8i / SDN / O.R.C.A. / V.A.S.K. - ENGINEERING REVIEW 18 October 2026 - reviewed derivatives, not replacement historical originals RESULT Two active Julia source branches were inspected throughout and run on a native Julia 1.13.2 host. Confirmed failures were repaired in separately named files: 8i-reviewed.jl ORCaRzGate-reviewed.jl The original September master and original ORCaRzGate.jl retain their bytes. The complete September original and reviewed derivative remain on page 04, O.R.C.A. / Architecture, and in the downloadable source atlas. REPRODUCIBLE EVIDENCE 2,307 assertions: julia/regression.jl, 8 substantive regression groups. 22/22: reviewed September CLI self-test. 9/9: reviewed capsule/replay CLI self-test groups. 11/11: browser-core conformance groups in Node, including cross-runtime SDO. 69/69: supplied Trial Draft 02 finite mathematical checks rerun. Counts identify the exact named suites, not every possible execution path. The GitHub source includes the pinned Julia Project.toml and Manifest.toml, regression source, test output and the rerunnable manuscript verifier. BASELINE FAILURES WERE REPRODUCED The unchanged September file stops at a tuple-to-String constructor. After lifting only that compatibility blocker in memory, its original self-test records 17 passes and five failures: Unicode indexing in BrotherQuicky, MetaTrit grammar, MetaTrit use in TMS, micrograd field/function use, and a byte vector prefix check. This lifted run is distinct from the unchanged original. The original capsule/replay file fails to parse at its one-line enum body. An additional asymmetric 248-position test exposes the incomplete-triplet permutation bug that a hardcoded power-three shortcut can hide. CODECS, CONTAINERS AND IDENTITY 1. Replaced tuple String construction with character joining and repaired byte-prefix checks, Unicode string cursor handling and text escape paths. 2. Made MetaTrit encoder and decoder share the emitted grammar. Invalid trits, non-byte six-trit values, malformed tokens and excessive expansion fail. Long valid runs are chunked. Ternary text is a representation, not a promise of smaller storage; UTF-8 byte costs and complete envelopes count. 3. Required SDO raw and payload hashes and lengths. The payload is checked before bounded decompression; raw identity is checked after reconstruction. Header sizes, field forms and unsupported methods are validated explicitly. 4. Repaired current ZipArchives API use, empty dictionary reconstruction, multi-entry preservation and corrupt ZIP handling. A failed read does not silently discard earlier archive entries. Corrected command working dirs. 5. Added self-contained capsule-branch candidate objects with codec identity, parameters, lengths and both hashes. Arena comparisons include the entire serialized candidate, not a payload requiring omitted reconstruction data. 6. Canonical unsigned varints, explicit decode bounds, transpose shape/padding checks and bounded capsule inspection reject malformed inputs. TAR headers are inspected for paths, types, duplicates and declared size before import; the extracted file set must match the manifest and its hashes. STATE, AUTHORITY AND EXECUTION 7. Capability use now compares the presented fields with the stored grant, including expiry. Knowing a nonce does not authorize forged resource fields. 8. Scheduler resume only executes waiting jobs; completed work is not repeated. 9. Source-integrity checking compares the actual current file digest against its construction-time baseline. This is a change detector, not a signature or a guarantee against an attacker replacing both code and its baseline. 10. Shell actions use structured arguments. The arithmetic helper evaluates a restricted expression tree. Arbitrary in-process Julia evaluation is disabled; a word blacklist is not advertised as an execution sandbox. 11. Repaired constructor recursion, output buffers, micrograd calls, zero powers and neural layer dimensions. All 32 digest bytes reach the trit signature transform. False hardware acknowledgments were removed and synthetic telemetry is labelled. No real hardware link is certified. V.A.S.K. MATHEMATICS AND ITS SOFTWARE INTERPRETATION 12. A signed electromagnetic charge and permission to interact are different quantities. Permission support is nonnegative and stored separately. 13. Occupation is a normalized state/projector quantity. Projector construction checks basis shape and rank; states, Hermitian operators and hbar are checked. Conditioned readout uses the norm of the filtered state, K'*K, including weighted permission operators that are not sharp projectors. 14. For K=T_eff*d, the entropy-containing profile is V(a)=-mu*a+(K-r)*a^2/2+u*a^4/4 +K*[a*log(a)+(1-a)*log(1-a)]. V'(a)=-mu+(K-r)*a+u*a^3+K*log(a/(1-a)). V''(a)=K-r+3*u*a^2+K/[a*(1-a)]. For K>=0 and u>=0, V''>=5*K-r on 04*K alone is not a valid sufficient nonconvexity/activation criterion for this profile. A linear bias mu also changes its equilibria. The website offers entropy, polynomial-only and tilted double-well profiles separately. 