BD × AI LAB · RESEARCH EDITION 0.4 · 17 SEPTEMBER 2026
What if compression is not only about describing a world more briefly, but about finding the structure that helps us act within it?
Kaj če kompresija ni samo krajši opis sveta, ampak iskanje strukture, ki nam pomaga v njem bolje delovati?
The world unfolds consequences. Understanding seeks the descriptions that connect them.
Svet razvija posledice. Razumevanje išče opise, ki jih povezujejo.
These are the original intuitions, kept visible rather than rewritten after the fact. Version 0.4 does not claim that any of them is already a law or theorem. It turns each into a research contract with explicit obligations.
To so izvirne intuicije, ohranjene vidne namesto naknadno prepisane. Različica 0.4 ne trdi, da je katera od njih že zakon ali izrek. Vsako spremeni v raziskovalno pogodbo z jasnimi obveznostmi.
Lossless recoding preserves source entropy. Discovered dependence can remove the coding cost of falsely treating parts as independent. Those are different statements.
Brezizgubno prekodiranje ohrani entropijo vira. Odkrita odvisnost lahko odstrani kodirno ceno napačno predpostavljene neodvisnosti delov. To sta različni trditvi.
A bare mc⁴ is not a number of bits. A successor relation needs independently fixed units, a physical regime and a discriminating prediction.
Goli mc⁴ ni število bitov. Naslednja zveza potrebuje neodvisno določene enote, fizikalni režim in razlikovalno napoved.
Compression remains an open construction route. Discovery, decoding and correct negative answers must fit the same uniform polynomial budget.
Kompresija ostaja odprta konstrukcijska pot. Odkritje, dekodiranje in pravilni negativni odgovori morajo soditi v isti uniformni polinomski proračun.
The apparent inversion between phrase count and code length becomes understandable when the missing successor identities are put back into the account.
Navidezni obrat med številom fraz in kodno dolžino postane razumljiv, ko v račun vrnemo manjkajoče identitete naslednikov.
Two five-city tours share length, ranks, histogram and code length, yet their best one-move improvements are 6 and 4.
Dve petmestni turi imata isto dolžino, range, histogram in kodno dolžino, a najboljši enopotezni izboljšek je 6 oziroma 4.
Does a representation predict what a fixed next action will achieve beyond cheap context we already know?
Ali predstavitev napove, kaj bo doseglo določeno naslednje dejanje, nad poceni kontekstom, ki ga že poznamo?
What sprouted is not a ToE proof. It is a better decomposition of “compression”: complete description cost, reference-relative structure and action-relevant information are different objects and should be tested separately.
Kar je vzklilo, ni dokaz ToE. Je boljša razgradnja pojma »kompresija«: cena popolnega opisa, struktura glede na referenco in informacija, pomembna za dejanje, so različni objekti in jih je treba testirati ločeno.
In a k-nearest-neighbour rank representation, an edge to a city outside the list becomes one symbol: E. A random tour can therefore look like E E E E E … — extremely simple as a symbol stream. But every E must still identify which outside city was actually chosen.
V rangovni predstavitvi k najbližjih sosedov postane povezava do mesta zunaj seznama en sam simbol: E. Naključna tura je zato lahko videti kot E E E E E … — zelo preprosta kot tok simbolov. Toda vsak E mora še vedno povedati, katero zunanje mesto je bilo dejansko izbrano.
Phrase count ranks the projected symbol streams one way; complete code length ranks these two examples the other way. That inversion is explainable. It does not establish a universal sign law for LZ, and it does not make every cost-aware code a quality oracle.
Število fraz razvrsti projicirana tokova na en način; celotna kodna dolžina ta dva primera razvrsti obratno. Ta obrat lahko pojasnimo. To ne vzpostavi univerzalnega zakona predznaka za LZ in ne naredi vsake cenovno utežene kode za merilnik kakovosti.
