STUDIO RECORD / PUBLIC EDITION

Visual stages, comparisons and decisions from the original run. Internal prompts, operating documents and downloadable implementation files are omitted. Some interactive comparisons are shown as separate images.

Round complete: six transfer studies, findings and full ledger →

Construction batch completed: results and updated ledger →

Development round / Batch 1 of 3

Can the instrument
do what we ask?

Twelve fixtures, three paper seeds each. One consolidated checkpoint. U03 was not changed.

Verdict: proceed to construction, with limits

All 36 fixture executions passed their stated numerical criteria. Visual inspection still identified a defect: sparse paths bead along their edges. The original 10% tone-tolerance check passed with 9–9.4% variation, but was too permissive to judge edge quality.

Corrected in an opt-in adapter: uniform arc-length sampling removes the visible beading for the tested straight path. Nine follow-up density/seed combinations agree within a scalar tolerance of 0.000001. The adapter is ready to trial in construction tests; it has not replaced the production planner or been validated for every mark family.

Do not promote native lifting to full removal. Twelve lifts recover about 97.86% of the deposited tone, yet leave a readable outline. Use explicit digital erasure when complete removal is intended. This finding separates tone recovery from perceptual success.

The controls now justify a construction comparison. They do not establish expressive drawing, anatomy, autonomous artistic judgment or superiority to other systems.

Progress ledger

Batch	Planned outputs	Completed	Status
Instrument control	12	12	Complete · 36 executions
Drawing construction	12	12	Complete · see construction review
Artistic transfer	6	6	Complete · see transfer review

The sampling correction adds nine focused checks to this batch. These are robustness checks, not nine additional artworks. The 30-output plan remains a bounded exploratory round.

Next checkpoint: four passages × three approaches—current method, finer observation, selective reconstruction. Use hands, contour, fabric supporting weight, and an arm–torso opening. Compare actual renders in context, preserve failed results, and choose methods only when the whole drawing benefits.

Required: construction comparison and contextual judgment. Optional: more media, larger seed sweeps and physical charcoal simulation. No further fixture expansion unless a new change or unresolved result calls for it. .

Process and methods

The harness exercises the existing Hands-based native canvas directly with known paths and mark parameters. It varies one principal control at a time and repeats with paper seeds 11, 29 and 73. The charcoal vocabulary's first mark supplies the base settings; this does not test its entire repertoire.

Each figure shows seed 11. Tone is measured as scalar distance from the paper value. Support uses an explicit contrast threshold. Boundaries and replay use exact pixel/state comparisons. Engineering thresholds are declared in the harness and are not published perceptual standards.

Most fixtures bypass the planner and apply known commands. This isolates the instrument but leaves planner choice, anatomy, pressure along a single stroke, complex curvature and artistic quality unproven. The numerical verdict and the assistant's visual judgment are separate.

The correction resamples a supplied polyline at uniform arc-length intervals. It cannot recover curvature missing from the supplied points. Intentional dots and broken marks must not be silently normalized into continuous strokes.

See corrected straight-path rendering

Continuous stroke after uniform path spacing

Contact sheet

Fixture results

01-pressure · Pressure response