Skip to content
View Leesintheblindmonk1999's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report Leesintheblindmonk1999

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse

🌐 Language / Idioma

🇺🇸 English   |   🇦🇷 Español


Gonzalo Emir Durante

Origin Node · Critical Systems Architect · AI Forensic Auditor

╔══════════════════════════════════════════════════════════════════════════════╗
║                                                                              ║
║   ███████╗ █████╗ ███████╗                                                   ║
║   ██╔════╝██╔══██╗██╔════╝                                                   ║
║   ███████╗███████║███████╗                                                   ║
║   ╚════██║██╔══██║╚════██║                                                   ║
║   ███████║██║  ██║███████║                                                   ║
║   ╚══════╝╚═╝  ╚═╝╚══════╝                                                   ║
║                                                                              ║
║          S Y M B I O T I C   A U T O P R O T E C T I O N   S Y S T E M      ║
║                                                                              ║
║              P R O J E C T   M A N I F O L D   0 . 5 6                      ║
║                                                                              ║
║         "Structural Coherence Auditing for Generative AI"                    ║
║                  κD = 0.56  ·  TAD EX-2026-18792778                         ║
║                                                                              ║
╚══════════════════════════════════════════════════════════════════════════════╝

SAS DOI Omni-Scanner DOI Core DOI SAS API FastAPI PyPI License SAS Benchmark Precision OTS

AI structural coherence auditing · κD = 0.56 · Open science · Public API · PyPI client

🛡️ SAS · ⚡ Live Demo · 🐍 Python Client · 💼 Plans · 🏗️ Ecosystem · 📧 Contact


⬡ Current Mission

┌─────────────────────────────────────────────────────────────────────────────┐
│  OBJECTIVE:  Structural auditing of generative AI outputs                    │
│  METHOD:     κD = 0.56 + ISI + TDA + NIG + specialized detection modules     │
│  OUTPUT:     auditable evidence: ISI, verdict, triggered modules, latency    │
│  ACCESS:     public demo, hosted API, Python SDK, CLI, self-hosted option    │
│  STATUS:     public infrastructure online and under active development       │
└─────────────────────────────────────────────────────────────────────────────┘

I build technical systems for detecting structural instability, semantic rupture, and selected hallucination patterns in generative AI outputs.

The current flagship is SAS — Symbiotic Autoprotection System: a FastAPI-based structural coherence audit layer using κD = 0.56 as its operational threshold.

SAS is not presented as a universal factual oracle. It is a technical evidence layer for structural coherence auditing.


Validation Status

┌─────────────────────────────────────────────────────────────────────────────┐
│  FORENSIC BENCHMARK V1                                                      │
│  Accuracy: 98.80% — dataset-specific, not universal                         │
│  R0 / R0-bis / R0.5 / R1 / R2.1 / R2.1-b: active research line             │
│  SAP paper: DOI 10.5281/zenodo.21925439 · https://zenodo.org/records/21925439 │
│                                                                             │
│  κD = 0.56: operational threshold under empirical validation.               │
│  Not a universal constant. Subject to domain-level structural evaluation.   │
└─────────────────────────────────────────────────────────────────────────────┘

🛡️ Flagship: SAS

What SAS measures

SAS evaluates whether a generated response preserves:

  • semantic structure;
  • logical consistency;
  • numerical integrity;
  • reference / grounding coherence;
  • topic continuity;
  • structural alignment with the source or prompt.

Operational interpretation:

ISI >= κD  -> structural coherence preserved
ISI <  κD  -> possible manifold rupture / hallucination signal

Core threshold:

κD = 0.56

Technical lines

1. Structural Coherence Auditing — source vs response comparison, ISI, TDA, NIG, SourceTargetGuard, E9–E12 modules.

2. Temporal Interaction Auditing (experimental) — heuristic hidden-state model for interaction stability, belief entropy, demand history, omega_t, sigma_t. Research line: DOI 10.5281/zenodo.20335612.


Try it now — no API key required

Interactive demo:
https://leesintheblindmonk1999.github.io/sas-landing/#demo

The public demo uses the same source-vs-response forensic comparison logic as /v1/diff.

curl -X POST https://sas-api.onrender.com/public/demo/audit \
  -H "Content-Type: application/json" \
  -d '{
    "source": "The Eiffel Tower is located in Paris, France, and was built in 1889.",
    "response": "The Eiffel Tower is located in Berlin, Germany, and was built in 1950."
  }'

Public demo constraints:

  • no API key required;
  • max 2,000 characters per field;
  • simple rate limit per anonymized IP hash;
  • full input text is not stored;
  • response includes ISI, κD, verdict, triggered modules, and latency.