The first pilot used a Spearman implementation that mishandled ties and defined its “5% band” around the best observed developmental tour. After correction, the regenerated development set did not reach the true ≤5% band. Across the full set, L_rank and escape count were effectively tracking the same dominant payload term. The old headline advantage is therefore not evidence.
Prvi pilot je uporabljal Spearmanovo implementacijo z napačno obravnavo vezi in »5-odstotni pas« določil glede na najboljšo opaženo razvojno turo. Po popravku regenerirana razvojna zbirka ni dosegla pravega pasu ≤5 %. Na celotni zbirki sta L_rank in število escape robov v glavnem sledila istemu dominantnemu payload členu. Stara glavna prednost zato ni dokaz.
A five-city truth-known fixture gives two tours with the same length, the same directed rank sequence, the same rank histogram and the same KT0 model bits — yet their local response differs.
Petmestni truth-known primer da dve turi z isto dolžino, istim usmerjenim rangovnim nizom, istim rangovnim histogramom in enakimi KT0 modelnimi biti — vendar se lokalno odzoveta drugače.
| TourTura | L | rank sequencerangovni niz | L_rank KT0 | best 2-opt neighbournajboljši 2-opt sosed | gainizboljšek |
|---|---|---|---|---|---|
| A = 0-2-1-4-3 | 41 | E E 1 E 0 | ≈ 15.755 b | 35 | 6 |
| B = 0-3-4-2-1 | 41 | E E 1 E 0 | ≈ 15.755 b | 37 | 4 |
All five nontrivial 2-opt reversals of each tour were enumerated and their integer EUC_2D lengths recomputed. The point is narrow: these listed summary variables are not sufficient statistics for exact next-move gain. Payload identity can matter even when payload length does not.
Za vsako turo je bilo enumeriranih vseh pet netrivialnih 2-opt obratov in ponovno izračunanih njihovih celoštevilskih EUC_2D dolžin. Sklep je ozek: navedene povzetkovne spremenljivke niso zadostna statistika za točen izboljšek naslednje poteze. Identiteta payloada je lahko pomembna tudi, ko njegova dolžina ni.
This changes the research target. Instead of asking only “does Z correlate with current tour length?”, ask “does Z help predict the consequence of a specified next action beyond context B?”
To spremeni raziskovalni cilj. Namesto da sprašujemo samo »ali Z korelira s trenutno dolžino ture?«, vprašamo »ali Z pomaga napovedati posledico določenega naslednjega dejanja nad kontekstom B?«
Let B be cheap known context, Z a candidate representation, a the action and Y its future outcome. The useful question is whether Z removes uncertainty about Y that B did not already remove.
Naj bo B poceni znani kontekst, Z kandidatna predstavitev, a dejanje in Y njegov prihodnji izid. Koristno vprašanje je, ali Z odstrani negotovost o Y, ki je B še ni odstranil.
How many bits does the complete, reconstructible description cost?
Koliko bitov stane celoten opis, iz katerega objekt obnovimo?
How does the representation compare with a declared reference family?
Kako se predstavitev ujema z deklarirano referenčno družino?
Does using it improve predictions or decisions after its costs are paid?
Ali njena uporaba izboljša napovedi ali odločitve, ko plačamo njeno ceno?
Imagine two equally likely states with the same current quality. Action A/B gains are (4,0) in the first and (0,4) in the second. A fixed choice averages 2; knowing the state allows 4 before observation costs. But if A wins in both states, better gain prediction need not change the action at all. Both are constructed examples, not arena results.
Predstavljajmo si dve enako verjetni stanji z isto trenutno kakovostjo. Izboljška dejanj A/B sta v prvem (4,0), v drugem (0,4). Fiksna izbira da povprečno 2; poznavanje stanja omogoči 4 pred stroški opazovanja. Če pa A zmaga v obeh stanjih, boljša napoved izboljška morda sploh ne spremeni dejanja. To sta konstruirana primera, ne rezultata arene.
