Research project / PRJ-2026-003

File City and the Vibe64 System Model

Research into whether a large software system can be reconstructed from implementation evidence and made understandable as a navigable three- dimensional world, without requiring the user to read source code.

running FY 2026-27
Project periodJul 2026 to Jun 2027
Core activities1
Recorded experiments2
Field of research4609: Information and computing sciences

Project objective

Determine whether the physical and logical structure of a substantial software system can be extracted from its implementation and presented as a stable, interactive spatial model which helps a systems-minded person understand ownership, boundaries, dependencies and code concentration without first reading the source.

The immediate subject is Vibe64 itself. The proposed System tool sits beside Files and Diff inside an active Vibe64 session. It is intended to show the same working tree from another altitude: files and directories as physical evidence; subsystems, contracts and connections as the architecture those files realise.

Why experimentation is required

Static analysis can recover imports, declarations and framework registrations. Diagramming systems can display nodes and edges. Neither fact establishes that several hundred files, a directory hierarchy, logical subsystem ownership and different kinds of dependency can coexist in one spatial representation which remains truthful, legible and responsive.

The central difficulty is representational rather than merely cosmetic. A filesystem view is exact but says little about purpose. An architecture diagram can explain purpose but may conceal the evidence and become an attractive fiction. Showing every relationship produces a web; hiding relationships can erase the very architectural problem the user needs to see. Three-dimensional space adds useful axes, but also adds navigation, occlusion and performance failure modes.

Core activity CA-01

Evidence-backed spatial representation of software systems tests whether a small common model, populated by Vibe64-owned framework adapters, can project a real working tree into an interactive visual world where every displayed claim has a known evidentiary basis.

The activity includes the common ontology, the JSKIT extraction trial, spatial encodings for directories, files and subsystems, rendering at repository scale, and methods for reducing dense relationships without inventing structure. Routine product integration and visual polishing are recorded separately where they do not test a technical uncertainty.

Records retained

  • the dated design and implementation conversation;
  • the evolving implementation guide and rejected visual approaches;
  • the isolated system-world-v1 worktree and Git history;
  • the generated source, subsystem and relationship model;
  • screenshots and live human-evaluation notes for renderer revisions;
  • focused layout, model, route and browser tests; and
  • private time, remuneration, compute and expenditure records.

Current position

File City exists as a working Vibe64 prototype and has been merged into the main codebase. The work established several useful mechanisms, including repository-scale instanced rendering, evidence-typed connections and subsystem-aware bundling. Five discrete architectural levels were implemented, but their usability was not established. The project did not prove that the world improves human understanding compared with a conventional tree, map or source browser. That requires repeatable performance measurements and controlled comprehension tasks rather than the inventor’s own live evaluation.

Entity considerations

Ownership, control and financial risk

Project-level facts relevant to determining which entity conducted the R&D activities for its own benefit.

01

Effective ownership

Mobily Enterprises developed the System model and File City prototype in its Vibe64 codebase and can use, modify, publish and integrate the resulting open-source work. Supporting company records retain the relevant contractual and IP details.

02

Direction and control

Mobily Enterprises selected the research questions, chose Vibe64 as the trial codebase, designed the visual and evidence rules, directed the iterations and decided when a proposed representation was rejected.

03

Financial burden

Mobily Enterprises bore the researchers' time and model-compute costs for the work recorded here. Detailed remuneration, expenditure and apportionment records are retained privately rather than on this public site.

Research entries

Experiments in this project

Each entry covers one experiment or material iteration and keeps its own dates and supporting records.

RD-2026-006 concluded

Can six hundred source files become a usable three-dimensional city?

We turned Vibe64's 600-file working tree into an interactive File City and tested directory terrain, instanced rendering, exact dependency evidence, subsystem bundling and architectural layers. The prototype worked; whether it improves understanding remains unproven.

12 Jul to 15 Jul 2026 4 days 30 research hours · estimated Core
#ai#software-visualisation#three-js#file-city
RD-2026-005 concluded

Can software architecture become the primary view of a codebase?

We tried to reconstruct Vibe64 as an evidence-backed 3D system of responsibilities, interfaces and connections. The model was viable, but the first visual world was too abstract to serve as the primary way into the software.

11 Jul to 14 Jul 2026 4 days 28 research hours · estimated Core
#ai#software-architecture#visualisation#system-model