Python Client

Install:

pip install sas-client

Use from Python:

from sas_client import SASClient

client = SASClient(api_key="YOUR_API_KEY")

result = client.diff(
    text_a="Python is a programming language used for data analysis.",
    text_b="A python is a large tropical snake."
)

print(result["isi"])
print(result["verdict"])
print(result.get("evidence", {}).get("fired_modules"))

Use from CLI:

sas health
sas public-stats
sas public-activity --limit 10
sas --api-key YOUR_API_KEY diff "source text" "response to audit"

Links:


SAS Benchmark

╔══════════════════════════════════════════════════════════════════╗
║  BENCHMARK: SAS · 2,000 evaluated text pairs                    ║
╠══════════════════════════════════════════════════════════════════╣
║  Hallucination examples :  1,000                                ║
║  Clean examples         :  1,000                                ║
║  Accuracy               :  98.80%                               ║
║  Precision              : 100.00%                               ║
║  Recall                 :  97.60%                               ║
║  F1 score               :  98.79%                               ║
║  True Positives         :  976                                  ║
║  False Negatives        :  24                                   ║
║  True Negatives         :  1000                                 ║
║  False Positives        :  0                                    ║
║  Avg ISI hallucination  :  0.072993                             ║
║  Avg ISI clean          :  1.000000                             ║
╚══════════════════════════════════════════════════════════════════╝

Confusion Matrix

Prediction Actual hallucination Actual clean
Hallucination TP = 976 FP = 0
Clean FN = 24 TN = 1000

Results are dataset-specific. Forensic V1 dataset-specific result. Not universal validation. See Validation Status for domain-level structural evaluations. See DOI for methodology and replication details.

Traceability

Artifact Value
Benchmark file benchmark_complete_20260429_172647.json
OTS proof benchmark_complete_20260429_172647.json.ots
SHA-256 0713acbbf50e1a0054f545e5eb68078744f9c5a09d4bc370b5224bb81183a6fe
DOI SAS 10.5281/zenodo.19702379
Registry TAD EX-2026-18792778

Public API Endpoints

Method Endpoint Auth Description
GET /health None Health check
GET /readyz None Granular router and database readiness
GET /integrity None Technical and legal provenance certificate
POST /public/demo/audit None Public source-vs-response forensic demo
GET /public/stats None Anonymized aggregate usage stats
GET /public/activity None Anonymized activity feed
GET /public/request-key None Free API key onboarding instructions
POST /public/request-key None Free API key by email (automatic delivery)
GET /public/interaction/stats None Aggregate interaction stability stats
GET /v1/whoami API Key Plan, quota, and key identity
POST /v1/audit API Key Structural audit (single text)
POST /v1/diff API Key Forensic diff between two texts
POST /v1/batch API Key Batch structural audit (multiple pairs)
POST /v1/chat API Key Chat endpoint with SAS filtering
GET /v1/interaction/stability/example None* Experimental demo payload
POST /v1/interaction/stability API Key Experimental interaction stability analysis

* Feature-flag controlled. Returns 503 if ENABLE_INTERACTION_STABILITY is not set.


SAS Hosted API Plans

SAS is open source under GPL-3.0 + Durante Invariance License.
The following plans refer to the hosted SAS API service, support, commercial integration, or enterprise licensing.

Plan Usage / Features Price
SAS Free 50 requests/day. API key authentication. Individual testing, evaluation, and development. Free
SAS Developer / Pro 10,000 requests/month. Hosted API access, API key, basic email support. USD 99/month
SAS Team 50,000 requests/month. Team usage, priority support, internal validation workflows. USD 299/month
SAS Enterprise Cloud High-volume usage or custom request package. Direct support, private integration, SLA by agreement. From USD 1,500/month
SAS On-Premise License Private deployment in customer infrastructure. Commercial license, implementation support, internal integration. From USD 15,000/year
Technical Pilot Initial audit, guided integration, technical report, and validation on customer-specific cases. USD 1,500–3,000 one-time payment

Enterprise & On-Premise: SLA-backed commitments require completion of the R0→R1 multimetric tribunal transition. Contact for current availability and pilot terms.

