Epistemic Architecture Research — Indonesia
Meaning Formation,
Not Prediction
An epistemic reasoning architecture that rejects next-token prediction in favor of structured, grounded meaning formation.
NINMENI (Navigating Interpretable Neural Meaning Encoding and Integration) is an attempt to build a language system from the inside out: meaning first, reasoning follows, language last. Not a transformer in disguise.
Paradigm
Meaning comes first,
not next-token prediction.
Meaning-First Architecture
GROUNDEDNINMENI inverts the classic LLM direction: meaning → knowledge → reasoning → language. Language is the interface, not the center. The system rejects surface prediction in favor of grounded structure.
Implementation Status
SPECULATIVEFoundation 36 Laws FROZEN — proven and functioning. Three forward tracks (Real Corpus, Large Ecology, Living World-State) are still in validation. The system hasn't been given a passport to the outside world yet.
Honest Ignorance
NINMENI doesn't guess. The system explicitly marks what it doesn't know (ABSTAIN) when grounding fails. Epistemic honesty matters more than baseless fluency.
Boundaries
Not This, Not That
NINMENI was built by throwing out the mechanisms that cause epistemic dishonesty. Not transformation, but foundation demolition.
- Not a Transformer — No tokens, vocab, next-token prediction, decoder, attention/QKV
- Not Vector Similarity — No cosine, nearest-phrase, embedding-selector as judge
- Not Confidence Score — No trust score, decay, TTL, importance-score, ranking
- Not Registry Lookup — No answer-table lookup during inference
- Not Co-occurrence Judge — Frequency doesn't decide; structural gate is the judge
- Not Self-Anchor — Provenance must be external; no grounding from within
- Not Repair/Healing — NOT AUTHORIZED: connector repair, crystal healing, auto correction
Zero Tolerance Scope Creep
GOVERNANCEFoundation 36 Laws and Phase A-S FROZEN. No new features, no changes, no "small adjustments" without new falsification probe and explicit reviewer decision.
Evolution
From crisis to crisis
Every law is born from experiments that kill assumptions, not defend them. This isn't a feature list, it's a graveyard of wrong ideas.
Total Demolition — Law 1-36
Old NINMENI (transformer + cosine + scores) was completely demolished. From zero, an epistemic foundation was built: meaning = structure, not token. External grounding. Status follows lineage. 36 Laws FROZEN.
Crystal Lifecycle — Audit End-to-End
Kristalisasi connector (v1→v4.2 Anchored). Ledger, lifecycle, ecology, dynamics, containment, routing, suspect propagation, policy unified. Semua di-audit: 17/17 PASS. Observation-only, no repair.
Memory & Causality Era
Memory audit: append-only, immutability, prefix-stability. Topology: explicit structure, not similarity. Causality: cause from fact diff, conflict = effect. Decision trace: Evidence → Episode → Decision.
FREEZE v0.1 — Foundation CLOSED
Foundation sealed. Real-world LOCKED. Three forward tracks awaiting authorization: Real Corpus Transition, Large Knowledge Ecology, Living World-State. Zero tolerance scope creep.
Epistemic Note
FROZEN means: Discovered → Falsified → Fixed → Proven → Frozen = finished, proven, functioning. Not "freeze then ask again every 3 months."
Governance
Binding rules
NINMENI is governed by strict discipline. Not for restriction, but for epistemic honesty.
Re-Audit Prohibited
NO RE-AUDIT of Law 1-36 or Phase A-S unless there's NEW EVIDENCE that directly refutes the freeze. New evidence = new experiment / new observation / new falsification. Not document interpretation / auditor opinion.
Forward Tracks PENDING AUTHORIZATION
- Real Corpus Transition — Synthetic → Controlled Real Corpus (artikel/regulasi/BPS/berita)
- Large Knowledge Ecology — Scale 100k → 1M → 10M relasi. Prove NINMENI lives at scale.
- Living World-State — Time-evolving: day 1→2→30→365. Epistemology reacts to changing evidence.
Disclaimer
Not claims, but foundation
FROZEN What Is Proven
- 36 Epistemic Laws: from meaning as structure to defeasible reasoning.
- Phase A-S audit: 17/17 PASS with strict determinism.
- Memory & Causality: append-only, immutability, cause from fact diff.
- Architecture stable under adversarial evidence.
- Adversarial-resistant guard (Law 35) passed evasion test.
ABSTAIN What Is NOT Claimed
- No claim of SOTA or any benchmark.
- No claim of human-equivalent language understanding.
- No claim that NMUs are "genuine" semantics (just structured structures).
- No claim that real-world implementation is ready.
- No claim of superiority over any LLM.
- No claim this is AGI or "consciousness".
Transformer Replacement?
NOT YET CLAIMED. No evidence yet demands that claim. NINMENI is currently an epistemically honest reasoning substrate stable under defeasible conflict — not a transformer replacement.
Public Research Documents
Sanitized for public release
Internal audit logs, scripts, and full training data are not published. Only documents that can stand alone without misleading context are released here.
NINMENI Epistemic Architecture v0.1
Architecture constitution: 36 epistemic laws from meaning as structure to defeasible reasoning. Contract to be honored before autonomous inquiry.
Internal Only →Crystal Lifecycle & Policy Audit Report
Documentation of 17 audit phases (C-S): ledger, lifecycle, ecology, dynamics, containment, routing, suspect propagation, unified policy. Deterministic, byte-identical.
Internal Only →Memory & Causality Audit Summary
Memory & Causality era: append-only, explicit topology, cause from fact diff. Observation-only, not a hidden repair engine.
Internal Only →Internal Documents
Complete audit logs, probe code, golden files, and training data are not published. Only information that can stand on its own without misleading context is released.
Origin & Lineage
From Indonesia
NINMENI is an independent research project conducted from Indonesia by Paulus Femi Leunufna (Emylton Leunufna). Not affiliated with any university, corporate AI lab, or funding body.
Intellectual lineage: cognitive science, formal semantics, philosophy of mind, and computational linguistics — not from modern scaling-law deep learning tradition. A deliberate choice, not a limitation.
The dominant AI research paradigm structurally cannot ask certain questions about meaning. NINMENI is an attempt to ask them from outside that paradigm, with strict architecture and honest epistemic accountability.
Indonesia-Native: Uses Plus Jakarta Sans, formal Indonesian language, and neat structure (order). Indonesian discipline in tidiness.
Independent Research
GROUNDEDNo ongoing funding guarantee, no institutional backing, no external timeline pressure. Progress is measured by what the architecture teaches us, not by publication deadlines.
Paulus Femi Leunufna (Emylton Leunufna)
Contact
Contact the Lab
This isn't a startup. No sales team. If you have genuine research questions, collaboration proposals, or architectural critique — write.
Important Note
Real-world implementation LOCKED. No demo, no API, no "try it out". The system hasn't been given a passport to the outside world.
Questions about architecture, methodology, or public documents.
Independent researchers, philosophers of mind, cognitive scientists welcome.
Methodological critique, falsifiability challenges.