Skip to main content
Ctrl K

welcome-to-github

Public portal and research gateway for an edge-first AI research system spanning agent research, evidence engineering, research infrastructure, digital archaeology, and deterministic knowledge lifecycles

1
contributor
Get started
2252 commitsLast commit ≈ 2 days ago0 stars0 forks

Cite this software

Description

lostlight | 2025-2026 Year in Review

Open Research / 开放科研 · Research Template / 科研模板

DOI

"Small and Stable; Quiet and Pragmatic."

Welcome to the digital portal of lostlight, an Edge AI practitioner focusing on Python, Google Vertex AI, MediaPipe, and the Huawei Ascend ecosystems.

My current public work connects edge-first engineering, agent systems, scientific evidence, research infrastructure, and long-horizon technical observation. The repositories below are maintained as independent projects: each keeps its own implementation, evidence boundaries, documentation, and publication identity.

Live Portal

The public presentation layer is available at:

lostlight.github.io/welcome-to-github

The portal is an independent display surface. Repository workflows and research systems run separately.

Public Research and Software Map

RepositoryPublic roleArchived software publication
welcome-to-githubPersonal portal, edge-first research entry point, and deterministic knowledge-lifecycle host10.5281/zenodo.22790907
zero-entropy-labReliability-oriented agent research, memory/state experiments, and evidence-preserving engineering10.5281/zenodo.22791081
Axiom-0Executable reference architecture, formalized contracts, ADRs, methodology, and bounded evaluation10.5281/zenodo.22791103
reflective-continuumReflective graph state, persistence identity, drift analysis, entropy metrics, and runtime evidence10.5281/zenodo.22791141
agent-foundationsBilingual foundational research, canonical source registry, claim/evidence mapping, and reproducibility discipline10.5281/zenodo.22791169
auto-doc-engineStructured research documents, artifact identity, lineage, diagnostics, and research-object packaging10.5281/zenodo.22791404
epistemic-pipelineClaim/evidence processing, provenance, transfer boundaries, runtime policy, and evidence envelopes10.5281/zenodo.22791463
sci-render-kitScientific communication, figure evidence, uncertainty metadata, accessibility, and publisher-target checks10.5281/zenodo.22791375
agentic-frontier-observatoryGlobal agentic-systems observatory covering protocols, runtimes, evaluation, safety, governance, and ecosystems10.5281/zenodo.22791334
china-agentic-observatoryChina-focused observatory covering policy, standards, infrastructure, industry, evaluation, and lifecycle governance10.5281/zenodo.22791309

The Zenodo records identify archived software publications. They provide stable citation targets for those archived publications; they do not replace the evolving state of each repository's current main branch.

2026 Public Milestones

The current public repository record supports a few durable portfolio-level milestones:

  • Ten independent software publications — on 2026-09-16, each repository in the map above received its own Zenodo software publication and DOI rather than being collapsed into one umbrella record.
  • A ten-repository research architecture — the portfolio now spans five foundational/experimental repositories, three research-infrastructure repositories, and two outward-facing observatories, while preserving independent repository identity and evidence boundaries.
  • Long-lived technical knowledge surfaces — Axiom-0 and Reflective Continuum maintain explicit ADR/Methodology systems; Agent Foundations maintains a structured bilingual claim/source/provenance core; the three research-infrastructure repositories maintain active architecture and research-contract layers; the observatories maintain persistent scope, taxonomy, methodology, source, and workstream definitions alongside time-series research.
  • Public citation identity — repository citation metadata links the public author identity to ORCID 0009-0001-3617-0832, while each repository retains its own citation and release metadata.
  • Preserved development history — this repository keeps a separate digital-archaeology layer for earlier project phases instead of rewriting old project narratives to match the current ten-repository system.

These are statements about public repository and publication records. They are not claims of universal scientific validation, product certification, or external adoption.

Repository-Native Guides

The repository itself has several public layers with different roles:

The internal docs/brain/** knowledge-lifecycle implementation remains a separate host subsystem and is not summarized here as if it were the same thing as Horizon research or digital archaeology.

Engineering and Research Principles

Across the public repositories, a few principles remain deliberately stable:

  • Evidence before narrative: implementation, source identity, observed execution, interpretation, and inference are kept separate where they materially differ.
  • Reproducible identity: revisions, artifacts, sources, dates, and runtime observations are identified explicitly rather than inferred from filenames or later state.
  • Bounded claims: a successful command, checker, render, source retrieval, hash, citation, or DOI supports only the property it actually demonstrates.
  • Historical preservation: later corrections refine current interpretation without silently rewriting point-in-time research history.
  • Human-review boundary: automation and AI assistance can research, generate, validate, and maintain, while public evidence and final repository changes remain inspectable and reviewable.
  • Independent repository identity: the ten projects share research concerns but retain their own architecture, contracts, evidence semantics, and release history.

Engineering Standards and Architecture

This repository remains the entry point for an edge-first research system and a deterministic knowledge lifecycle.

  • Bilingual Accessibility: Chinese navigation and English technical content remain readable without requiring a specialized client.
  • Standard-Library Core: The Python lifecycle uses the standard library. Explicitly declared browser assets may be delivered separately by the presentation layer.
  • Append-Only Evidence: Historical evidence is retained under the archive policy while active indexes remain reproducible from canonical ledgers.
  • Validation Boundary: Untrusted observations are accepted only after provenance, format, lifecycle, and human-review checks. This is a repository policy boundary, not a public network security service.
  • Continuous System Pulse: Scheduled and change-triggered checks keep the current repository state testable and auditable.

Citation and Reuse

Each repository exposes its own citation and release metadata where applicable. Use the repository-specific DOI above when citing an archived software publication, and use the relevant repository revision when reproducibility depends on exact implementation state.

A DOI establishes a stable publication identity; it does not by itself establish scientific validity, runtime reproduction, or equivalence with later repository revisions.

Security

Security reports must use the private process described in SECURITY.md.

Do not publish credentials, private data, or exploit details in a public issue.

Digital Museum

The repository also preserves its own development history as a public digital archaeology layer:


© 2026 lostlight | Built for the Edge, Built for the Future.

Keywords
Programming languages
  • Python 71%
  • HTML 19%
  • JavaScript 10%
License
</>Source code

Contributors

Contact person

XJ

Xuanyi Jiang

Independent Researcher and Research Software Maintainer
Independent Researcher
0009-0001-3617-0832Mail Xuanyi
XJ
Independent Researcher and Research Software Maintainer
Independent Researcher
0009-0001-3617-0832