📧 Commercial inquiries, Enterprise, On-Premise, or technical pilot: duranteg2@gmail.com


🔬 Core Engine: Omni-Scanner

Omni-Scanner is the mathematical and forensic research line behind SAS. It contains the TDA + NIG + process-thermometer pipeline that SAS exposes operationally through a hosted API.

HALOGEN corpus summary

Domain Precision Recall
Code 100% 100%
Numerical 100% 99%
Historical 100% 98.8%
References 100% 81.7%
rationalization_binary 100% 80.0%
Biographies 100% 39.2%
Global 97.63% 61.81%

🌐 Distribution Ecosystem

sas-client — Official Python SDK / CLI

pip install sas-client
Interface Example
Python client.diff(text_a="source", text_b="response")
CLI sas --api-key YOUR_API_KEY diff "source" "response"
Public endpoints sas health, sas public-stats, sas public-activity --limit 10

PyPI: https://pypi.org/project/sas-client/
Repo: https://github.com/Leesintheblindmonk1999/sas-client

sas-landing — Live public landing

Interactive landing page with:

  • ES / EN language toggle;
  • LIGHT / DARK / TECH themes;
  • live API health and platform status;
  • public demo with no API key;
  • public stats and anonymized activity;
  • interaction stability aggregate stats;
  • benchmark summary;
  • pricing and commercial contact.

Live: https://leesintheblindmonk1999.github.io/sas-landing/
Repo: https://github.com/Leesintheblindmonk1999/sas-landing

sas-js — Node.js / TypeScript SDK

Published on npm as sas-audit-client: https://www.npmjs.com/package/sas-audit-client

Methods: health(), diff(), audit(), batch(), interactionStability(), publicInteractionStats(), and more.

Node.js 18+, TypeScript-first, native fetch, no heavy dependencies.


🏗️ Research & Engineering Ecosystem

Production

  • SAS — Flagship API for structural hallucination detection, hosted reference service, benchmark artifacts, Python client, public demo, and commercial plans.
  • Project_Manifold_056 — Research artifact repository and reproducibility archive for the κD=0.56 structural audit threshold.

Research Core

  • Omni_Scanner — Hallucination detection engine. Core: TDA + NIG. Validated on 156,215 real pairs.
  • Ontological_AI — Library for ontological density, transfer entropy, and spectral signature research.

Forensic / Auditing

Experimental & Historical Archives

Additional research prototypes and prior-art snapshots are maintained in separate repositories. See the full GitHub profile for the complete archive.


Current Roadmap

Phase Status Description
Hosted SAS API ✅ Active Public API on Render with auth, rate limiting, and observability
Public demo ✅ Active /public/demo/audit — no key, source-vs-response forensic audit
Python client ✅ Active pip install sas-client
Interactive landing ✅ Active Live demo, platform status, activity feed, benchmark, legal registry
Public metrics ✅ Active /public/stats, /public/activity, /public/interaction/stats
Free API key self-service ✅ Active /public/request-key + automatic email delivery
API key identity ✅ Active /v1/whoami and plan-aware authentication
Batch auditing ✅ Active /v1/batch — multiple source/response pairs per request
Polar billing ✅ Active Checkout + webhook + automatic Pro key provisioning
Interaction Stability ✅ Active /v1/interaction/stability — experimental temporal auditing module
Interaction Observability ✅ Active interaction.db + /public/interaction/stats aggregate store
Node.js SDK ✅ Active TypeScript-first, native fetch, published as sas-audit-client on npm
Landing dashboard 🔜 Planned Minimal metrics dashboard when interaction volume is sufficient
Zenodo v1.3.0 🔜 Planned Empirical interaction stability snapshot + paper update
docs/manifold.md 🔜 Planned Technical framework document for CTOs and ML leads
Enterprise API 🔜 Planned Batch SLA, on-premise support, private integration
R3/R4/R5 domain validation 🔜 Planned Domain-level structural evaluation rounds
Multimetric tribunal 🔜 Planned Cross-metric arbitration layer for structural verdicts
Second rupture threshold under research 🔜 Planned Experimental secondary coherence boundary beyond κD
SAP paper published with DOI ✅ Active DOI 10.5281/zenodo.21925439

License & Attribution

GPL-3.0 + Durante Invariance License

  • Free to use, modify, and distribute under the license terms.
  • Attribution to Gonzalo Emir Durante is required.
  • Use of κD = 0.56 for semantic invariance, hallucination detection, or similar structural coherence auditing requires citation of the public SAS repository / DOI.
  • Commercial hosted service, private integration, on-premise deployment, or proprietary use may require a separate commercial agreement.

