Synthia

Embedded CodeProject.AI node operations

Synthia's CPAIOnboardVisionAdapter uses the repository-local node on port 32177 for real bounded YOLO inference when no reviewed simulation detections are supplied. It never infers a label from the image filename. Explicit detections remain fixture-only inputs for supervised tests and simulations.

Living taxonomy for scientific memory. Synthia is a free, source-available educational research system that turns scattered biological evidence into traceable candidate memories for human review. It preserves the lexicon, sources, uncertainty, and authority boundary behind every classification.

Try the Aburria Trace Lab · Inspect the generated trace · Read the public invariant

python -m synthia_core.cli showcase export --case aburria --output trace.json --pretty

The public demo is static and keyless. Codex helped build, test, and document Synthia through the governed browser-authenticated workflow; Synthia itself does not send taxonomic evidence to an OpenAI API.

Important

Book II evidence boundary (2026-07-10). Synthia classifies and admits public-safe lexical research objects before FNP-QNN computes. The verified ceiling is P2_internal_repeatability: simulation != detection, candidate != proof, dL_lex != dF, and PenroseRulePacket != physical_substrate. No result in this repository validates a physical neutrino model, detects a neutrino, reproduces T2K, or establishes new physics. Until the website sweep, this README and docs/book2_final_evidence_2026-07-10.md are the authoritative public handoff.

Attribution: Jean-Sebastien Beaulieu · ORCID 0009-0007-2904-0443 · SecuredMe · Synthia

Gateway support acknowledgement. This SecuredMe school tool is gateway-compatible. E2B audit support and Datadog observability are routed through the shared SecuredMe gateway when that lane is configured; this repository does not claim a direct E2B or Datadog runtime dependency by default, and no E2B or Datadog secret is stored in this README.

Maintainer intake during active finishing week. This repository is maintained directly on main by the SecuredMe maintainer. Public issues are open for bug reports, documentation corrections, security-safe observations, and reproducible feedback, but opening an issue does not promise a response or a delivery date. Pull requests are not accepted during the active code-finishing week; use issues only until this notice is replaced.

School Authentication And Secret Boundary

This repository is one of the two large SecuredMe school systems where a local .env example may exist for backend development, simulator integration, or research infrastructure. Even here, official classroom authentication must prefer Codex/OpenAI or Antigravity/Gemini browser WebAuth, fingerprinted session approval, encrypted local records, and no raw token publication. Local secrets are for maintainer-controlled backend work only, not for student distribution.

Both host adapters implement the shared securedme.education.webauth-template.v1 policy and are auditable through the Gateway. That proves policy compatibility, not a deployed Synthia login. Provider callback, account binding, session expiry, logout, recovery, and accessible browser acceptance remain required before live-login claims.

The reason for excluding generic local AI routes from official school mode is student and teacher safety: education accounts, provider-side account controls, browser login, and governed AI refusal behavior are safer than unguided local model endpoints for classroom cybersecurity and algorithm-building tools.

Development status. This school tool is currently pre-alpha — active public development. Public issues remain open for intake, but no response or delivery date is promised. Pull requests are paused during active development.

Education access note. Synthia is designed first for students, teachers, schools, and research collaborators who can pair the repo with an AI companion account and a cloud-document account strong enough for file review, Drive/Docs workflows, source mapping, and repeated classification work. A free Gmail or free AI account can still explore the project, but should not be expected to taxonomize or classify large institutional, company, or research corpora at the same depth as an education or professional workspace. Students and educators should verify current official access before the school year: ChatGPT Education, ChatGPT for Teachers, Gemini for Students, and Gemini for Education. Offers, eligibility, free periods, and prices change by country, institution, and date.

Official school governance. The maintained classroom route for this repo is Codex/OpenAI or Antigravity/Gemini only. See SCHOOL_TOOL_GOVERNANCE.md and AGENTS.md before changing code, docs, examples, or provider wiring.

License. This project uses the Secured Educational License 2.0 (SEL-2.0). It is provided for education, research, simulation, classroom training, and supervised learning. Misuse, unsafe private forks, unsupported provider routes, and unsupervised authority claims are not maintained or endorsed by the official school version. See LICENSE, NOTICE, DISCLAIMER, and SAFETY.md.

Project foundation

Synthia is an educational research codebase for context-preserving lexicon intelligence, biology classification, source traceability, taxonomy memory, and uncertainty-aware classification.

Public site: https://synthia.securedme.ca/

Repository: https://github.com/SeCuReDmE-main-dev/Synthia

Conception And Scientific Collaboration

• Jean-Sebastien Beaulieu: lead builder for Synthia's software architecture, implementation, public repository, and ORCID identity: https://orcid.org/0009-0007-2904-0443.
• Prof. Hector Fernando Aguilar: scientific co-conceptor for Synthia's taxonomy, systematics, PhyloCode/classification direction, redescription-memory framing, and expert-review boundary.