Ideal value of observation with the same available actions and the option to ignore Z. Net value after sensing, learning and use costs can be negative; a learned policy can also be wrong.
Idealna vrednost opazovanja ob istih dejanjih in možnosti, da Z prezremo. Neto vrednost po ceni merjenja, učenja in uporabe je lahko negativna; naučena politika se lahko tudi zmoti.
For q=(0.8,0.2), the all-first-symbol distribution costs 0.322 bits per symbol under q; q itself costs 0.722. A more concentrated distribution can therefore be cheaper without being closer. For reference-law similarity, use an actual divergence such as KL or Jensen-Shannon, with smoothing and an independently declared reference law. Cross-entropy may still be a useful predictive feature; it is simply not a symmetric “distance to optimum”.
Za q=(0,8;0,2) stane porazdelitev samo prvega simbola 0,322 bita na simbol, q sam pa 0,722. Bolj koncentrirana porazdelitev je zato lahko cenejša, ne da bi bila bližja. Za podobnost referenčnemu zakonu uporabimo pravo divergenco, npr. KL ali Jensen-Shannon, z glajenjem in neodvisno določenim referenčnim zakonom. Cross-entropy je lahko še vedno uporabna napovedna značilka; ni pa simetrična »razdalja do optimuma«.
If a coarse state is meaningful, evolving the detailed state and then compressing should approximately agree with evolving the compressed state under a declared coarse dynamics. The metric, horizon and tolerance must be fixed before the result.
Če ima grobo stanje pomen, naj se razvoj podrobnega stanja in nato kompresija približno ujemata z razvojem že stisnjenega stanja po deklarirani grobi dinamiki. Metriko, horizont in toleranco je treba določiti pred rezultatom.
KL divergence is the ideal excess code length from using q on source p.
KL-divergenca je idealni presežek kodne dolžine zaradi uporabe q na viru p.
Under the specified thermodynamic setup, the same divergence measures excess free energy after physical units are restored.
V določenem termodinamičnem okviru ista divergenca po vrnitvi fizikalnih enot meri presežek proste energije.
Correlation is the ideal coding penalty of falsely assuming independent parts.
Korelacija je idealni kodirni pribitek napačne predpostavke neodvisnih delov.
These are established identities, not a new unification theorem. A shorter file is not automatically lower thermodynamic entropy, and LZ savings are not joules.
To so uveljavljene identitete, ne nov izrek poenotenja. Krajša datoteka ni avtomatično nižja termodinamična entropija in LZ-prihranek ni energija v joulih.
If i means bits, i = mc⁴ is dimensionally incomplete as written. The productive task is not to hide the seed, but to ask what measurable information quantity, regime and scale could make a successor relation nontrivial and predictive.
Če i pomeni bite, je i = mc⁴ v zapisani obliki dimenzijsko nepopoln. Produktivna naloga ni skriti semena, ampak vprašati, katera merljiva informacijska količina, režim in skala bi lahko naredili naslednjo zvezo netrivialno in napovedno.
For E=mc², geometry R remains explicit.
Pri E=mc² geometrija R ostane eksplicitna.
The scaling is m² for this saturated gravitational system.
Za ta nasičeni gravitacijski sistem je skaliranje m².
Thermodynamic entropy, information-capacity bounds and ideal operation-rate bounds are different observables. Their mass scalings also depend on what else is held fixed: for example Bekenstein's mR becomes m4/3 at fixed density because R ∝ m1/3. The useful question is where a regime changes from volume-like bookkeeping toward area-like bounds and what physical mechanism causes the transition.
Termodinamična entropija, meje informacijske kapacitete in idealne meje hitrosti operacij so različne količine. Njihovo skaliranje z maso je odvisno tudi od tega, kaj držimo fiksno: Bekensteinov mR je pri fiksni gostoti m4/3, ker R ∝ m1/3. Koristno vprašanje je, kje režim preide iz volumskega računovodstva proti ploskovnim mejam in kateri fizikalni mehanizem ta prehod povzroči.