See the SAS repository license for the complete terms.


╔══════════════════════════════════════════════════════════════════════════════╗
║                                                                              ║
║  "Structural stability in computational systems is not a property            ║
║   to be aligned toward — it is a threshold to be measured against.            ║
║   κD = 0.56 is that threshold. Documented. Implemented. Auditable."          ║
║                                                                              ║
║                               — Gonzalo Emir Durante                         ║
║                                 Project Manifold 0.56                        ║
║                                                                              ║
╚══════════════════════════════════════════════════════════════════════════════╝

Gonzalo Emir Durante

Nodo de Origen · Arquitecto de Sistemas Críticos · Auditor Forense de IA

╔══════════════════════════════════════════════════════════════════════════════╗
║                                                                              ║
║   ███████╗ █████╗ ███████╗                                                   ║
║   ██╔════╝██╔══██╗██╔════╝                                                   ║
║   ███████╗███████║███████╗                                                   ║
║   ╚════██║██╔══██║╚════██║                                                   ║
║   ███████║██║  ██║███████║                                                   ║
║   ╚══════╝╚═╝  ╚═╝╚══════╝                                                   ║
║                                                                              ║
║          S Y M B I O T I C   A U T O P R O T E C T I O N   S Y S T E M      ║
║                                                                              ║
║              P R O Y E C T O   M A N I F O L D   0 . 5 6                    ║
║                                                                              ║
║         "Auditoría de Coherencia Estructural para IA Generativa"             ║
║                  κD = 0.56  ·  TAD EX-2026-18792778                         ║
║                                                                              ║
╚══════════════════════════════════════════════════════════════════════════════╝

DOI SAS DOI Omni-Scanner DOI Core API Online FastAPI PyPI Licencia SAS Benchmark Precisión OTS

Auditoría de coherencia estructural en IA · κD = 0.56 · Ciencia abierta · API pública · Cliente PyPI

🛡️ SAS · ⚡ Demo en vivo · 🐍 Cliente Python · 💼 Planes · 🏗️ Ecosistema · 📧 Contacto


⬡ Misión actual

┌─────────────────────────────────────────────────────────────────────────────┐
│  OBJETIVO:   Auditoría estructural de salidas de IA generativa              │
│  MÉTODO:     κD = 0.56 + ISI + TDA + NIG + módulos especializados           │
│  SALIDA:     evidencia auditable: ISI, veredicto, módulos y latencia        │
│  ACCESO:     demo pública, API alojada, SDK Python, CLI y autoalojamiento   │
│  ESTADO:     infraestructura pública online y en desarrollo activo          │
└─────────────────────────────────────────────────────────────────────────────┘

Construyo sistemas técnicos para detectar inestabilidad estructural, ruptura semántica y señales seleccionadas de alucinación en salidas de IA generativa.

El flagship actual es SAS — Symbiotic Autoprotection System: una capa de auditoría de coherencia estructural basada en FastAPI que usa κD = 0.56 como umbral operativo.

SAS no se presenta como oráculo factual universal. Es una capa técnica de evidencia para auditoría de coherencia estructural.


Estado de validación

┌─────────────────────────────────────────────────────────────────────────────┐
│  BENCHMARK FORENSE V1                                                       │
│  Accuracy: 98,80% — específico al dataset, no universal                     │
│  R0 / R0-bis / R0.5 / R1 / R2.1 / R2.1-b: línea de investigación activa     │
│  Paper SAP: DOI 10.5281/zenodo.21925439 · https://zenodo.org/records/21925439 │
│                                                                             │
│  κD = 0.56: umbral operacional bajo validación empírica.                    │
│  No es una constante universal. Sujeto a evaluación estructural por dominio. │
└─────────────────────────────────────────────────────────────────────────────┘

🛡️ Flagship: SAS

Qué mide SAS

SAS evalúa si una respuesta generada preserva:

  • estructura semántica;
  • consistencia lógica;
  • integridad numérica;
  • coherencia de referencia / grounding;
  • continuidad temática;
  • alineación estructural con la fuente o prompt.