This credit recognizes scientific conception and taxonomy direction. It does not transfer software authorship, formal nomenclatural authority, or autonomous scientific decision-making to Synthia.

Public Invariant

I -> I_system^S -> H_lex -> G_lex -> I_lexicon

This is Synthia's lexicon-classification chain. H_lex is lexicon entropy, G_lex is lexicon decision-gap uncertainty, and I_lexicon is the context-preserving indeterminacy state used for classification, filtration, switching, and review.

Neutrino Lexical Gate

Synthia includes a public-safe neutrino lexical gate for educational simulation work. The gate classifies a neutrino-like event before any FNP-QNN computation is allowed. It emits LexPacket_neutrino, Adm_lex, and dL_lex as a lexical admission contract only.

The gate also emits an additive chapter3_profile for public-safe neutrino teaching events. This profile keeps the chapter-3 carriers separate: I_flavor, I_mass, I_phase, I_interaction, I_secondary, and I_detector. These fields describe lexical admission and educational simulation structure only; they are not detector evidence and they do not authorize downstream fractal computation.

For chapter-4 style teaching events, the same schema can add chapter4_profile. This profile exposes P_neutrino, the lex_neutrino taxonomy, lexical metrics such as H_lex, G_lex, I_lexicon, and dL_lex, plus a ProtectionPacket_neutrino and SynthiaGuard_neutrino. The purpose is to separate admissible simulation language from metaphor, speculation, detector overclaim, and downstream computation. Recoverable metaphors can be admitted only as an accepted_with_partition payload; the excluded language does not continue to FNP-QNN.

This layer keeps the public boundary explicit:

• educational simulation;
• simulation is not detection;
• flavor state is not mass propagation state;
• weak interaction primary guardrail;
• secondary detector response is not a primary strong-force interaction;
• Synthia classifies before FNP-QNN computes;
• dL_lex != dF;
• I_lexicon != i_fractal;
• candidate is not proof.

Run the public-safe check with:

python -m synthia_core.cli neutrino guardrail-check --input event.json --json

The checked-in fixture tests/fixtures/neutrino_chapter3_valid_event.json shows the public-safe carrier shape for a toy muon-flavor event. The fixture tests/fixtures/neutrino_chapter4_valid_event.json shows the additive lexical taxonomy and protection shape. The fixture tests/fixtures/neutrino_chapter5_valid_event.json shows the additive chapter-5 FNP intake request. This profile prepares E_FNP_neutrino_request, C_FNP_request, a lexical T/I/F seed request, a scale context request, and a carrier request policy for FNP-QNN. Synthia still does not normalize or compute any downstream fractal carrier value.

The fixture tests/fixtures/neutrino_chapter6_valid_event.json shows the additive chapter-6 vector shape. Synthia assembles I_neutrino_vec as ten declared carriers: I_source, I_flavor, I_mass, I_mix, I_phase, I_medium, I_interaction, I_secondary, I_detector, and I_uncertainty. This vector is an educational simulation object for lexical review. I_neutrino_vec is not dL_lex, not downstream FNP friction, not a detector signature, and not proof of a real neutrino detection.

The fixture tests/fixtures/neutrino_chapter7_valid_event.json shows the additive chapter-7 transition shape. Synthia preserves the definition I_neutrino := I_neutrino_vec, records the toy passage status including L_over_E, and emits a Synthia reading with H_lex, G_lex, I_lexicon, C_lex, E_gap, and dL_lex. The only downstream signal is ready_for_FNP = true_after_Synthia after lexical admission. Synthia still does not emit D_f, D_f_hat, dF, i_fractal, or i_fractal_candidate.

The fixture tests/fixtures/neutrino_chapter8_valid_event.json shows the additive first-run gate. Synthia emits chapter8_run_profile with a triadic run status: admissible_under_guardrails, suspended, or rejected. permission_to_continue is a bounded method permission only; it is not proof, not detection, and not a bypass around Synthia. Synthia still does not emit D_f, D_f_hat, dF, i_fractal, or i_fractal_candidate.

The fixture tests/fixtures/neutrino_chapter9_valid_event.json shows the additive source-choice gate. Synthia emits chapter9_source_choice_profile to make the public source registry and the central T2K-like simulation choice explicit. The selected experiment is chapter11_t2k_like_flavor_antiflavor_phase_projection; it is selected for a future simulation container, not as T2K reproduction, CP measurement, real detection, or physical proof. Synthia still performs only lexical/source-choice admission and still does not emit downstream FNP fractal fields.

