run record~/research/cliste/runsAll Cliste runs

Type a section name and press Enter. Use “help” for the list.

Figures: softened for dark viewingDisplay only; image files are unchanged.
← All Cliste runs

Cliste experiment

Native continuing life: partial perception, recall and acquired cleanup

Registered 8 Oct 2026 · Snapshot 12 Oct 2026

Execution: completedScientific verdict: bounded integration engineering pass

The question

Connect ordinary native-image admission, retained VSA content, partial-cue recall, working use and restart in one durable life.

Why this experiment?

Remembered content needs a role in current activity and a relation to earlier and later events. Supplied controls let those mechanisms be tested separately.

How it works

Supply image fragments and visibility masks to recall, then test working-state hold/release and temporal replay. The addresses, clocks and control commands are supplied; encountered images provide the content.

Follow this part of the investigation: Recall something, hold it, then recover what followed →
Implementation and comparison details

Ordinary native intake, dense eight-bit retained content, causal acquired cleanup, partial-input query, working use and exact continuation controls in a durable life.

What happened

Engineering controls passed. Final no-cleanup queries accepted 25/32 correctly, 0 wrongly and abstained on 7. Before admission, 0/8 absent cues were falsely accepted. Cleanup outcomes are recorded separately against no-cleanup; these results do not close milestone 1.3.

Measurements
Measurements — values preserved from this run
Recorded fieldValue
verdictbounded_integration_engineering_pass
engineering passedPass
no cleanup final
{
  "stage": "final32",
  "mode": "none",
  "group": "all",
  "queries": 32,
  "accepted": 25,
  "abstained": 7,
  "accepted_correct": 25,
  "accepted_wrong": 0,
  "absent_false_accept": 0,
  "candidate_correct": 32,
  "statuses": {
    "accepted": 25,
    "ambiguous": 7
  }
}
no cleanup absent
{
  "stage": "acquisition16",
  "mode": "none",
  "group": "absent",
  "queries": 8,
  "accepted": 0,
  "abstained": 8,
  "accepted_correct": 0,
  "accepted_wrong": 0,
  "absent_false_accept": 0,
  "candidate_correct": 0,
  "statuses": {
    "ambiguous": 8
  }
}

Results and visual artifacts

Open a figure for its input, comparison, and result. Click an image to enlarge it.

Cliste stores whole images as structured VSA memory, then receives a small local view as a recall cue. Each row compares the same cue with and without acquired cleanup. Cleanup replaces a noisy feature vector with a sufficiently close acquired prototype; it is a separate operation from selecting a whole remembered image. The source image at left is an answer reference for the evaluator, not an input to retrieval. A correct answer must select the source memory; an abstention means the acceptance rule declines to return any memory. Scores are similarities and gaps between competing candidates, not probabilities.

Recalling an image from a local view after 16 encounters — examples 1
Recalling an image from a local view after 16 encounters — examples 1
Left to right: source image, the presented 80 × 80-pixel view, recall without cleanup, and recall after cleanup with learned prototypes. The whole local field is visible, but the rest of the source image is unavailable to the query. “Source owner” is the expected memory number, used only to score the answer; “candidate” is the selected memory. A blank response labeled rejected means no new image was recalled. These fixed examples include images not yet stored at the earlier stage, so declining those cues is appropriate. Compare corresponding rows across the 16- and 32-encounter stages to see what changes as memory grows.

Display derivative: WebP quality 90; maximum edge 2400 px. Original scientific pixels remain in the source run; do not measure this preview.

Recalling an image from a local view after 16 encounters — examples 2
Recalling an image from a local view after 16 encounters — examples 2
Left to right: source image, the presented 80 × 80-pixel view, recall without cleanup, and recall after cleanup with learned prototypes. The whole local field is visible, but the rest of the source image is unavailable to the query. “Source owner” is the expected memory number, used only to score the answer; “candidate” is the selected memory. A blank response labeled rejected means no new image was recalled. These fixed examples include images not yet stored at the earlier stage, so declining those cues is appropriate. Compare corresponding rows across the 16- and 32-encounter stages to see what changes as memory grows.

Display derivative: WebP quality 90; maximum edge 2400 px. Original scientific pixels remain in the source run; do not measure this preview.

Recalling an image from a local view after 32 encounters — examples 1
Recalling an image from a local view after 32 encounters — examples 1
Left to right: source image, the presented 80 × 80-pixel view, recall without cleanup, and recall after cleanup with learned prototypes. The whole local field is visible, but the rest of the source image is unavailable to the query. “Source owner” is the expected memory number, used only to score the answer; “candidate” is the selected memory. A blank response labeled rejected means no new image was recalled. These fixed examples include images not yet stored at the earlier stage, so declining those cues is appropriate. Compare corresponding rows across the 16- and 32-encounter stages to see what changes as memory grows.

Display derivative: WebP quality 90; maximum edge 2400 px. Original scientific pixels remain in the source run; do not measure this preview.

Recalling an image from a local view after 32 encounters — examples 2
Recalling an image from a local view after 32 encounters — examples 2
Left to right: source image, the presented 80 × 80-pixel view, recall without cleanup, and recall after cleanup with learned prototypes. The whole local field is visible, but the rest of the source image is unavailable to the query. “Source owner” is the expected memory number, used only to score the answer; “candidate” is the selected memory. A blank response labeled rejected means no new image was recalled. These fixed examples include images not yet stored at the earlier stage, so declining those cues is appropriate. Compare corresponding rows across the 16- and 32-encounter stages to see what changes as memory grows.

Display derivative: WebP quality 90; maximum edge 2400 px. Original scientific pixels remain in the source run; do not measure this preview.

Limits and next question

Previously examined natural inputs and supplied visibility; not a fresh generalization benchmark. Dense eight-bit storage and the immutable codec are provisional machinery. A 512-entry acquired bank is bounded. Engineering continuity and scientific cleanup/novelty outcomes are separate; no complete milestone 1.3 claim or unseen final-panel evaluation is assumed.

Next question. Use observed integrated failures to choose the next cognitive mechanism; no automatic codec optimization or complete 1.3 claim.

Inspect the evidence

Download the methods and measurements for this experiment. Large training and saved-state files are not included.