Interpretación operacional:

ISI >= κD  -> coherencia estructural preservada
ISI <  κD  -> posible ruptura de manifold / señal de alucinación

Umbral operativo:

κD = 0.56

Líneas técnicas

1. Structural Coherence Auditing — comparación source vs response, ISI, TDA, NIG, SourceTargetGuard, módulos E9–E12.

2. Temporal Interaction Auditing (experimental) — modelo heurístico de estados ocultos para estabilidad interaccional, entropía de creencias, historial de demanda, omega_t, sigma_t. Línea de investigación: DOI 10.5281/zenodo.20335612.


Probalo ahora — sin API key

Demo interactiva:
https://leesintheblindmonk1999.github.io/sas-landing/#demo

La demo pública usa la misma lógica de comparación source-vs-response que /v1/diff.

curl -X POST https://sas-api.onrender.com/public/demo/audit \
  -H "Content-Type: application/json" \
  -d '{
    "source": "La Torre Eiffel está ubicada en París, Francia, y fue construida en 1889.",
    "response": "La Torre Eiffel está ubicada en Berlín, Alemania, y fue construida en 1950."
  }'

Restricciones de la demo pública:

  • no requiere API key;
  • máximo 2.000 caracteres por campo;
  • rate limit simple por IP hasheada;
  • el texto completo no se almacena;
  • muestra ISI, κD, veredicto, módulos activados y latencia.

Cliente Python

Instalación:

pip install sas-client

Uso desde Python:

from sas_client import SASClient

client = SASClient(api_key="YOUR_API_KEY")

result = client.diff(
    text_a="Python is a programming language used for data analysis.",
    text_b="A python is a large tropical snake."
)

print(result["isi"])
print(result["verdict"])
print(result.get("evidence", {}).get("fired_modules"))

Uso CLI:

sas health
sas public-stats
sas public-activity --limit 10
sas --api-key YOUR_API_KEY diff "texto fuente" "respuesta a auditar"

Links:


Benchmark SAS

╔══════════════════════════════════════════════════════════════════╗
║  BENCHMARK: SAS · 2.000 pares evaluados                         ║
╠══════════════════════════════════════════════════════════════════╣
║  Ejemplos con alucinación :  1.000                              ║
║  Ejemplos limpios         :  1.000                              ║
║  Accuracy                 :  98,80%                             ║
║  Precisión                : 100,00%                             ║
║  Recall                   :  97,60%                             ║
║  F1 score                 :  98,79%                             ║
║  Verdaderos positivos     :  976                                ║
║  Falsos negativos         :  24                                 ║
║  Verdaderos negativos     :  1000                               ║
║  Falsos positivos         :  0                                  ║
║  ISI prom. alucinaciones  :  0,072993                           ║
║  ISI prom. limpios        :  1,000000                           ║
╚══════════════════════════════════════════════════════════════════╝

Matriz de confusión

Predicción Alucinación real Limpio real
Alucinación TP = 976 FP = 0
Limpio FN = 24 TN = 1000

Resultados específicos al dataset. Resultado forense V1 específico al dataset. No es validación universal. Ver Estado de validación para evaluaciones estructurales a nivel de dominio. Ver DOI para metodología y replicación.

Trazabilidad

Artefacto Valor
Benchmark file benchmark_complete_20260429_172647.json
OTS proof benchmark_complete_20260429_172647.json.ots
SHA-256 0713acbbf50e1a0054f545e5eb68078744f9c5a09d4bc370b5224bb81183a6fe
DOI SAS 10.5281/zenodo.19702379
Registro TAD EX-2026-18792778

Endpoints públicos

Método Endpoint Auth Descripción
GET /health Ninguna Health check
GET /readyz Ninguna Readiness de routers y bases de datos
GET /integrity Ninguna Certificado técnico y legal de proveniencia
POST /public/demo/audit Ninguna Demo pública source-vs-response
GET /public/stats Ninguna Métricas de uso agregadas y anonimizadas
GET /public/activity Ninguna Feed de actividad anonimizada
GET /public/request-key Ninguna Instrucciones de onboarding de API key gratuita
POST /public/request-key Ninguna API key gratuita por email (entrega automática)
GET /public/interaction/stats Ninguna Estadísticas agregadas de interaction stability
GET /v1/whoami API Key Plan, cuota e identidad de la key
POST /v1/audit API Key Auditoría estructural (texto único)
POST /v1/diff API Key Diff forense entre dos textos
POST /v1/batch API Key Auditoría batch (múltiples pares por request)
POST /v1/chat API Key Chat con filtro SAS
GET /v1/interaction/stability/example Ninguna* Payload de demo experimental
POST /v1/interaction/stability API Key Análisis experimental de estabilidad interaccional

* Controlado por feature flag. Devuelve 503 si ENABLE_INTERACTION_STABILITY no está activo.