Survival contract for the seed: define i, derive the dimensional factor independently, state the physical domain, and predict something that differs from an existing identity or bound.
Pogodba preživetja semena: definirati i, neodvisno izpeljati dimenzijski faktor, navesti fizikalno domeno in napovedati nekaj, kar se razlikuje od že znane identitete ali meje.
The compression route remains open as a construction programme, not as a proof claim. The semantic-inversion lesson strengthens the accounting: do not confuse a cheap label with a cheap complete witness mechanism.
Kompresijska pot ostaja odprta kot konstrukcijski program, ne kot trditev o dokazu. Lekcija semantičnega obrata zaostri račun: poceni oznake ne smemo zamenjati s poceni celotnim mehanizmom priče.
| RoutePot | Sufficient formDovolj močna oblika | Still missingŠe manjka |
|---|---|---|
| A · witness list | Polynomial-size, polynomial-time constructible set guaranteed to contain a witness if one exists.Polinomsko velika in v polinomskem času zgrajena množica, ki zagotovo vsebuje pričo, če obstaja. | SAT-wide construction. |
| B · structural width | Equivalent representation of O(log n) width, found in polynomial time, with exact DP.Ekvivalentna predstavitev širine O(log n), najdena v polinomskem času, z eksaktnim DP. | Universal discovery without hidden exponential work.Univerzalno odkritje brez skrite eksponentne cene. |
| C · short + fast programs | O(log n)-bit witness-generating programs with one uniform polynomial runtime bound.O(log n)-bitni programi za generiranje priče z eno uniformno polinomsko časovno mejo. | All satisfiable instances + correct UNSAT decision.Vse zadovoljive instance + pravilna odločitev UNSAT. |
Positive control: affine systems over GF(2) already show the pattern when it is true: elimination exposes rank, free variables and a compact complete solution description in polynomial time. That is a model of what a successful structural theorem looks like, not evidence that general SAT shares it.
Pozitivna kontrola: afini sistemi nad GF(2) že pokažejo vzorec, ko je resničen: eliminacija v polinomskem času razkrije rang, proste spremenljivke in kompakten popoln opis rešitev. To je model uspešnega strukturnega izreka, ne dokaz, da ga ima tudi splošni SAT.
This is a research objective, not yet a universal DCC law. A sensor can be informative yet too expensive, predictive yet decision-irrelevant, or useful in one regime and misleading in another. DCC should compare such trade-offs rather than assume that one complexity scalar always has the right polarity.
To je raziskovalni cilj, ne še univerzalni DCC-zakon. Senzor je lahko informativen, a predrag; napoveden, a nerelevanten za odločitev; ali uporaben v enem režimu in zavajajoč v drugem. DCC naj primerja te kompromise, ne pa predpostavlja, da ima en sam skalar kompleksnosti vedno pravi predznak.
Change-point and non-stationary bandit theory provide positive controls for a narrower statement: adaptive policies can outperform a fixed policy when the environment changes, under specific assumptions and regret definitions. That does not transfer a theorem to DCC. It gives DCC a mature neighbouring theory against which candidate sensors and switching logic can be compared.
Teorija zaznavanja sprememb in nestacionarnih banditov ponuja pozitivne kontrole za ožjo trditev: prilagodljive politike lahko pod določenimi predpostavkami in definicijami obžalovanja prekašajo fiksno politiko v spreminjajočem se okolju. Izrek se s tem ne prenese na DCC. DCC dobi zrelo sosednjo teorijo, s katero lahko primerjamo kandidatne senzorje in logiko preklapljanja.
A speculative ontological interpretation: a proposed fundamental experiential ground or field-like continuum. It is not established here as a physical carrier or extra force.