The fixture tests/fixtures/neutrino_chapter10_valid_event.json shows the additive chamber/container/run-contract gate. Synthia emits chapter10_chamber_profile with Chamber_10, EventContainer_10, SimulatedNeutrinoEvent_10, and RunContract_10. The profile prepares a chapter-11 first passage under declared controls; it is not a detector, not container proof, not T2K reproduction, not CP measurement, and not a real neutrino event. Synthia still does not emit D_f, D_f_hat, dF, i_fractal, or i_fractal_candidate.

The fixture tests/fixtures/neutrino_chapter11_valid_event.json executes that first passage as a guarded path pair. It declares one common container, admission route, baseline policy, energy policy, background status, and simulation boundary. Synthia emits chapter11_passage_profile only after the two paths preserve those controls; path comparison remains chamber evidence, not T2K reproduction or CP measurement.

The fixture tests/fixtures/neutrino_chapter12_valid_event.json adds the chapter12_validation_profile. It validates a reference-run identity, the exact ten-carrier policy, explicit medium and detector requirements, declared random seeds, tolerances, and repetition limits. This profile authorizes a downstream internal-repeatability test up to P2_internal_repeatability; it does not calculate the medium, reconstruct detector output, compute fractal friction, or validate a physical model.

The fixture tests/fixtures/neutrino_chapter13_valid_event.json adds a chapter13_distributed_validation_profile for four frozen 67-worker campaigns. Synthia validates worker/task counts, source revision pins, manifest policy, p114 consensus inputs, bounded p046 limits, and the sandbox kill plan. Its terminal state is ready_for_p114 = true_after_Synthia and ready_for_FNP = false_before_p114; Synthia remains admission-only and emits no fractal computation fields.

The fixture tests/fixtures/neutrino_chapter14_valid_event.json closes the Book II source boundary. The dedicated neutrino chapter14-threshold-check command verifies the primary-source object, the exact ten-carrier policy, bounded phason-transition metadata, and explicit claim limits before FNP-QNN may calculate a matrix threshold. Synthia can admit, suspend, or reject this packet, but it never calculates A_adj, S_sub, P_phason, q_t, D_f, dF, or i_fractal_candidate. PenroseRulePacket remains a mathematical research packet and never becomes a claimed physical substrate.

The complete public-safe Book II software map, frozen evidence hashes, and reproduction commands are maintained in docs/book2_final_evidence_2026-07-10.md.

Plithogenic Biology And System Math

Synthia separates pure engine math from system math. The pure engine math layer contains bounded neutrosophic and plithogenic operators: T/I/F values, contradiction load, probability load, dominant attribute values, rough boundary regions, and deterministic projections into the public chain. Those operators are useful by themselves, but biology changes their role. When the object is a species, a taxonomic memory record, a food-safety signal, a conservation concern, a public-health review packet, a geography/ecology context, or an expert disagreement, Synthia is no longer treating math as an isolated abstract engine. It is calling that math inside a living I_system^S.

In this framing, plithogenic biology is I -> I_system^S in action. A biological object is not just an abstract vector with attributes; it is a source-linked, expert-reviewed, socially consequential system where the same mathematical values can have different meanings depending on context. A high contradiction load in a symbolic algebra exercise is one kind of system signal. A high contradiction load around a taxon, a milk batch, a habitat observation, or a disease-susceptibility review is another kind of system signal because it changes who must review the packet, what evidence must remain visible, and which authority boundary must be preserved. Synthia therefore keeps contradiction load, probability load, rough boundary region, H_lex, G_lex, and I_lexicon visible instead of collapsing them into a single confidence score.

