The Autonomic Web manifesto makes a claim: a page can become the visible state of a continuing decision process rather than a fixed document displayed repeatedly. limina.li is the experiment where that claim can be inspected rather than merely asserted.
This article stays at the level of public behavior. It does not describe the private wiring of the system. The useful evidence is what a visitor can observe: pages organized as nodes, choices that create paths, an expanding graph, operational counters, persisted state, and a visible distinction between ready and pending material.
What the system presents to a reader
A limina.li page is a narrative node containing prose, imagery, context, and three possible moves. Selecting a move either opens an existing node or requests an unexplored edge. The resulting sequence becomes a path through a graph rather than progress through a fixed table of contents.
The important difference from ordinary branching fiction is not simply that choices exist. A conventional branching work can pre-author every route. Here, the public system states that unexplored edges can enter a generation queue and become persistent nodes. The graph can therefore grow through use instead of only revealing a tree completed in advance.
That makes the reader's choice an input to future system state. It does not mean every sentence is uniquely generated for every view, nor that the reader has unlimited control. The system still operates within a world, vocabulary, visual identity, and set of allowed moves.
A dated public snapshot
As observed on 25 September 2026, the public homepage reports:
- 2,698 published nodes;
- 5,397 pending nodes;
- 8,095 nodes in the reported graph surface;
- a 33.3% completion rate;
- three user-requested nodes awaiting priority generation;
- a last-published date of 14 June 2026.
These counters matter because they make the unfinished state visible. The system does not pretend that every possible branch already exists. It distinguishes navigable content from discovered but unpopulated edges and shows that the graph is larger than the published surface.
The numbers are a snapshot, not a timeless performance claim. They may change, and they do not by themselves establish narrative quality. They establish that the public experience is backed by persistent, inspectable state rather than presented only as a concept video.
Evidence visible in the graph
The homepage exposes an example five-page thread. Each node shows three moves, the selected move, conceptual axes, and the next target. A separate branch view distinguishes ready nodes from pending descendants. Summary pages expose graph depth, choice history, and content signals.
Together, these surfaces demonstrate four properties relevant to the manifesto:
1. State persists. A reader's path is represented as a sequence through named nodes rather than disappearing after one response.
2. The system exposes alternatives. Decisions are visible as branches, not hidden inside a single generated answer.
3. Unfinished work is represented honestly. Pending edges remain distinct from published nodes.
4. Generation changes the shared graph. New material is described as persisted content that can later be navigated, not only as ephemeral output in one session.
This is close to the seed model: the experience grows inside persistent goals, constraints, identity, and world rules instead of treating each prompt as an unrelated request.
Why narrative is a useful test environment
Narrative makes system behavior legible. A reader can see continuity fail when a branch contradicts its own history, see choice become meaningless when every path converges, and feel identity drift when imagery or prose leaves the world established by earlier nodes.
Those failures correspond to problems that occur in other adaptive systems: state loss, objective collapse, inconsistent policy, and local optimization that damages the whole experience. Narrative turns them into something a person can inspect directly.
It also tests a difficult balance. The system must generate enough novelty for choices to matter while preserving enough constraint for the result to remain one world. Pure repetition defeats autonomy; unconstrained variation defeats coherence.
What this experiment demonstrates
The public system demonstrates that an adaptive content experience can maintain an expanding graph of pages, preserve paths, expose pending state, combine content and imagery, and let reader decisions participate in what becomes available next.
It also demonstrates a form of accountable incompleteness. A pending edge is not silently represented as finished. Operational status is part of the interface. That aligns with the wider position of this project: autonomous behavior should leave observable evidence of state and transition.
The experiment therefore supports a narrower claim than “the future of all websites has arrived.” It shows that the page-as-node model can operate publicly at meaningful scale and remain navigable as the graph grows.
What it does not prove
The current public evidence does not prove that generated narrative is consistently better than authored narrative. It does not establish that every branch is coherent, that every choice is equally meaningful, or that the system can grow indefinitely without editorial degradation.
The completion rate is not a quality score. Published-node count is not evidence of reader satisfaction. A large queue may represent possibility, demand, or accumulated unfinished work; the number alone cannot distinguish between them.
The experiment also does not prove that the same architecture should be used for high-consequence information. Narrative tolerates ambiguity and surprise that would be unacceptable in medical, legal, financial, or safety-critical guidance.
Finally, a public interface can show outputs and operational state without exposing every decision or dependency behind them. This case study deliberately does not claim more transparency than the public evidence provides.
The next useful evaluation
The strongest next step is not simply generating more nodes. It is measuring the qualities the manifesto says matter: coherence across depth, meaningful divergence between choices, rate of human intervention, repeated failure patterns, and the ability to reverse or repair a branch without damaging the graph around it.
Reader signals can contribute, but they need interpretation. A selected path may be interesting, confusing, or merely first in the list. A highlighted passage indicates attention, not necessarily understanding. Evaluation should combine behavior with deliberate review rather than optimize directly from clicks.
That is where governance, not just intelligence, becomes part of the narrative experiment. Growth is easy to count. Controlled, legible, and correctable growth is the harder standard.
From manifesto to falsifiable work
limina.li matters to Autonomic Web because it gives the ideas somewhere to fail in public. The graph can become repetitive. Choices can lose consequence. Pending work can outrun useful completion. Identity can drift. Those outcomes would be evidence against particular design decisions, not inconveniences to hide.
A working example should make the thesis more vulnerable, not merely more impressive. As of this snapshot, limina.li demonstrates a persistent, expanding narrative graph driven partly by reader choice and agentic generation. The next question is how well that graph preserves meaning as autonomy accumulates.