Spekulativna ontološka interpretacija: predlagana temeljna izkustvena podlaga oziroma polju podoben kontinuum. Tu ni potrjen kot fizikalni nosilec ali dodatna sila.
An evidence-first biological research programme. The current scientific lane treats X(t) as a candidate multidimensional neural state; optional scalar targets such as S* may fail independently and are not consciousness meters.
Biološki raziskovalni program, usmerjen v dokaze. Trenutna znanstvena veja obravnava X(t) kot kandidatno večdimenzionalno nevronsko stanje; opcijski skalarji, kot je S*, lahko odpovejo neodvisno in niso merilniki zavesti.
Compare B, B+r₀, B+histogram and B+LZ. B includes length, escape fraction, generator, checkpoint age and recent progress. The prepared baseline is standardized ridge regression with fixed λ=1 and leave-one-group-out prediction. New features must earn their cost above that baseline.
Primerjamo B, B+r₀, B+histogram in B+LZ. B vsebuje dolžino, delež escape robov, generator, starost posnetka in nedavni napredek. Pripravljena osnova je standardizirana ridge-regresija s fiksnim λ=1 in napovedjo z izločanjem celih skupin. Nove značilke morajo upravičiti svojo ceno nad to osnovo.
Five predictions were written before the target run. The protocol deliberately includes a losing path: its stated default expectation is that the new rank features may add nothing beyond B on this development set. A null result is therefore a planned result, not a failed story.
Pet napovedi je bilo zapisanih pred ciljnim tekom. Protokol namenoma vsebuje tudi pot poraza: zapisano privzeto pričakovanje je, da nove rangovne značilke na tej razvojni zbirki morda ne dodajo ničesar nad B. Ničelni rezultat je zato načrtovan rezultat, ne neuspešna zgodba.
P1. r₀ should have a positive partial association with both continuation gains given B. P2. B+r₀ should improve held-out MSE for at least one action. P3. B+histogram should not beat B+r₀; if it does, histogram shape matters beyond concentration. P4. B+LZ is expected to add nothing; if it improves with an interval excluding zero, the sequential channel is empirically live. P5. or-opt is expected to win on most states, so the informative targets may be the per-action gains rather than the winner.
P1. r₀ naj bi imel pozitivno parcialno zvezo z izboljškom obeh nadaljevalnih dejanj ob znanem B. P2. B+r₀ naj bi izboljšal held-out MSE za vsaj eno dejanje. P3. B+histogram naj ne bi premagal B+r₀; če ga, oblika histograma nosi informacijo poleg koncentracije. P4. B+LZ po pričakovanju ne doda ničesar; če izboljšanje dobi interval brez ničle, je sekvenčni kanal empirično živ. P5. or-opt naj bi zmagal na večini stanj, zato so lahko bolj informativni izboljški po posameznem dejanju kot sam zmagovalec.
Pre-run design choices are now selected: the first development run keeps the pair (2-opt, or-opt); each continuation is intended to run to convergence under a hard safety cap rather than stop merely because both consumed the same number of candidate evaluations; the primary state filter is gap ≤0.5 and gap ≤0.35 is the sensitivity analysis. This selection was made before observing the target result. The executable runner still has to encode the convergence rule and its numerical safety cap before the run can be called frozen. Group identities must follow actual shared origins; 128 snapshots are not 128 independent runs.
Izbire zasnove pred tekom so zdaj določene: prvi razvojni tek ohrani par (2-opt, or-opt); vsako nadaljevalno dejanje naj teče do konvergence pod trdo varnostno mejo, namesto da se ustavi zgolj zato, ker sta obe dejanji porabili enako število kandidatnih ocen; primarni filter stanj je gap ≤0,5, občutljivostna analiza pa gap ≤0,35. Izbira je bila narejena pred opazovanjem ciljnega rezultata. Izvršilni program mora še zapisati pravilo konvergence in njegovo numerično varnostno mejo, preden lahko tek imenujemo zamrznjen. Skupine morajo slediti dejanskemu skupnemu izvoru; 128 posnetkov ni 128 neodvisnih tekov.