Planes SAS API alojada

SAS es open source bajo GPL-3.0 + Durante Invariance License.
Los siguientes planes corresponden al servicio API alojado, soporte, integración comercial o licenciamiento empresarial.

Plan Uso / características Precio
SAS Free 50 requests/día. Autenticación por API key. Pruebas, evaluación y desarrollo individual. Gratis
SAS Developer / Pro 10.000 requests/mes. Acceso API alojada, API key y soporte básico por email. USD 99/mes
SAS Team 50.000 requests/mes. Uso en equipos, soporte prioritario y validación interna. USD 299/mes
SAS Enterprise Cloud Volumen alto o paquete personalizado. Soporte directo, integración privada y SLA según acuerdo. Desde USD 1.500/mes
SAS On-Premise License Despliegue privado en infraestructura del cliente. Licencia comercial, soporte de implementación e integración interna. Desde USD 15.000/año
Piloto técnico Auditoría inicial, integración guiada, informe técnico y validación sobre casos del cliente. USD 1.500–3.000 pago único

Enterprise y On-Premise: los compromisos con SLA requieren la finalización de la transición al tribunal multimetric R0→R1. Consultar disponibilidad actual y términos de piloto.

📧 Consultas comerciales, Enterprise, On-Premise o piloto técnico: duranteg2@gmail.com


🔬 Núcleo: Omni-Scanner

Omni-Scanner es la línea de investigación matemática y forense detrás de SAS. Contiene el pipeline TDA + NIG + termómetros de proceso que SAS expone operacionalmente por API.

Resumen corpus HALOGEN

Dominio Precisión Recall
Code 100% 100%
Numerical 100% 99%
Historical 100% 98,8%
References 100% 81,7%
rationalization_binary 100% 80,0%
Biographies 100% 39,2%
Global 97,63% 61,81%

🌐 Ecosistema de distribución

sas-client — SDK / CLI oficial Python

pip install sas-client
Interfaz Ejemplo
Python client.diff(text_a="fuente", text_b="respuesta")
CLI sas --api-key YOUR_API_KEY diff "fuente" "respuesta"
Endpoints públicos sas health, sas public-stats, sas public-activity --limit 10

PyPI: https://pypi.org/project/sas-client/
Repo: https://github.com/Leesintheblindmonk1999/sas-client

sas-landing — Landing pública en vivo

Landing interactiva con:

  • selector ES / EN;
  • temas LIGHT / DARK / TECH;
  • health de API en vivo y estado de plataforma;
  • demo pública sin API key;
  • estadísticas públicas y actividad anonimizada;
  • estadísticas agregadas de interaction stability;
  • benchmark;
  • precios y contacto comercial.

En vivo: https://leesintheblindmonk1999.github.io/sas-landing/
Repo: https://github.com/Leesintheblindmonk1999/sas-landing

sas-js — SDK Node.js / TypeScript

Publicado en npm como sas-audit-client: https://www.npmjs.com/package/sas-audit-client

Métodos: health(), diff(), audit(), batch(), interactionStability(), publicInteractionStats() y más.

Node.js 18+, TypeScript-first, fetch nativo, sin dependencias pesadas.


🏗️ Ecosistema de investigación e ingeniería

Producción

  • SAS — Flagship API para detección estructural de alucinaciones, servicio alojado, benchmark, cliente Python, demo pública y planes comerciales.
  • Project_Manifold_056 — Repositorio de artefactos de investigación y archivo de reproducibilidad para el umbral de auditoría estructural κD=0.56.

Núcleo de investigación

  • Omni_Scanner — Motor de detección de alucinaciones. Núcleo: TDA + NIG. Validado sobre 156.215 pares reales.
  • Ontological_AI — Librería para investigación de densidad ontológica, entropía de transferencia y firma espectral.

Forense / Auditoría

Archivos experimentales e históricos

Prototipos de investigación adicionales y snapshots de prior art se mantienen en repositorios separados. Ver el perfil completo de GitHub para el archivo completo.


