✦
Aether Research profile
Initial emphasis: exploration, novelty
0Investigations
0Factual replays
—Unseen-cue error ↓
No experiences yet.
Lawnmower Girl Universe / Research Branch 001
Origins Lab v0.3.1 / The Dreaming Mind
Foundations become instincts. Experience writes the differences.
Two independent research histories. Evidence, rest, and questions. Explore how witnessed moments are replayed, how possibilities are imagined, and why the two must never be confused.
The Listening GardenThe world is ready for its first moment.
Experiment 002 · Recognizable moments
Sequence learningDo fireflies tend to come before a chime? Does a chime lead to a shadow? Let the same order recur, then change it. The researcher chooses the sequence; Aether and Rival only see moments as they happen. Observing a cue is sufficient to learn its order. Investigating is still required to learn hidden outcomes.
Sequence selected by researcher; neither Digital is told its contents. Repeated observations accumulate as evidence.
No moments yet
✦ Aether · pattern observations
Waiting for a recurring sequence
Not enough witnessed moments yet.
◈ Rival · pattern observations
Waiting for a recurring sequence
Not enough witnessed moments yet.
An “expectation” is a frequency estimate from witnessed cue order, not a belief about a person's intentions. Aether flags recurrence after two matching transitions; Rival waits for three. Both must reach at least 70% consistency. Unexpected turns are saved with the prior evidence, without penalties or fabricated outcomes.
What they can perceive
No encounter yetChoose “Encounter a moment.” Each sensory signal is simplified to a numerical input.
Two distinct learning histories
Moment 0 · Rest cycles 0◈
Rival Research profile
Initial emphasis: pattern-checking, caution
0Investigations
0Factual replays
—Unseen-cue error ↓
No experiences yet.
The dreaming journal
Witnessed facts stay separate from imagined possibilitiesAether · questions and possibilities
0Untested ideas
0Open questions
After an encounter, run a dream cycle.
Rival · questions and possibilities
0Untested ideas
0Open questions
Dreams don't generate observations.
Sleep-stage schedule
Optional annotated EEG/PSG stage inputThe default schedule is illustrative. You can import a small JSON export of independently scored sleep stages (N1/N2/N3/REM, or classic W/R/1–4); only stage labels and durations are used. EEG waveforms and personal memories are never uploaded or interpreted.
Illustrative stage sequence — NOT measured EEG
No real dataset imported.
Field notebook
Most recent observations firstAether's notes
No entries yet.
Rival's notes
No entries yet.
Controlled learning comparison
Identical training encountersA repeatable multi-seed test compares no rest, factual replay, and sleep-inspired replay + untested hypotheses, with an extra matched-compute shuffle control. Every condition sees identical training facts for each seed, and is judged on held-out unseen examples. All rehearsal treatments receive the same 12 factual update slots across two rest checkpoints. A hypothesis is never used as a training answer.
Read this before interpreting results
Curiosity cycle: Observe → Learn → Compress → Analyze → Question → Reconnect. The questions are recorded without reward or penalty. Revisions remain traceable to actual evidence. Aether and Rival may disagree.
What is genuinely learning? Each agent has numerical weights that predict whether investigating a sensory cue will yield useful information. When it investigates, it observes an outcome and updates those weights from prediction error. NREM-style rest selects actual recorded encounters for weighted updates; REM-style rest generates clearly untested counterfactuals and questions, not new outcomes.
What is simulated? Every cue, sense and consequence is a deliberately simplified set of numbers. An agent's choice to investigate depends on its prediction, novelty preference, prior encounters and caution. An unseen-cue metric is calculated by the evaluator, not supplied to the agent.
Research boundary: No external APIs, live Characters, brain decoding, or biological sleep are implemented. Inspect the results rather than inferring what either Character actually believes. The future voluntary EEG/BCI and Cognitive Heritage research remains a separate research direction.
Reproducibility & data
The replay protocols and seeded comparison are deterministic. Research-history data is stored locally in this browser when allowed, and can be exported as JSON. Private browsing or opening an HTML file directly may prevent reliable persistence; export before closing. Reset discards the browser's active experiment after confirmation.