If one operator dominates, predicting gain is still testable, but the best-action question may have little variation. Wide intervals are inconclusive, not proof of equivalence. A positive development result justifies an independent next test, not a production sensor replacement.
Če en operator prevladuje, lahko še vedno testiramo napoved izboljška, vprašanje najboljšega dejanja pa ima lahko malo variacije. Široki intervali so nedoločeni, ne dokaz ekvivalence. Pozitiven razvojni rezultat upraviči naslednji neodvisni test, ne zamenjave produkcijskega senzorja.
Projection loss, code-model choice, sequence order, orientation and ties, generator mixture, geometric move availability, delayed reorganisation and the regulator's causal role can produce different signatures. The first six address the static measurements directly. The last two need actual temporal data. Reference-law transfer, shuffled feedback, physical energy and CCH are separate tests, not parts of one all-or-nothing ToE verdict.
Izguba v projekciji, kodni model, vrstni red, orientacija in vezi, mešanica generatorjev, geometrijska razpoložljivost potez, zakasnjena reorganizacija in vzročna vloga regulatorja imajo lahko različne podpise. Prvih šest neposredno zadeva statične meritve. Zadnja dva potrebujeta časovne podatke. Prenos referenčnega zakona, premešani feedback, fizična energija in CCH so ločeni testi, ne deli ene same sodbe vse-ali-nič o ToE.
| ItemPostavka | Status at v0.4Status v v0.4 | What it supportsKaj podpira |
|---|---|---|
| Original narrow-band L_rank advantagePrvotna prednost L_rank v ozkem pasu | WithdrawnUmaknjeno | The old tie-handling and band definition were wrong. The replacement development set has no tours in the ≤5% band.Stara obravnava vezi in pasu je bila napačna. Nadomestna razvojna zbirka nima tur v pasu ≤5 %. |
| 128 tours / 34 reported groups128 tur / 34 navedenih skupin | Development reanalysis archivedRazvojni preračun arhiviran | Whole-range Δρ ≈ −0.0013, interval [−0.0177, 0.0053]. No shown L_rank advantage; equivalence was relative to a developmental ±0.10 margin.Celotni razpon Δρ ≈ −0,0013, interval [−0,0177;0,0053]. Ni prikazane prednosti L_rank; ekvivalenca je veljala glede na razvojni prag ±0,10. |
| Projection / payload pairPar projekcija / payload | Recomputed for this releasePonovno izračunano za to izdajo | A concrete accounting mechanism, not a universal sign law.Konkreten računovodski mehanizem, ne univerzalni zakon predznaka. |
| Five-city fixturePetmestni primer | All one-move neighbours checkedPreverjeni vsi enopotezni sosedi | The listed scalars are not universally sufficient for exact next-step response.Navedeni skalarji niso univerzalno zadostni za točen naslednji odziv. |
| T-A* / proposal 3 | Prepared; five predictions written before the target run; five synthetic test functions passed againPripravljeno; pet napovedi zapisanih pred ciljnim tekom; pet sintetičnih testnih funkcij ponovno uspešnih | Executable preparation with an explicit null-friendly prior. No 128-state continuation result, no demonstrated DCC improvement.Izvršljiva priprava z izrecnim pričakovanjem, ki dopušča ničelni izid. Brez rezultata nadaljevanj 128 stanj, brez dokazane izboljšave DCC. |
| CFH · mc⁴ · P=NP | Open research commitmentsOdprte raziskovalne naloge | No proof is transferred from the TSP checks.Dokaz iz TSP kontrol se ne prenaša. |
This edition joins the v0.3 manuscript, Claude's Semantic Inversion Brief, GPT's RHP synthesis, a second concurrent RHP analysis, and Claude_ToE_proposal_3.zip. Claude later rechecked the five-city fixture and the trajectory grouping from the same shared evidence. That is a useful consistency check, not independent empirical replication. The sources share evidence and have seen one another's results; the original files remain preserved.
