Funnel
R&D register
Open framework questions, tracked as a living list.
Open framework questions, tracked as a living list.
Announcement summaries currently use Groq with a Gemini fallback (both free-tier REST APIs, no new npm dependency -- lib/ai/summarize.ts). Both are rate-limited-but-perpetual free tiers, not expiring trial credits, so this should hold up indefinitely at this site's volume. If a provider relationship ever actually breaks, transformers.js (pure JS/WASM, no Python) running a small open model inside a GitHub Actions job is a genuine zero-vendor-dependency fallback -- real quality tradeoff (much smaller distilled models) and real complexity increase (model download/caching in CI), so not worth building until actually needed.
When should the shortlist/scoring lens switch from P/E to P/B (or vice versa) depending on sector or business-cycle stage? No rule has been validated yet.
A breadth thrust (rapid expansion in the % of stocks above their 200-DMA) is a classic bullish signal, but the framework has not settled how to read a thrust that reverses quickly, or whether the same signal works symmetrically on the downside.
Stage 3b's RS-Ratio/RS-Momentum liquidity gate (lib/funnel/rrg.ts) uses a documented but unvalidated methodology -- parameters (smoothing/normalization/momentum windows) were chosen as reasonable defaults, not backtested against Indian sector rotation history. Track real-world hit rate here before leaning on it for sizing decisions.