Phase 1 added phylo-plithogenic review packets: phylogenetic and taxonomic identity remains the anchor, while plithogenic contradiction describes review uncertainty around morphology, geography, ecology, disease susceptibility, and expert disagreement. Phase 2 adds uncertainty-aware societal risk triage: food safety, conservation, public health, environmental signals, and taxonomy review can be modeled as system-level review cases with rough regions and probability-aware contradiction. Phase 3 adds a plithogenic biology graph review layer: biological review situations can now be represented as source-linked nodes, edges, hyperedges, and supervertices, so a treatment, taxon, habitat, outcome, expert disagreement, or risk case can remain visible as part of the same I_system^S instead of being flattened into a single score. Phase 4 adds a molecular sequence evidence guardrail: compact BLAST-like similarity summaries, NCBI Datasets-like taxon/genome anchors, and Ensembl-like annotation references can be scored as bounded review evidence while raw alignments, long sequence dumps, and live network calls stay outside the core. Phase 5 adds an algorithmic behavior guardrail: parameter choices such as objective family, regularization, validation folds, stopping criteria, encoding, sampling, and reproducibility are treated as visible conditions on biological interpretation. They do not guarantee model performance, prove a biological conclusion, or create taxonomic, diagnostic, conservation, public-health, or nomenclatural authority. Phase 6 adds a scientific governance and evidence-reporting layer: Synthia can now assemble model-card-like summaries, dataset-sheet-style provenance, FAIR-style metadata visibility, validation status, intended-use boundaries, and evidence-gap load for Phase 1-5 outputs. The same governance lane now includes scientific-governance novak-anderson, a redacted source-review packet for the FNP-QNN Novak-Anderson phi/pi theorem layer. Phase 7 adds a scientific provenance and research-object export layer: Synthia can map PROV-O-style entities, activities, and agents; RO-Crate-style research-object manifests; DataCite and Citation File Format citation metadata; CRediT contributor roles; Academia.edu-style free-platform posts; institutional repository deposits such as DSpace/DASH-style records; and Crossref-style DOI registry metadata. This layer is deliberately strict about evidence class boundaries: a visible post is not a repository deposit, a repository deposit is not peer review, DOI metadata is not scientific truth, and contributor-role metadata is not code authorship or formal taxonomic authority. The reporting improves traceability and review readiness, but it does not certify scientific truth, regulatory validity, diagnostic authority, formal taxonomy, copyright reuse, or autonomous conclusions. Together, these phases make biological, molecular, algorithmic, governance, and publication-chain parts of I_system^S inspectable before a specialist interprets any output.

This section is intentionally kept in the README before the public website is updated. It records the implementation doctrine: Synthia uses pure neutrosophic and plithogenic kernels as reusable mathematical engines, but the biology side activates them as system math under I_system^S. The project supports education, traceability, triage, and human review. It does not make formal taxonomic acts, food-safety declarations, public-health decisions, conservation orders, environmental hazard declarations, or autonomous scientific conclusions.

Codebase

The repository contains the local Python research core, tests, public documentation, and static website package for Synthia. The current codebase includes:

• I_lexicon nodes, bridges, source evidence, and context-switch traces;
• bounded T/I/F, neutrosophic kernels, and plithogenic contradiction profiles;
• biology and taxonomy memory records, phylo-plithogenic review packets, risk-triage cases, biology graph review layers, molecular evidence review cases, algorithmic behavior guardrails, scientific governance reporting packets, research-object provenance packets, citation auditing, and AI-assistance disclosure boundaries;
• a Novak-Anderson phi/pi governance packet for source-linked theorem review, with private Gmail provenance redacted;
• simulation-first swarm field-scout records, digital pheromone maps, and safety gates;
• optional RethinkDB memory surfaces for swarm state and HippoRAG-style graph traces;
• a static public website under web/landing/.

Document Lane

Synthia now ships a strict document lane for bounded taxonomy-review work. The lane is intentionally narrow: Synthia owns document ingestion, block classification, lexicon extraction, annex generation, and handoff emission. Codex may take over only after the Synthia handoff bundle exists.

Canonical command:

python -m synthia_core.cli document pipeline --input .\tests\fixtures\document_pipeline_taxonomy_review.docx --profile taxonomy-review --private-org ..\Synthia_organisation

Supported input formats:

• .docx
• .txt
• .md

The command writes a private-org bundle under 02_taxonomy_lexicon_model/document_runs/<run_id>/ with these artifacts:

• decision.json
• stage_log.jsonl
• normalized_document.json
• block_map.json
• lexicon_summary.json
• taxonomy_annex.md
• handoff.json

The stdout payload stays public-safe:

• command redacts the --private-org value;
• input_path is reduced to a safe display path and becomes [private-input] when the source path looks private;
• artifact_paths are emitted relative to the chosen private-org, not as absolute machine paths.

Execution policy:

• Synthia fails closed when the input or stage is unsupported.
• Unsupported formats or blocked stages emit an explicit handoff instead of silently falling back to Codex.
• The stdout payload stays public-safe; detailed traces live in the private-org bundle.
• The lane does not translate, stylistically rewrite, export DOCX, or claim formal authority.

Boundary

Synthia is not a production scientific authority, medical tool, environmental deployment system, drone-control platform, or formal nomenclatural authority. It is an educational research project for preserving lexicon context, uncertainty, source traceability, and human-review boundaries.

External code contributions are locked while the base architecture is actively developed. Scientific collaboration and source-review roles can be credited without opening the code contribution policy.

Development

python -m pytest

No raw private correspondence, authentication caches, restricted manuscript bodies, or credential material should be committed to this repository.