# Polaris > Polaris (Звёздочка, Zvezdochka) is an independent research project building a digital organism that can sustain a life, learn from its own experience, preserve acquired knowledge and repair itself. This is a research goal, not a claim of consciousness, AGI or demonstrated general autonomous self-improvement. Canonical website: https://polaris.spectorn.ai/ Languages: Russian and English. Contact: polaris@spectorn.ai. Public site snapshot: 16 November 2026. The recorded astra1 state is from 14 November 2026, UTC−08; this is not live telemetry. The project distinguishes a real persistent native life, isolated component qualification, research testbeds and a browser illustration. These are different evidence levels. Testbed survival improvements are not automatically capabilities of the current native organism. The role and added benefit of the NCA field are not generally established; useful behaviour can be explained by a conventional controller. The site reports negative and inconclusive results alongside positive ones. ## Start here - [Russian project page](https://polaris.spectorn.ai/): purpose, research history, anatomy, results, limitations and collaboration. - [English project page](https://polaris.spectorn.ai/en/): English version of the same public account. - [Russian text](https://polaris.spectorn.ai/index.md): readable text generated from the published page, including results and controls. - [English text](https://polaris.spectorn.ai/en/index.md): text generated from the English page. - [Complete public text bundle](https://polaris.spectorn.ai/llms-full.txt): both project pages and all twelve published evidence excerpts. It contains public material only, not the repository or full raw evidence. ## Current native life and recovery The latest accepted astra1 snapshot is tick 436, 194 ATP, paused. Sixty paid repairs are recorded in its history. NTHRET1 preserved a paid calculation through a prescribed crash; the researcher prescribed Think, and the recommendation was not applied. CENV1 capsule-order qualification covers a test executable, one saved historical state and Idle. Production-binary qualification, power-loss durability, a live writer validation failure and interruption during event adoption remain open. There is no new native tick implied by publishing this website. - [NTHRET1](https://polaris.spectorn.ai/evidence/nthret1.md): paid Think, 7,236 operations, crash recovery and limits of the claim. - [CENV1](https://polaris.spectorn.ai/evidence/cenv1.md): capsule-order component qualification and remaining recovery boundaries. ## Research evidence These are selected saved reports and protocols. Dates, exact controls and SHA256 source pins are included. They are not a new run or an independent replication of every claim. Historical report wording is preserved; the main project page provides corrections and qualifications. - [KDRY1](https://polaris.spectorn.ai/evidence/kdry1.md): changing environments, map revision and survival; generation-four comparison concerns reaction to a loss. - [KBLIND1](https://polaris.spectorn.ai/evidence/kblind1.md): cost of periodic map revision in a stationary world. - [KTHIRST3](https://polaris.spectorn.ai/evidence/kthirst3.md): exploration and held goals, including their survival cost. - [KTIME2](https://polaris.spectorn.ai/evidence/ktime2.md): inherited season knowledge and uncertain survival benefit. - [KPOISON1](https://polaris.spectorn.ai/evidence/kpoison1.md): poison-history inheritance; no demonstrated positive survival gain in the reported comparison. - [KSEVEN2](https://polaris.spectorn.ai/evidence/kseven2.md): policy revision comparison; reported normal interval and exact p = 0.071 are different statistical summaries. - [KCIRC1](https://polaris.spectorn.ai/evidence/kcirc1.md): feeding-circuit geometry and costs. The formula was derived after earlier observations, then fixed for fresh worlds. One income cell differs by 0.07, so universal agreement within 0.03 is not claimed. - [KLAW1](https://polaris.spectorn.ai/evidence/klaw1.md): identifying a planted law; knowledge identification is distinct from a survival advantage. - [KVALROOM2](https://polaris.spectorn.ai/evidence/kvalroom2.md): valuation and transfer limits. - [KCOMP1](https://polaris.spectorn.ai/evidence/kcomp1.md): composed rules and limits of measured benefit. ## Evidence interpretation - ATP is the model's accounted resource, not a direct measurement of physical energy or electricity. - Ticks, rows, temporary copies and repeated observations are not independent organism births. - A result on a family of graph worlds is not proof of unrestricted real-world generalisation. - Reading a trace of the past, using it in an action, benefiting from learning, and preserving it through repair are separate claims. - Matching implementations or surviving a process abort does not establish model validity or resistance to arbitrary hardware failures. - The portrait is an AI-generated artistic representation, not a recording or a visualisation of measured internal state. - The project does not claim that useful independent speech, consciousness, unrestricted self-improvement or scientific priority has been established. ## Discovery and collaboration - [Source excerpt manifest](https://polaris.spectorn.ai/evidence/manifest.json): SHA256 pins and excerpt scope. - [Sitemap](https://polaris.spectorn.ai/sitemap.xml): canonical public HTML pages. - [Collaboration](https://polaris.spectorn.ai/#contact): research, engineering, joint experiments and independent verification. Email polaris@spectorn.ai. This file is a reading guide. robots.txt governs crawler access; this document neither grants runtime authority nor changes the research evidence.