The spectral description cost of the Fermat K3 has been measured for the first time from the framework's own computed spectrum: the symmetry multiplets compress the first ~440 modes from ~7000 bits (generic surface) to ~400 to 1400 bits; T⁴ compresses to a ~10-bit formula. The T⁴ paradox confirms the elimination structure: more compressible than K3 and still hard-eliminated — stability precedes simplicity. Concentration test: within the mini-landscape K3 carries a weight share of 1 − 10⁻²⁷¹⁵ — "everything exists but one dominates" now has a number.

Corrected. The Def 6.1 example counted K_moduli(K3) with h^1,1 = 20 (997 bits), contradicting the framework's own definition and the K3 reference card (dim 𝓜 = 58): revised to 2890 bits, K3 total 1947 → 3840 bits. And the weight ratio stood as e^{−ΔK/ln 2}; correct is 2^−ΔK. The qualitative selection is unchanged; the numbers are — and that is noted here as it should be.

"Selecting and eliminating". There is no separate elimination principle: it is the same slope from two sides. Hard elimination (the unstable does not exist — no membership of 𝔄) and soft elimination (the expensive dilutes exponentially — small weight in μ), with the asymmetry that selecting seems expensive but eliminating is free. That is why nature can "choose" without anything that chooses.

"The detour walker". Why you cannot defy the law: physically (every step follows least action), structurally (resistance is not distance; human weights = motivation/goals/habits, and a habit is a worn riverbed), in principle (preference not prohibition). The attempt to step outside the law is itself an instance of it. With the honest boundary that the after-the-fact reading is always true per event (hence cheap) but becomes testable across cases — and explicitly marked as interpretive lens, not testable core.