The original working phrase “I Don’t Know” was judged philosophically strong but poor as a primary public software name because it is extremely generic and difficult to search or distinguish.
An exact-name GitHub check for idkmesh / IDKMesh returned no existing repositories at the time of selection, which supported choosing IDKMesh.
SwarmForge was considered as a candidate but was already used in the AI-agent ecosystem, so it was rejected to reduce confusion.
The owner’s existing public MSKazemi/novafabric repository was inspected before naming IDKMesh. NovaFabric is a different project: an open-source execution-capsule/replay system for AI and HPC workloads focused on capture, sealing, replay, diffing, audit evidence, provenance, and reproducibility.
Conclusion: IDKMesh should remain a distinct project rather than reusing the NovaFabric name. Future integration between the projects may be possible because provenance/replay could eventually be useful inside a distributed collective-intelligence system, but the product theses are different.
The motivating question “Can 100 smaller-model vibe coders guarantee the quality of one strong coder/model?” has no general guarantee merely from increasing the number of contributors or agents.
Reasons include:
This produced an architectural principle: proposal generation and independent verification must scale together.
Traditional project management assumes a sufficiently clear goal that can be decomposed into tasks. IDKMesh starts one level earlier: participants may disagree about what the goal means.
Therefore the system should eventually represent:
A single issue tracker or flat backlog is likely insufficient as the core abstraction.
No single mechanism — voting, market pricing, blockchain consensus, reputation, or one optimization objective — appears suitable for every IDKMesh decision.
Likely separation:
The first useful prototype should not begin with millions of real machines. A single process or small cluster can simulate thousands of logical agents and nodes, making it possible to test task assignment, failure, reputation, diversity, verification, and communication policies cheaply and reproducibly.
A promising first research program is to compare coordination policies under controlled workloads and adversarial conditions before building a real volunteer-compute substrate.