Small modules.
Room to grow.

One storage language, three arrangements

Drawer bays keep small parts contained; shelf bays keep frequently used pieces accessible. Top plates, raised feet, links, keys and a base coupler complete the assembly.

• 2×2 · the starting point

  Two drawer bays and two shelf bays, arranged diagonally. The canonical package contains six drawers, their tag plates and the connecting parts.

• 2×3 · more vertical storage

  Three drawer bays and three shelf bays over three tiers. Taller geometry received its own export and sampled service checks.

• 4×2 · spread across the desk

  Four drawer bays and four shelf bays in adjacent clusters. The wider base and far seam remain part of the documented evidence limits.

These are measured outputs from the delivered manifest, not a load rating or proof of arbitrary stacking.

Directed by a person. Coordinated through Dots.

OpenAI Dots coordinated Antigravity planning and building, and separate evidence-based reviews. The workflow connected a human brief to native Blender construction, measured checks and revisions.

1. Human direction sets the target

   The owner supplied the product direction and feedback. Decisions about arrangement, appearance, mechanisms and presentation shaped subsequent revisions.

2. Antigravity plans and builds

   Antigravity worked with the project scripts and Blender native assets. Models, materials, scene setup and rendering stayed traceable to saved project artifacts.

3. Dots coordinates review and follow-through

   Separate reviews checked geometry, exports and sampled mechanics against their evidence. Findings led to refinements, including the v16b base-seating revision.

Coordination helps keep tasks and evidence connected. It does not replace engineering judgment, human feedback or physical tests. Mention of OpenAI describes the tools used; it does not imply endorsement.

What the digital evidence establishes

The recorded geometry/export checks pass for 32, 45 and 68 parts. Their scope includes topology, dimensions, wall-ray screening and STL export with fresh reopen.

Sampled drawer motion covers a nominal 0–70 mm stroke. Service probes cover key removal, staged unstacking and the base route. These are bounded sampled tests, not continuous swept-volume or force/friction proofs.

The corrected retainer route slides 26 mm before dropping. The earlier 24 mm route retains a collision failure; the builder regression that misses that intermediate collision remains unresolved.

The canonical delivery scene was renamed after audit. Its identity receipt links unchanged meshes, bases and origins to the audited source; the gate does not directly carry the final delivery hash.

There are 64 / 90 / 136 unchecked entries across the three configurations: undeclared bore-feature checks and report-only overhang area. Raw seating/seam intersections and numerical-contact caveats remain documented. A sampled wall-ray screen does not establish a continuous minimum wall on every surface.

Read the public evidence summary

The next step is a physical print test

Printer, material, nozzle, shrinkage, layer adhesion, support scars, fit, friction, fatigue, strength, drawer loads and tip stability remain unverified. No rated height or payload is assigned.

1. Choose a real printer, material, profile and intended payload.
2. Print and measure small mating subsets before a complete tower.
3. Record fit and service results, then plan bounded load and stability tests.

The default 256 mm bed and PLA/FDM 0.4 mm assumptions in the digital specifications are metadata, not validated process choices. Tag plates are geometry; film lettering is not manufactured STL lettering, and physical label attachment remains unspecified.

A film explains the system; a new base improves the candidate

The existing 30-second Full HD film uses 1,800 native rendered states at 60 fps. It is an intermediate presentation; the expanded 56-second film was still in progress when this story was prepared.

Canonical media uses frozen E304. The v16b print-test candidate changes base seating: 2.8 mm engagement, a 0.2 mm floor gap and nominal 0.4 mm lateral clearance per side. Those digital measurements do not establish fit or strength.

Exploded, catalog and blueprint views are explanatory presentation modes. Their positions and module transitions should not be read as verified physical assembly paths or authoritative manufacturing drawings.

This story distributes no reference photographs, Blender files, STLs or geometry archives. Derived renders are also held while reference-design redistribution rights remain unverified.

Reference and provenance

The reference design is NODESTACK // SERVICE PACK by SMACKMAX_ (@SMACKMAX). This project is a Blender reconstruction and local engineering study; it does not claim authorship of the original design.

The reference-design redistribution license has not been verified. The repository’s MIT software license does not establish redistribution rights for that design. Geometry downloads are held at the owner’s direction.

The public summary is derived from the delivered v16b manifest and handoff. It retains artifact hashes and scope limits while excluding private paths, internal conversations and unrelated project data.

Questions about the project

Built within design:os

The Blender repository carries the build workflow, native-asset policy and evidence gates used for these studies.

• Explore the Blender worked builds
• Explore the design:os toolkit
• Read the Blender AI workflow
• Read the publication evidence rules