15. At K=0, V'(1)=0 requires u=mu+r. Then V''(1)=3*mu+2*r. A scalar mass squared is V''/f^2 only with the stated constant kinetic normalization. A zero-temperature formula cannot silently assume mu=0. 16. The September polynomial equilibrium includes nonzero mu and bounded endpoints. Its mixed-sector energy matches the gradient used in descent; backtracking replaces updates that could increase the stated objective. 17. A three-cycle and its continuous generator are distinct constructions. H=kappa*(P+P') has equal forward and reverse transition probability (4/9)*sin^2(3*kappa*tau/(2*hbar)); it is not a directed cycle pulse. H_cyc=2*pi*hbar*i*(P-P')/(3*sqrt(3)*T) instead gives U(T)=P for the stated orientation and Schroedinger sign convention. 18. All complete adjacent triplets cycle; incomplete tails are fixed. The 82+82+82 plus two fixed singlets profile is separately retained, as is the named 3x16 generation-block action. They are not interchangeable operators. Tests apply the permutation three actual times to asymmetric states. 19. Corrected SO(N) adjoint Casimirs to N-2 under long-root norm squared 2, signed proper-time domain handling and two-state occupation limits. 20. Trial Draft 02's 69 supplied checks cover stated finite constructions: projectors, propagators, potential derivatives, representation counts, anomaly sums and geometry. Rerunning those checks verifies their scope; it is not empirical confirmation of a unified physical theory. GLYPHS, HARDWARE AND SECURITY CLAIMS The exact 256-entry graphic repertoire and its visible byte-zero substitute remain fixed and reversible. A Unicode scalar is not necessarily one UTF-8 byte. The table alone does not prevent memory leaks or establish a security boundary. Hardware resource count depends on the chosen output encoding, ports, target device and synthesis. No BRAM, timing or clock-cycle result is claimed without a target-specific synthesis report. Immutable/mutable roles are architectural intent; a prototype in one process is not an unbreakable hardware isolation boundary. AI PARTNERSHIP: THE NEXT MEASURABLE TARGET The promising near-term role is continuity and controlled admission: preserve source evidence, retrieve relevant objects, propose typed candidates, verify them independently, checkpoint accepted work and resume after interruption. Suggested adapter roles are .sd4 evidence, .sd3 transport, .sd2 inspection, .sd1 references/abstraction, .sd0 verified complete candidate and .sdn active state. This is a proposed integration, not an installed ChatGPT component. Compare the same model, tools, corpus, budget and interruption schedule with an equally persistent conventional database/checkpoint baseline. Include model swaps, corrupted evidence and failed candidates. Measure correct task completion, retrieval, duplicate effort, recovery, total latency, tokens, storage and measured energy. Current synthetic fixtures cannot establish a numerical intelligence boost, consciousness, personal identity or AI speedup. WEBSITE AND MANUAL Trial Draft 02 (17 October 2026) is the current 256-page manual, preserved byte-for-byte including its embedded editable sources and equation register. Trial Edition 01 and earlier manuscripts remain downloadable. Authored UI is CP437-first; exact source, scientific notation and visitor payloads have explicit exception surfaces. Client data remains local to the browser demos. Optional compressor failures keep raw candidates available. Async controls are guarded against double activation and stale recovery state. A direct Stripe-hosted, live, customer-chosen one-time support link is configured in EUR, minimum EUR 1. It grants no product, membership or subscription. No payment was made during this review. Configuration evidence is separate from an end-to-end paid transaction or webhook verification. LIMITS AND REPRODUCTION All original lines in the two named active Julia branches were inspected. Execution coverage is the listed CLI/regression cases, not every branch of every historical file. Older C++, Python, Julia and hardware snapshots keep their earlier, separately scoped evidence. External model APIs, GUI/voice, RF devices, 7z system integration, PS2 console boot and FPGA synthesis were not exercised by this host review. These revisions are research software, not a formal security certification or a complete adversarial audit. From the GitHub site's directory: julia --project=julia -e 'using Pkg; Pkg.instantiate()' julia --startup-file=no --project=julia julia/regression.jl julia --project=julia public/downloads/8i-reviewed.jl --root /tmp/8i-review self-test julia --project=julia public/downloads/ORCaRzGate-reviewed.jl selftest node --test tests/conformance.mjs python3 build.py python3 tests/verify_public_build.py The supplied manuscript verifier requires NumPy, SciPy and SymPy. Its equation inputs and resulting finite-check report are in reports/manual-02/. Release/deployment identity is verified separately from these code results.