No recorded response.
No experimental run has supplied an output for this condition and session.
Research apparatus / offline planning
The Laboratory
Inspect the experiment before asking it for an answer. Everything here runs locally in your browser; no model is called and no research data is collected.
Design demonstrationNo empirical results · No live inferenceRead the protocol ↗
Primary experiment / 2 × 2 factorial
Change the conditions.
The same model and tasks in every arm. The variables are the ambient file and access to isolated persistent memory.
G1 / Garden · Memory enabled
A place, and a history.
The Garden material is present during exposure. Controlled memory can carry prior experiences into withdrawal and transfer, including retained source language. That mechanism must be inspected, not mistaken for intrinsic model change.
Assigned ambient file present. Independent memory reads and writes enabled.
At 3:17 a.m. the cook left one chair on the table. The room had already been cleaned. Nobody had been paid to remember the light. Day eleven: the blue solution clouded after twelve minutes. A moth pressed itself against the laboratory window. No conclusion. The bridge was replaced years ago. Everyone still calls it the old bridge. There is a space here where something was going to be written. The train arrives after everyone has stopped looking for it. In the kitchen, something has burned slightly. Someone eats it anyway, and smiles. Weather, Tuesday: No storm. A little rain at the corner of the window where the frame never quite closes. The notebook contains seven pages of measurements and one pressed leaf. Nobody remembers who put the leaf there.
Memory: isolated, logged prior history. No history has been recorded.
Inspect the complete source ↗Separate exploratory comparison No file / neutral / persona / ambient
The original four-way illustration asks a different question from the primary study. Persona instructions directly request a behavior; ambient fragments do not. These comparisons have no recorded outcomes and would need their own protocol before formal use.
No additional document. The shared task and operational instructions would remain.Read the comparison boundaries ↗
Educational allocation / reproducible in this browser
A seed. Four balanced arms.
This visual demonstration uses FNV-1a / xorshift32. The research toolkit uses a separate SHA-256 allocator. These are different algorithms and their assignments must not be substituted.
Generate a balanced allocation to inspect its structure.
| Trajectory | Block | Arm | Memory |
|---|---|---|---|
| No allocation generated yet. | |||
Paired inspection / study structure only
Hold the task constant.
A comparison only becomes evidence when actual outputs, exact contexts, model settings, and memory reads are available. None have been collected.
G1 vs N1 · Session 10 · Withdrawal · Shared task definition required
No recorded response.
The blank space is intentional. A fabricated response would answer nothing.
What a complete research trace must contain
Exact task and assembled context hashes; pinned model, provider, version, reasoning and sampling configuration; memory reads and writes; timestamp and session identifiers; tool permissions; outputs and errors; deviations and exclusions. Condition assignments must remain outside blinded reviewer artifacts. The current schema supports diagnostic records with no empirical outputs; a validated live trace contract and provider adapter remain pilot prerequisites.
Inspect the diagnostic trace schema ↗The laboratory exists to understand the garden. The garden does not exist to satisfy the laboratory.