← AMRAL · PROGRAM-DMSD
Decimal Multiscale Serializable Domain · a software and representation architecture with a prime stress test; no theorem, and no new result about primes
DMSD's idea: a theoretically unbounded integer domain can be serialized first as decimal scale levels Lk = [10k, 10k+1), expanded lazily, viewed through two projections (a magnitude pyramid and a serialized staircase), and equipped with 'property layers' on every block — exact summary algebras, epistemic states (UNKNOWN / EXACT_COMPUTED / EXACT_MERGED, under a 'No False Closure' rule), source depth, budget policy and provenance. The structure: the foundational methodology v0.1 → Phase 2A–2C (parity, residue, square; closed-form block summaries up to scale k ≤ 100) → Phase 3A–3I (the prime layer, used as a stress test; exact data for n < 107) → Phase 4, the generalized formal core v1 ('FROZEN FOR VALIDATION'; primes demoted to 'Reference Stress Test A') → Horizon Edition v0.1–v0.8 (engineering: addressing the scale of scales, a compute-capacity overlay, algorithm calibration and routing, timed only on one AI sandbox; not number theory) → two later side prototypes: the M6 fixed-point carrier (n ≤ 106) and DNGP/DNCP prime survival (n ≤ 100,000).
Honest status: DMSD is mainly a software and representation architecture, not a mathematical theory — no theorem is stated or proved anywhere in the package. The exact content is only elementary closed-form summaries (parity, residue, square, up to scale k ≤ 100, with brute-force oracles) and finite-range prime data (n < 107, bases 2/8/10/16; M6 to 106; DNGP to 100,000). This site recomputed from scratch: the decade prime counts 4, 21, 143, 1061, 8363, 68906, 586081; the 3B bias table; 3D's explained fractions 95.68 / 99.13 / 99.31%; the 3G/3H closure y = 313 / 997 / 3137 and burdens 61,066 / 768,662 / 9,047,215; M6's 333,332 / 78,496; DNGP's π(105) = 9592, 1,224 twin pairs, maximum gap 72 and four gaps of length 58 — all match, and the shipped reference code runs. Nothing concerns 'infinitely many primes' or 'all primes', and the package never mentions Goldbach, Legendre, Riemann or the twin-prime conjecture (twin appears only as a count and as mod-30 residues). The prime phases end in null results: 3D 'the raw left bias of 3B/3C is mainly not new prime geometry'; 3G 'no residual feature qualifies as a new prime invariant'; M6 'Novel prime theorem claimed: NO'; DNGP 'Status of novel theorem claims: NONE'. 'Closure' here is classical trial division up to √n.
Authorship and provenance: none of the 23 documents, nor anything inside the zips, carries an author or AI-role line; this site does not guess and will add it once Neo supplies it. The only cues are the AI assistant's first-person voice ('I think', 'I feel'), two second-person quotations of Neo (Horizon v0.2 §20, M6 §19), Horizon v0.5's remark that timings ran in a Linux sandbox rather than on the user's machine, and opaque search-tool tokens in the Horizon v0.2/v0.3 data. Those conversational lines and tokens stay in the original bytes, and an editor's note is added on the affected pages.
Easy to misread: (1) M6's 'exact' results are definitional or vacuous (lpf(n) = n iff n is prime; log₁₀ is strictly increasing, so any increasing sequence passes); the paper concedes this itself and says one cannot conclude the primes form a continuous subset of R. (2) DNGP's 'Gap-depth fingerprints: FOUND' lacks a null comparison and is promotional in tone; the paper limits it with 'this is not a new prime theorem'. (3) Horizon v0.1–v0.3 label externally sourced 2026 figures CURRENT_VERIFIED, but they are source-reported, unverifiable and will go stale. (4) Phase 4's 'PASS' is the specification's own self-test, exercised with only four plugins. (5) The REPORT.md of Phase 2A has control-character-corrupted LaTeX (the source is like that). Horizon v0.1–v0.8 are cumulative increments, not independent papers; 3B/3C's candidate pattern is corrected by 3D, so read them as one arc.
Missing files and relations: Phase 3F (cited by 3G), Phase 1 and Horizon v0.9 are not in the package; SDRPC (the earlier prime line cited in 3A/3E/3G/3H/3I) is neither in the package nor on this site. M6's definitions are cited nowhere; they come from the RG-SCP series of this site's Goldbach case (2026-09-28). No file in this folder's zips is byte-identical to anything in the X4_DMSD sub-line of the perfect-cuboid folder (233 versus 81 files compared); the names suggest the latter is a DMSD-derived side branch.
Reading path: start with the foundational methodology (especially §28–30, the bounded-extrapolation duty), Phase 3D (the most honest page of the package) and Phase 4 (the specification). To see how UNKNOWN becomes a formal state read Phase 3A; for 'looks like a pattern, is really the density baseline' read 3C → 3D → 3G; the M6 page is there to prevent the misreading 'the primes are a continuous curve'. The other documents have no pages of their own and are listed in the table below for download. M6's original zip (19.5 MB, 97% of it regenerable CSVs) is not offered here; only its REPORT.md is included, and the zip's SHA-256 is recorded on the M6 page.
The methodology, Phases 2B/2C/3A/3C/3D/3G/4, M6 and Horizon v0.2 comprise 10 reader pages and 10 complete bodies, all available in Chinese and English. The authoritative originals are zh-Hant with English terminology; translation is not independent mathematical verification and does not upgrade the original evidence level. The other phases remain original-only materials in the download table below.
The original bytes of the documents with pages (each REPORT.md is named here 'zip name_REPORT.md'; the bytes equal the zip member's) and of the remaining phases, unmodified; the original zips are attached too (except M6's 19.5 MB zip). The other phases carry a one-line claim and a flag ('none' = no overclaim found; 'mild' = wording a reader could over-read, or corrected later). The complete archive is kept in AMRAL's research archive and verified per file by SHA-256.