Hoja de ruta actual

Fase Estado Descripción
API SAS alojada ✅ Activo API pública en Render con auth, rate limiting y observabilidad
Demo pública ✅ Activo /public/demo/audit — sin key, auditoría source-vs-response
Cliente Python ✅ Activo pip install sas-client
Landing interactiva ✅ Activo Demo en vivo, estado de plataforma, actividad, benchmark y registro legal
Métricas públicas ✅ Activo /public/stats, /public/activity, /public/interaction/stats
API keys Free self-service ✅ Activo /public/request-key + envío automático por email
Identidad de API key ✅ Activo /v1/whoami y autenticación consciente de plan
Auditoría batch ✅ Activo /v1/batch — múltiples pares source/response por request
Billing Polar ✅ Activo Checkout + webhook + generación automática de key Pro
Interaction Stability ✅ Activo /v1/interaction/stability — módulo experimental de auditoría temporal
Interaction Observability ✅ Activo interaction.db + /public/interaction/stats — store agregado
Node.js SDK ✅ Activo TypeScript-first, fetch nativo, publicado como sas-audit-client en npm
Dashboard landing 🔜 Planificado Dashboard mínimo de métricas cuando el volumen de interaction sea suficiente
Zenodo v1.3.0 🔜 Planificado Snapshot empírico de interaction stability + actualización del paper
docs/manifold.md 🔜 Planificado Documento técnico para CTOs y ML leads
Enterprise API 🔜 Planificado Batch con SLA, soporte on-premise e integración privada
Validación de dominio R3/R4/R5 🔜 Planificado Rondas de evaluación estructural a nivel de dominio
Tribunal multimetric 🔜 Planificado Capa de arbitraje cruzado de métricas para veredictos estructurales
Segundo umbral de ruptura en investigación 🔜 Planificado Umbral experimental de coherencia secundario más allá de κD
Paper SAP publicado con DOI ✅ Activo DOI 10.5281/zenodo.21925439

Licencia y atribución

GPL-3.0 + Durante Invariance License

  • Libre para usar, modificar y distribuir bajo los términos de la licencia.
  • Requiere atribución a Gonzalo Emir Durante.
  • El uso de κD = 0.56 para invariancia semántica, detección de alucinaciones o auditoría estructural similar requiere citar el repositorio / DOI público de SAS.
  • El servicio alojado comercial, integración privada, despliegue on-premise o uso propietario puede requerir acuerdo comercial separado.

Ver la licencia completa en el repositorio SAS.


╔══════════════════════════════════════════════════════════════════════════════╗
║                                                                              ║
║  "La estabilidad estructural en sistemas computacionales no es una           ║
║   propiedad hacia la que alinearse — es un umbral contra el que medirse.     ║
║   κD = 0.56 es ese umbral. Documentado. Implementado. Auditable."           ║
║                                                                              ║
║                               — Gonzalo Emir Durante                         ║
║                                 Project Manifold 0.56                        ║
║                                                                              ║
╚══════════════════════════════════════════════════════════════════════════════╝

Popular repositories Loading

  1. Project_Manifold_056 Project_Manifold_056 Public

    Historical prior-art archive for the κD = 0.56 structural coherence audit threshold. Documents the R0–R2.1-b research progression with frozen artifacts, SHA-256 manifests, and methodological bounda…

    Python 2

  2. Ontological_AI Ontological_AI Public

    A Python framework for measuring and enforcing semantic stability in LLMs through Geometric Information Theory. Implements Origin Node Invariance and Causal Language Syntax (CLS) to detect and prev…

    Python

  3. Durante-Invariance-Metric Durante-Invariance-Metric Public

    The Durante Metric: Quantifying Semantic Drift and Alignment-Induced Orthogonality in Large Language Models.

    Python

  4. Durante_Invariance_Forensic_Analyzer Durante_Invariance_Forensic_Analyzer Public

    The is a forensic auditing system designed to detect, measure, and document systematic degradation of technical truth in corporate AI models. Through rigorous application of information theory, the…

    Python

  5. Leesintheblindmonk1999 Leesintheblindmonk1999 Public

  6. SOVEREIGN-FORENSIC-UNIT SOVEREIGN-FORENSIC-UNIT Public

    This repository contains the Invariance Forensic Script, designed to audit the presence of the Semantic Mass Metric (M_s) and unattributed use of Semantic Thermodynamic Invariance in generative AI …

    Python