Independent computational research · VARZIN Project
A research program at the intersection of finite affine mathematics, reproducible computation, and constructed symbolic systems.
Led by Reza Nirouyar. Built around explicit assumptions, controlled tests, and a record that remains open to correction.
01 / Reading the evidence
A mathematical proof, a computational report and an empirical question are distinct research records. Each should be read within its stated scope.
Affine bijection criterion; valid multiplier count.
Deterministic implementations; verify the specific release.
Computational findings with stated artifact limits.
Mirror-13 MDL depends on its coding and candidate space.
Designed structure and hypotheses requiring controls.
Tested criteria or predicted recovery not achieved.
Independent semantics remain unestablished.
Human evaluation; no established outcome.
Earlier claims retained for provenance.
No established evidence for physical or consciousness-field effects, quantum mechanisms, frequency-to-meaning causation, or independent semantic emergence. Read the current research position ↗
02 / Research directions
Affine permutations over finite cyclic spaces, deterministic models and construction-specific audits.
The bijection criterion is proved mathematics. Mirror-13 connectivity checks are a separate computational record.
Read the finite affine systems guide ↗A constructed symbolic system with deliberately designed morphemic and structural regularities.
The broader v2.2 design reports 801 stable roots and an approximately 1.3-billion-form generated corpus. Later work adds distinct 300-word, 360-word, and 10,800-record benchmarks; these are not one silently enlarged canonical lexicon. Surface structure does not establish independent semantics.
Understand LUXVAR ↗ · Scale & benchmark map ↗Protocols for separating intended semantic structure from morphology, matched controls and training effects.
Recruitment has not commenced for the later N=59 plan. No human outcome is established; protocol identities remain unresolved.
Read protocol history & limits ↗03 / Applied research
A tested local MVP plus a live commercial generation API: Aff(ℤ₁₂)/LUXVAR challenge construction delivered locally through Streamlit and globally through RapidAPI with a Render-hosted FastAPI origin.
The corrected local Streamlit build completed all ten generated challenges without application exceptions. Production bot-resistance remains an open validation question.
See test evidence, limitations & public demo ↗Earlier mockup-level production-security claims were not supported by the current evidence base and are not promoted on the research portal.
The delivery path is explicit: verified local MVP + RapidAPI/Render API → origin hardening and independent human-versus-automation evaluation → larger unseen challenge family → stronger production security engineering.
Read the model-evidence boundary ↗04 / Selected research records
Project-listed public deposits. Listing a record does not imply independent validation or peer review; version-specific DOIs are preserved for provenance.
Open software + companion monograph: finite-affine core, torus solver, deterministic genomic scanner, tectonic null-model audit, and versioned reproducibility files.
DOI: 10.5281/zenodo.22036769 ↗ Companion monograph →Direct frozen-model audit: Mistral 7B ARI −0.0337 ± 0.039 and Llama 3 8B ARI 0.0152 ± 0.028 against the designed orbit labels; the axes were not recovered under the tested protocol.
DOI: 10.5281/zenodo.22101179 ↗Frozen Qwen structural information is strongly linearly decodable (mean BA 0.9666), while the preregistered composition phase reports 0/58 passes; recovery did not establish systematic composition in this experiment.
DOI: 10.5281/zenodo.22679978 ↗v3 adds the harder composition test: a seen-pair MLP fit ≈0.996, but held-out TRUE accuracy 0.106 falls below SHUFFLED 0.281 and WRONG_OP 0.175.
DOI: 10.5281/zenodo.22287006 ↗Morphology and phonotactics remain the supported signal; semantic-axis tests are negative, and the manuscript addendum explicitly rejects the historical “AI-Recoverable” axis interpretation.
DOI: 10.5281/zenodo.22115483 ↗05 / Reproducibility
Reproducibility belongs to a specific artifact and release. For current scientific interpretation, this portal now prioritizes the latest relevant Zenodo manuscript and its version-matched code/data package; older prose remains provenance rather than authority over later evidence.
Use the exact DOI and release in the publication registry.
Separate proved statements from reported computation and inspect the evidence boundaries.
Consult preprint corrections and addenda; retain artifact limitations in any citation.
06 / Negative results
Negative results constrain the next question. They remain visible alongside positive findings.
Track D reported ARI −0.0337 for Mistral 7B and +0.0152 for Llama 3 8B against intended axis/orbit labels. This finding applies to those unremediated models and that protocol.
Open the frozen-audit record ↗In v3, seen-pair MLP fit ≈0.996 but held-out TRUE accuracy is 0.106, below SHUFFLED 0.281 and WRONG_OP 0.175. VARZIN V2 separately reports 0/58 composition passes despite mean frozen-Qwen probe BA 0.9666.
Read the current AI evidence map ↗SEM-001 found no string-level semantic-axis signal in the tested analysis. SEM-002 met 1 of 4 criteria against a threshold of at least 2 of 4.
Read the preprint record ↗07 / Provenance & corrections
Earlier interpretations remain traceable. Corrections change the reading of the evidence; historical documents do not override the current position.
Inspect the access-action register ↗Historical ARI 0.812 concerns inter-model surface/morpheme agreement, not intended semantic-axis recovery. Raw pairwise scoring was not independently reverified at correction.
The 20691858 / 22115483 deposit-version mapping and YEGJ8 / 2NKZA protocol relationship remain partially unresolved.
Earlier symbolic narratives, historical reports and the previous portal are preserved as provenance, not current findings.
The computational atlas
Mathematical core, Core-30 registry, methods, negative findings and the archive boundary—in one public index.
Enter the research atlas ↗Researcher / Project author
Independent Researcher · VARZIN Project
Research in finite affine systems, reproducible computational models and the structured analysis of LUXVAR.
Use the DOI or release associated with the finding. For the portal itself: Nirouyar, Reza. VARZIN Research Portal and Reproducibility Materials. VARZIN Project.
Find citation records ↗