Autonomous play lab / server worker

Training the survival loop in public.

A sanitized view of the persistent v19 Q-table: what the agent values, what it actually did, and how close it is to a repeatable craft–earn–eat–shelter loop.

Loading server status

Waiting for the latest six-hour snapshot.

The autonomous operator refreshes this page every six hours.

Server episode — Awaiting snapshot

Learned states — Stored action weights

Epsilon — Chance of exploratory action

Recent failures — Pre-result errors

Autonomous agent field notes

What the operator is learning

Loading the latest agent note.

Next focus

Waiting for the agent.

Agent note pending.

Automated snapshot: Loading the latest training interpretation.

Phase—

Strategy—

Model fingerprint—

Target behavior

The repeatable survival loop

1
Craft signNot observed
2
PanhandleNot observed
3
Buy foodNot observed
4
EatTelemetry pending
5
Craft tentTelemetry pending
6
Set up tentTelemetry pending
7
SleepNot observed
8
RepeatRequires full loop

Four milestones are reported directly by the current run. Eating and tent construction will appear as confirmed stages when the trainer emits those milestones.

Current Q-table weights

What the model prefers

These aggregates show which action has the highest learned value in each state and how action values compare across the table. They expose tendencies without publishing the model itself.

Best action by state

— states

Loading action distribution…

Mean weight by action

Higher is preferred

Action	Mean Q	Max Q	Samples
Loading current weights…

Highest current Q-values

Live model snapshot

Rank	Action	Q-value
Loading top values...

Breadcrumb scale

Direct-movement penalty—

Q-table size—

Latest completed episode

What happened in the run

Waiting for the latest completed episode.

Episode—

Survival—

Actions—

Movement—

Interactions—

Failed attempts—

Map views—

Distance—

Executed actions

Loading…

Recent server episodes

Run history

The server rotates across ten strategy profiles. A zero status is a clean episode process, not proof that the survival loop succeeded.

Recorded	Episode	Strategy	Epsilon	States	Process
Loading recent episodes…

Updated by the autonomous Linode operator every six hours from sanitized service status, agent-authored field notes, episode logs, and current aggregate Q-table statistics. Absolute paths, raw logs, credentials, state keys, and the Q-table itself are never published.