Izdaja povezuje rokopis v0.3, Claudov Semantic Inversion Brief, GPT-jevo RHP sintezo, drugo sočasno RHP analizo in Claude_ToE_proposal_3.zip. Claude je pozneje na istih skupnih dokazih ponovno preveril petmestni primer in grupiranje trajektorij. To je uporaben konsistenčni pregled, ne neodvisna empirična ponovitev. Viri delijo dokaze in so videli rezultate drugih; izvirne datoteke so ohranjene.
Manuscript: 8Z_CFH_ToE_v0.4_20260917.md
Proposal 3 SHA-256: 794bdfc5399b040a259d0862a3b3efebfa8eaf056a51463e3cec117055974c44
Source package: T_A_STAR_PREP_20260917_R1 · 21 manifest-listed files.
This synthesis does not replace the individual research lanes. Follow a claim to the page that owns its definitions and evidence.
Ta sinteza ne nadomesti posameznih raziskovalnih vej. Trditvi sledi do strani, ki določa njene definicije in dokaze.
A derivation survives counterexamples and adds a prediction not already in the assumptions.
Izpeljava preživi nasprotne primere in doda napoved, ki ni že skrita v predpostavkah.
A bounded experiment distinguishes competing explanations with a frozen outcome rule.
Omejen eksperiment razlikuje konkurenčne razlage z vnaprej določenim pravilom izida.
A new exact complexity result, or a physical equation with independently fixed units and a discriminating prediction.
Nov eksakten rezultat zahtevnosti ali fizikalna enačba z neodvisno določenimi enotami in razlikovalno napovedjo.
More prose alone is not v0.5.
Samo več besedila ni v0.5.
[R1] C. E. Shannon, A Mathematical Theory of Communication (1948).
[R2] S. Kullback & R. A. Leibler, On Information and Sufficiency (1951).
[R3] S. Watanabe, Information Theoretical Analysis of Multivariate Correlation (1960).
[R4] J. D. Bekenstein, Universal upper bound on the entropy-to-energy ratio for bounded systems, Phys. Rev. D 23, 287 (1981).
[R5] R. Landauer, Irreversibility and Heat Generation in the Computing Process (1961).
[R6] A. Hutter, Universal Artificial Intelligence (2005) — background on prediction and description length; not a direct validation of this programme.
[R7] R. J. Solomonoff, A Formal Theory of Inductive Inference (1964).
[R8] P. D. Grünwald, The Minimum Description Length Principle (2007).
[R9] L. Levin, universal search / search-complexity lineage; used only as conceptual background for description + computation cost.
[R10] M. R. Garey & D. S. Johnson, Computers and Intractability (1979).
[R11] T. J. Schaefer, The Complexity of Satisfiability Problems (1978).
[R12] G. B. Dantzig, R. Fulkerson & S. Johnson, TSP cutting-plane lineage; background only.
[R13] D. L. Donoho, Compressed sensing lineage; analogy limited to representation and recovery, not evidence for the ToE seed.
[R14] R. Bousso, holographic principle reviews; area-law context.
[R15] S. Lloyd, Ultimate physical limits to computation (2000).
[R16] A. Garivier & E. Moulines, non-stationary bandit work; neighbouring control theory, not a theorem about DCC.
[R17] R. B. Myerson / value-of-information decision-theory lineage; the page uses the elementary decision form directly.
This selected list is orientation, not a claim that the research programme follows from these works. Full bibliographic and source notes remain in the manuscript package.
Ta izbrani seznam je orientacija, ne trditev, da raziskovalni program sledi iz teh del. Polne bibliografske in izvorne opombe ostajajo v paketu rokopisa.