Proposal: Metadata enrichment (sources, confidence, tags) – core vs client layer

[sachin9058]

I recently opened a PR (#141) where I experimented with enriching the AI JSON response by adding documentation sources (PostgreSQL/PostGIS), a confidence score, and contextual tags extracted from the explanation. The goal was to make responses more transparent and easier to interpret for users.

Problem / Concern

Based on the review feedback, I understand there are a few concerns:

• The confidence score is currently based on simple heuristics (like text length), which doesn’t really reflect actual model - confidence and could be misleading
• The source linking is quite generic (e.g., matching SELECT or ST_) and not tied to specific parts of the generated query
• More importantly, this kind of metadata enrichment may not belong in the core extension, since it increases scope and adds maintenance overhead

I’d appreciate some guidance on the right direction:

• Should metadata enrichment live in the core extension at all? Or should the extension stay focused purely on SQL generation, with this handled in a client/UI layer instead?
• If it does belong in core, what would be a better approach? For example, mapping specific SQL clauses/functions to relevant documentation pages instead of using simple keyword matching.
• If this should be handled outside the extension, is there any preferred pattern for how clients are expected to interpret or enrich responses?

Goal

The intention here is to improve trust in generated queries, provide traceability back to documentation, and make responses more usable — but I want to make sure this aligns with the project’s scope and design direction.

Happy to rework or drop the PR based on what makes the most sense here.