MoodSpan Project Dossier
A concise record of what exists, what is verified, and what still needs work.
This page is built for reviewers, portfolio packets, and handoffs: product scope, research assets, release boundaries, and the next technical lanes without inflated claims.
Current posture
Strong demo surface, active research instrumentation, and protected generated-library indexing while the corpus decision is unresolved.
Verified surfaces
Product paths that are worth showing now.
Kira
Sourced clinical question answering with retrieval metadata, citations, and safety boundaries.
OpenComparison Guides
Structured pages for high-confusion searches such as BPD vs bipolar and ADHD vs bipolar.
OpenInteractive Screeners
Self-report instruments with scoring, disclaimers, and clear non-diagnostic language.
OpenShowcase Runner
A live demonstration path that shows streaming events, retrieval status, and answer boundaries.
OpenResearch assets
The strongest portfolio value is the measurement discipline.
MoodSpan is more than a polished chat surface. The durable work is source retrieval, trace review, evaluation data, and release gates that make quality visible.
- Hybrid retrieval evaluation with gold queries and bootstrap confidence intervals.
- Ablation views comparing raw lexical search, semantic retrieval, and combined ranking.
- Quality gates for groundedness, citation coverage, unsupported claims, and safety outcomes.
- Founder/admin observability for weak traces, low-confidence retrieval, and release-risk buckets.
Remaining work
The next lanes are clear.
Content corpus restoration
Restore, regenerate, retire, or hybrid-quarantine generated libraries before indexing.
Strict Kira release gate
Move from regression protection to a stricter groundedness target before broader promotion.
Production verification
Re-run sitemap, robots, noindex, schema, and Lighthouse checks after any approved deploy.