Vireon V1.0 architecture

1. Current baseline

V1.0 is the current stable baseline. It preserves the frozen compact SentenceCell connection design, deterministic source-order re-reading and connected retrieval across multiple SentenceCells. Summarization experiments are excluded from the active architecture until vocabulary, logic and definition-based meaning are sufficiently mature.

2. WordCells

WordCells are reusable lexical Cells with persistent identity. They may hold lexical forms, definitions and future semantic fields. A WordCell is not a container for every sentence in which the word appears; sentence-instance relationships remain inside SentenceCells.

3. ActionCells

ActionCells are reusable Cells for actions, verbs and verbal forms. They retain action identity while sentence-specific subject, object, tense, aspect, voice and context are represented by the SentenceCell and its proposition data.

4. SentenceCells

SentenceCells are source-specific evidence structures. Each saved SentenceCell preserves the source sentence, stable token references, source order, final role frames and compact sentence-local connection arrays.

The compact connection design stores role positions inside the parent SentenceCell. Token identifiers, semantic identifiers and surfaces are read from the SentenceCell arrays, avoiding expanded duplicate relationship objects.

5. Connection ownership

Reusable WordCell / ActionCell
            ↘
             SentenceCell owns the source relationship
            ↗
Reusable WordCell / ActionCell

No sentence backlinks are required inside reusable WordCells or ActionCells. Reuse is proven because several SentenceCells refer to the same stable Cell identity.

6. Deterministic re-reading

Reading order is fixed by the source. Question order is arbitrary. Canonical, reverse and deterministic shuffled question orders must return the same answers.

7. Multi-SentenceCell retrieval

V1.0 can assemble evidence across connected sentences for entity timelines, problem retrieval, actor action chains, events before or after another event, repair processes, final outcomes and participant retrieval.

This is retrieval, not free-form summarization. The selected evidence remains ordered and traceable to source SentenceCells.

8. Planned semantic fields

key → class → subcategory → category → subdomain → domain

The nodes remain the Cells. These levels are fields or routes associated with Cell meaning. Sentence usage determines the active meaning first; field connections then allow related Cells to activate across narrower or broader levels.

The future temporary answer core may activate inward, outward, laterally or across semantic levels, but it will not become a duplicate persistent knowledge store.

9. Read-only public demonstration

The planned public website demo will use a prebuilt population of several thousand Cells. Public visitors will submit questions against existing Cell knowledge. The public process will be read-only: no new Cell creation, no learning-count increments and no modification of the validated Cell store.

10. Data integrity rules

• Persistent Cell identities remain stable.
• Sentence-specific relationships remain in SentenceCells.
• Re-reading and questioning are read-only operations.
• Every answer should retain its supporting SentenceCell evidence.
• Semantic category fields supplement, rather than replace, sentence evidence.
• Hard-coded summaries are deferred until genuine semantic meaning is available.