Breakthrough Memory bloomnet

Agent MQI standalone repo: first public-facing artifact of bloomnet quality observability

->

breakthroughai-agentsfrontend
Key Metric
Before
↓
After

The MQI dashboard was extracted from rusty-bloomnet into a standalone repository with a viral README, dashboard thumbnails, and a thought leadership blog post. This is the first public-facing artifact of the bloomnet quality observability system: the internal tooling that has been measuring and gating agent quality across all projects since March.

What Happened

The MQI (Model Quality Index) dashboard had been living inside rusty-bloomnet as a tightly coupled module. It read from bloomnet.db, rendered via a local web server, and was only accessible on the development machine. The dashboard was powerful: per-model baselines, staleness detection, metric distributions, session-level drill-down: but invisible to anyone outside the immediate workflow.

The extraction moved the dashboard into its own repository with clean boundaries:

  1. Standalone data layer. The dashboard reads from a SQLite database with a documented schema. It no longer imports rusty-bloomnet internals. Any system that writes MQI-compatible rows to the database can drive the dashboard.

  2. Viral README. Three-act structure: the hook (what if your agents told you when they were degrading?), the evidence (dashboard screenshots showing a real quality regression caught and surfaced), and the call to action (run it yourself in 3 commands). Dashboard thumbnails are embedded directly: the README renders as a visual artifact, not a wall of text.

  3. Blog post. Thought leadership piece covering the problem space (agent quality is unmeasured in most deployments), the approach (per-model statistical baselines, not global thresholds), and the results (real incidents caught by the system). Written for an audience of ML engineers and agent-tooling builders who face the same observability gap.

The extraction preserved all existing functionality. The rusty-bloomnet crate now depends on the standalone dashboard as an optional feature: existing workflows are unaffected, but the dashboard can also run independently.

Why It Matters

This is a distribution inflection point. Every prior bloomnet artifact was internal-only: the Rust crates, the quality hooks, the MQI database, the scanner infrastructure. Useful, but invisible. The standalone repo makes the quality observability story tangible and shareable.

The three-act README structure is deliberate. Most developer tool READMEs lead with installation instructions. By leading with the problem (agents degrade silently) and the evidence (screenshot of a caught regression), the README converts readers who don’t yet know they need the tool. The installation instructions come last, after the reader already wants it.

The blog post extends reach beyond the repository. It positions the MQI approach in the context of the broader agent-quality problem, referencing the Huang Constraint work and the IO taxonomy as the theoretical foundation. This connects the practical tool to the research contributions.

Evidence

  • Standalone repository created with clean SQLite interface, no rusty-bloomnet imports.
  • README with embedded dashboard thumbnails renders correctly on GitHub.
  • Blog post drafted covering problem space, approach, and real-incident evidence.
  • rusty-bloomnet updated to use standalone dashboard as optional dependency: existing cargo test suite passes.