apply_feedback, decay, get_mean_affinity, and engagement_score had no
(or broken) test coverage. Also fixes decay(): subtracting a fixed
DAILY_DECAY from a score smaller than that step flipped its sign
instead of settling at zero, causing oscillation on repeated decay
passes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KcD2BtNqehJxchKiuodTxi
sqlx's query!/query_as! macros only check raw SQL strings against the live
schema; Diesel's table!-derived DSL type-checks query structure itself at
compile time. SQLite has no native async driver, so diesel-async wraps a
blocking SqliteConnection via SyncConnectionWrapper, pooled with bb8.
Migrations move from sqlx's single-file-per-migration format to Diesel's
up.sql/down.sql pairs, run transactionally via diesel_migrations against a
throwaway sync connection at boot (MigrationHarness needs a sync
Connection), giving revertable migrations that sqlx::migrate! doesn't
support.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Was hardcoded to llama3.1, which isn't actually on this machine.
Switched to gemma4-e4b (5.0 GB, already pulled) as a speed/quality
balance for a call made once per shortlisted article on every
pipeline run.
Rust workspace with core (topic-affinity learning engine + relevance
scoring), db (sqlite/sqlx schema + repo), feeds (RSS/Atom fetch), llm
(rig + local Ollama embeddings/completion), and web (axum + Dioxus
fullstack UI, no separate JS stack). Two-stage relevance filtering
(embedding shortlist -> LLM judgment) and an engagement/surprise-based
topic affinity engine with daily decay. All crates compile and core's
affinity engine has passing unit tests; server and wasm client targets
of feedsignal-web both check clean.
