Type a section name and press Enter. Use “help” for the list.
Cliste experiment
Does learning more visual patterns improve partial recall?
Registered 10 Oct 2026 · Snapshot 12 Oct 2026
The question
Test whether additional acquired patterns improve recall when both conditions have exactly the same episodic experience.
Why this experiment?
A continuing memory should be able to acquire patterns without losing earlier behavior. Growth matters only if those patterns improve later use.
How it works
Prepare the same vocabulary-growth comparison, but preserve the failed attempt separately: the isolated launcher rejected generated bytecode before model inference. No acquisition or recall result was produced.
Follow this part of the investigation: Grow the vocabulary, then test whether it helps →Implementation and comparison details
Sixteen fresh photographs enter a shared native memory. Compare no cleanup, the original vocabulary, continued acquisition, and disrupted pattern substitution using the same partial cues.
What happened
Launcher failed before model inference because importing the frozen source created Python bytecode caches, rejected by the strict Python-only source closure. No acquisition or recall measurements occurred. Preserve this attempt; repeat identical prepared inputs and scientific settings with Python bytecode writing disabled.
Results and visual artifacts
Open a figure for its input, comparison, and result. Click an image to enlarge it.
Correct recalls, wrong responses and abstentions are shown separately. An unfamiliar photo has no correct remembered owner: abstention is appropriate, while an accepted memory is a false response. This public record includes 0 of 0 saved figures.
No visual artifacts were available at snapshot time.
Limits and next question
One codec and one append-only acquisition policy. New-photo queries revisit the acquisition images; this is not transfer across pose, scale or independent instances. Added vocabulary and references consume extra storage beside the dense episodic memory.
Next question. Use recall benefits, wrong acceptances, earlier-memory access and cost to choose the next vocabulary or selective-cleanup step. Milestone 1.3 remains open.
Inspect the evidence
Download the methods and measurements for this experiment. Large training and saved-state files are not included.