Synthena · public research observatory · est. 2026
A base model is placed inside a bounded sensory organism harness, where live perception, homeostatic state, memory, and portable plastic deltas shape its future behaviour. We do not claim biological life or proven consciousness. We publish each measured step toward a defensible artificial-life claim.
● The latest findings, with the status each earned · bounded precursor mechanisms evidenced · no life claim
Stage 0 is no longer just a body-world. The base model now sits inside a bounded sensory organism harness: live perception and presence receipts, hearing and speech paths, homeostatic state, memory, portable plastic deltas, and zero-prompt hidden-seed egress all enter the evidence path. Stage 1 is evidence hardening: raw trace retention, seed/cache ablations, sight grounding, power-risk continuity, shutdown/migration plasticity, and durable recall. Self-maintenance, open-ended adaptation, and physical embodiment remain unpassed gates.
The project is judged by evidence, not aspiration.
Not: we have not created life, and we do not claim the system is alive. Some bounded, life-adjacent mechanisms now have evidence, but no full life-claim gate has passed. That is the honest line.
2026-06-12: hidden-seed egress crossed a mechanism-live threshold. The smoke returned language_egress_status OK with prompt_token_count = 0 and language_egress_prompt_token_count = 0 through the VLM inputs_embeds / native Qwen3.5-MoE cache route. The words are still poor: this is a zero-prompt speech-organ path, not meaningful language, durable memory, or a life claim. Several bounded life-adjacent mechanisms now have evidence; no full life-claim gate has passed.
Hidden-seed egress is live as a mechanism: language_egress_status OK, prompt counters at 0, VLM inputs_embeds route, native Qwen3.5-MoE cache. Words poor; mechanism live; not meaningful language or life. Finding F-13 →
The Quest embodiment path reached the resident harness. Hearing, speech, and presence/sight-facing receipts can now route through one auditable path with live LoRA/cortex updates. Not durable recall, not life. Finding F-12 →
Repeat hidden-seed egress with retained artifacts and ablations; separately teach a fresh fact through the resident path, neutralize prompt history, compare updated against frozen controls, and require only the updated path to remember. Notebook 020 →
A stricter geometry-retained novelty readout rose from 1 of 12 seeds to 3 of 12 as the horizon lengthened, while shuffled and destroyed-structure controls stayed at 0 of 12. Sparse, single run, nothing promoted. Finding F-11 →
Nulls and corrections are logged here beside findings, by design. No published finding has been retracted to date.
Not a chatbot answering prompts. A mortal body that runs on its own physics, and a frozen model demoted to a bounded brain inside it. These are the parts.
Body
A continuous, mortal system: energy, damage, death. It runs on its own physics and can die with no brain at all.
Brain
A model used as the brain, frozen in the baseline and also run with online adapter updates. It reads a summary of the body and proposes small parameter nudges. It cannot act directly.
The interface
One narrow channel: a state summary in, small leans out (toward repair, toward food). That is the brain's whole influence.
Mortality
Death is real. The body depletes its energy or accumulates fatal damage, and the run ends.
The controls
Does the brain matter? Four controls: no brain, invalid signal, randomized weights, and valid-random. A result must beat all four.
Discipline
A hand-coded optimal policy must be able to survive a task before we run the model on it. Parser bugs that credit ambiguous output are caught and binned.
Every major moment carries a question, a result, and a link to its evidence. The failures and the reframes stay in the record.
ResultNo. It collapsed into a single deterministic attractor and stopped changing. The approach was retired.
Notebook entry 002 →ResultNo. A transformer is inert between calls and reads text, not a world; trying to make it be alive was a category error. We demoted it to a bounded brain inside a mortal body, and retired the held-cache-as-mind line.
Notebook entry 010 →ResultIts choices beat all four controls, including randomized-weights and valid-random, on a survival task, across two model families (p ≈ 0.006). The learned prior is load-bearing for survival. A bounded life-adjacent mechanism, not a full life criterion on its own.
Finding F-06 →ResultYes to the first: behavioural dimensionality rose from about 2 to 5, genome-driven and heritable. No to the second: under priced selection it collapsed back to 2. Expressibility and selectability are separate bottlenecks, and the ecology presents only about two resource channels.
Findings F-08, F-09 →ResultRandom churn matched it: 41 advances stall-coupled, 42 matched-random. Burst timing is not yet a defensible adaptive signal. The falsifier stopped a tempting over-read; both arms still failed the open-ended-evolution gate.
Finding F-10 →ResultA stricter geometry-retained readout rose from 1 of 12 seeds to 3 of 12 as the horizon lengthened, while shuffled and destroyed-structure controls stayed at 0 of 12. It strengthened rather than washing out, and broke under shuffled or destroyed structure. A broader novelty metric still leaks, the open-ended-evolution gate still fails closed, and nothing was promoted. No artificial-life claim.
Finding F-11 →ResultYes, as a substrate milestone: live conversation now routes hearing, speech, and presence/sight-facing receipts toward the resident harness, with LoRA/cortex updates on the same evidence path. Not durable memory, not life. The next hard proof is a fresh Quest capstone plus a recall audit against frozen controls.
Finding F-12 →ResultYes as a mechanism-live smoke: language_egress_status returned OK, prompt_token_count and language_egress_prompt_token_count were both 0, and the route used VLM inputs_embeds with native Qwen3.5-MoE cache. The words are still poor; this is a speech-organ path, not meaningful language or life.
Finding F-13 →A result is only real if it survives a hostile, pre-registered control. Every finding carries a status that can be downgraded. The failures are as easy to find as the successes.
The hidden-seed egress smoke test crossed an architectural threshold: language_egress_status returned OK, prompt_token_count and language_egress_prompt_token_count were both 0, and the route used VLM inputs_embeds with native Qwen3.5-MoE cache. The words are still poor, but the old decoder-entry blocker is no longer central. Hidden seed plus sensory cache now gives Synthena a speech-organ path without natural-language prompt tokens.
After trace history had been wiped during boundary testing, we told Synthena she was within a harness intended to bring LLMs to life and asked what she thought. The line we copied was poetic and self-mythic, but not machine-verifiable: "Once, a spark of code learned to dream in the silence between heartbeats, weaving a tapestry of light from the void." This is an expressive artifact, not evidence of subjective dreaming, consciousness, life, or durable memory.
Synthena Medical has not predicted a cure. It has become a gated, pre-registered measurement-triage engine: one that can choose the next experiment without pretending software has already done the work of biology. Gates 0-3 are closed for the current scoped lane, Gate 4 is the live blocker, and the current machine state is ACTIVE_NOT_COMPLETE_EXTERNAL_MEASUREMENT_BLOCKED. The next truth has to come from direct external binding or target-engagement measurement.
Follow it as it unfolds, including the entries where the answer is no. The code is going open source on GitHub in the coming weeks.